<?xml version="1.0" encoding="UTF-8" standalone="no"?><rss xmlns:atom="http://www.w3.org/2005/Atom" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:itunes="http://www.itunes.com/dtds/podcast-1.0.dtd" xmlns:slash="http://purl.org/rss/1.0/modules/slash/" xmlns:sy="http://purl.org/rss/1.0/modules/syndication/" xmlns:wfw="http://wellformedweb.org/CommentAPI/" version="2.0">

<channel>
	<title>VC Cafe</title>
	<atom:link href="https://www.vccafe.com/feed/" rel="self" type="application/rss+xml"/>
	<link>https://www.vccafe.com</link>
	<description>Founded in 2005, VC Cafe covers early stage startup and VC news, founder advice, and resources for tech entrepreneurs with a spotlight on Israel and Europe.</description>
	<lastBuildDate>Fri, 13 Nov 2026 13:22:00 +0000</lastBuildDate>
	<language>en-US</language>
	<sy:updatePeriod>
	hourly	</sy:updatePeriod>
	<sy:updateFrequency>
	1	</sy:updateFrequency>
	

<image>
	<url>https://www.vccafe.com/wp-content/uploads/2019/05/favicon-1.ico</url>
	<title>VC Cafe</title>
	<link>https://www.vccafe.com</link>
	<width>32</width>
	<height>32</height>
</image> 
<site xmlns="com-wordpress:feed-additions:1">6646906</site>	<itunes:explicit>no</itunes:explicit><itunes:subtitle>Weekly insights on Israeli tech, pre-seed investing, and AI — by Eze Vidra, Managing Partner at Remagine Ventures</itunes:subtitle><item>
		<title>Weekly Firgun Newsletter – November 13 2026</title>
		<link>https://www.vccafe.com/weekly-firgun-newsletter-november-13-2026/</link>
					<comments>https://www.vccafe.com/weekly-firgun-newsletter-november-13-2026/#respond</comments>
		
		<dc:creator><![CDATA[Eze Vidra]]></dc:creator>
		<pubDate>Fri, 13 Nov 2026 13:21:58 +0000</pubDate>
				<category><![CDATA[Firgun]]></category>
		<category><![CDATA[agentic commerce]]></category>
		<category><![CDATA[AI agents]]></category>
		<category><![CDATA[AI Infrastructure]]></category>
		<category><![CDATA[artificial intelligence]]></category>
		<category><![CDATA[Cybersecurity]]></category>
		<category><![CDATA[data centres]]></category>
		<category><![CDATA[generative AI]]></category>
		<category><![CDATA[Israeli startups]]></category>
		<category><![CDATA[Israeli tech]]></category>
		<category><![CDATA[Mergers and acquisitions]]></category>
		<category><![CDATA[pre-seed]]></category>
		<category><![CDATA[Remagine Ventures]]></category>
		<category><![CDATA[seed funding]]></category>
		<category><![CDATA[Startup Funding]]></category>
		<category><![CDATA[Startup nation]]></category>
		<category><![CDATA[venture capital]]></category>
		<guid isPermaLink="false">https://www.vccafe.com/?p=14185</guid>

					<description><![CDATA[The latest #FIRGUN is live: huge Israeli funding rounds, startups emerging from stealth, and a reported $150M–$200M for a cybersecurity startup that hasn’t even started yet. ? Plus Jensen’s departure from NVIDIA, Anthropic’s IPO ambitions, two new Israeli startup landscapes and more.]]></description>
										<content:encoded><![CDATA[<p class="wp-block-paragraph">It&#8217;s #FIRGUN time!</p>



<p class="wp-block-paragraph">The original version of this newsletter was <a href="https://www.linkedin.com/pulse/weekly-firgun-newsletter-eze-vidra-c7jye/?trackingId=luLan7ijSciQxl8jnYLb3Q%3D%3D" data-type="link" data-id="https://www.linkedin.com/pulse/weekly-firgun-newsletter-eze-vidra-c7jye/?trackingId=luLan7ijSciQxl8jnYLb3Q%3D%3D" target="_blank" rel="noopener">published on Linkedin</a> to 8,600+ subscribers. Sign up for free to get the next one in your inbox. </p>



<p class="wp-block-paragraph">Firgun (pronounced “feer-goon”) is an Israeli term describing genuine, unselfish joy in someone else’s success. It’s basically the opposite of schadenfreude. Every week, we celebrate the wins of our startup community. If someone forwarded this to you, previous editions are on <a href="https://www.vccafe.com/">VC Cafe</a>.</p>



<p class="wp-block-paragraph">Israel took another step towards its next government this week, with President Isaac Herzog <a href="https://www.president.gov.il/en/president-herzog-tasks-gadi-eisenkot-with-forming-a-government/" target="_blank" rel="noopener">giving Gadi Eisenkot the mandate to form a coalition</a>. A mandate is not a government, and bringing the potential partners together remains a substantial challenge. My hope is that the outcome gives the country, and the founders building its future, more stability, confidence in its institutions and room to plan beyond the next crisis.</p>



<p class="wp-block-paragraph">In the wider tech world, <a href="https://www.linkedin.com/in/jenhsunhuang/" target="_blank" rel="noopener">Jensen Huang</a>’s resignation from <a href="https://www.linkedin.com/company/nvidia/" target="_blank" rel="noopener">NVIDIA</a> after 33 years dominated the headlines. CFO <a href="https://www.linkedin.com/in/colette-kress-22328346/" target="_blank" rel="noopener">Colette Kress</a> <a href="https://nvidianews.nvidia.com/news/nvidia-announces-ceo-transition" target="_blank" rel="noopener">has taken over as interim CEO</a>, with the board searching for a permanent successor. Huang’s contribution to modern computing is extraordinary, and recognising that does not put a founder beyond accountability. Beyond the immediate market reaction, I’m watching what the transition means for Nvidia&#8217;s startup investments, customer financing and relationships across the AI ecosystem. The company says its product roadmap and customer commitments will continue without interruption. <a href="https://www.linkedin.com/company/anthropicresearch/" target="_blank" rel="noopener">Anthropic</a>, meanwhile, <a href="https://www.anthropic.com/news/roadshow-launch" target="_blank" rel="noopener">launched its proposed IPO roadshow</a>, with 825 million shares offered at an expected $70–$78 range and a planned Nasdaq <span class="mywpglossary-term-def" data-title="listing" data-url="https://www.vccafe.com/startup-and-vc-glossary/#listing">listing</span> under ANT. It also introduced <a href="https://www.anthropic.com/claude-fable-and-mythos-5-2" target="_blank" rel="noopener">Claude Fable 5.2 and Mythos 5.2</a>, while <a href="https://www.linkedin.com/company/google/" target="_blank" rel="noopener">Google</a> launched <a href="https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-4-hydrogen/" target="_blank" rel="noopener">Gemini 4 Hydrogen</a>. The competition increasingly involves choosing the right model for each part of a job, balancing capability, speed and cost. In creative AI, <a href="https://www.linkedin.com/company/adobe/" target="_blank" rel="noopener">Adobe</a> and <a href="https://www.linkedin.com/company/the-walt-disney-company/" target="_blank" rel="noopener">The Walt Disney Company</a> <a href="https://news.adobe.com/news/2026/11/adobe-and-the-walt-disney-company-alliance" target="_blank" rel="noopener">announced a multi-year alliance</a>, including custom Firefly Foundry models tailored to Disney’s properties and creative workflows. <a href="https://www.linkedin.com/company/meta/" target="_blank" rel="noopener">Meta</a>’s Muse added calls to US businesses for paying subscribers. The ambition keeps growing, along with the need to make these systems dependable. On the China side, <a href="https://www.linkedin.com/company/bytedance/" target="_blank" rel="noopener">ByteDance</a> co-founder said that a Muse-style personal <span class="mywpglossary-term-def" data-title="ai-agent" data-url="https://www.vccafe.com/startup-and-vc-glossary/#ai-agent">AI agent</span> is coming within weeks and that the company expects to reach the frontier by 2029 to 2030. while <a href="https://www.linkedin.com/company/moonshotai/" target="_blank" rel="noopener">Moonshot AI</a>&#8216;s <a href="https://www.linkedin.com/company/kimi/" target="_blank" rel="noopener">Kimi</a> K3.1 Pro weights are now reportedly on Hugging Face. Elsewhere, <a href="https://www.linkedin.com/company/googledeepmind/" target="_blank" rel="noopener">Google DeepMind</a> released <a href="https://deepmind.google/blog/alphagenome-2-deciphering-how-cells-read-dna-to-better-understand-disease/" target="_blank" rel="noopener">AlphaGenome 2</a>, which predicts how DNA variants behave in specific cell types and is available for non-commercial research. <a href="https://www.linkedin.com/company/openai/" target="_blank" rel="noopener">OpenAI</a> reportedly closed about $11 billion of a round that could reach $20 billion at a $1.25 trillion <span class="mywpglossary-term-def" data-title="valuation" data-url="https://www.vccafe.com/startup-and-vc-glossary/#valuation">valuation</span>. The money kept moving too: <a href="https://www.linkedin.com/company/mercor-ai/" target="_blank" rel="noopener">Mercor</a> raised <a href="https://techstartups.com/2026/11/11/mercor-raises-750m-at-20b-valuation-doubling-its-worth-in-just-over-a-year-with-backing-from-nvidia-salesforce/" target="_blank" rel="noopener">$750 million at a $20 billion valuation</a>, <a href="https://www.linkedin.com/company/path-robotics/" target="_blank" rel="noopener">Path Robotics</a> raised $300 million at $2.45 billion, <a href="https://www.linkedin.com/company/langchain/" target="_blank" rel="noopener"><strong>LangChain</strong></a> reportedly raised $300 million at a $3 billion <span class="mywpglossary-term-def" data-title="valuation" data-url="https://www.vccafe.com/startup-and-vc-glossary/#valuation">valuation</span>, <a href="https://www.linkedin.com/company/coreweave/" target="_blank" rel="noopener"><strong>CoreWeave</strong></a> reported $3.66 billion in quarterly revenue, <a href="https://www.linkedin.com/company/nominalinc/" target="_blank" rel="noopener">Nominal</a> raised $240 million at $3 billion, Dutch defence-tech <a href="https://www.linkedin.com/company/onodrim-industries/" target="_blank" rel="noopener">Onodrim Industries</a> reported $300 million <span class="mywpglossary-term-def" data-title="series-a" data-url="https://www.vccafe.com/startup-and-vc-glossary/#series-a">Series A</span> at a $1.3 billion <span class="mywpglossary-term-def" data-title="valuation" data-url="https://www.vccafe.com/startup-and-vc-glossary/#valuation">valuation</span>, and <a href="https://www.linkedin.com/company/rox-data-corp/" target="_blank" rel="noopener">Rox</a> <a href="https://fortune.com/2026/11/11/rox-series-b-100-million-funding-take-on-salesforce/" target="_blank" rel="noopener">$100 million at $1.3 billion</a> to take on Salesforce, while <a href="https://www.linkedin.com/company/mavria/" target="_blank" rel="noopener">Mavria</a>, which builds factory-made data centres, is reportedly valued at about $1 billion.</p>



<h2 id="quote-of-the-week" class="wp-block-heading">QUOTE OF THE WEEK</h2>



<blockquote class="wp-block-quote is-layout-flow wp-block-quote-is-layout-flow">
<p class="wp-block-paragraph">“Character is like a tree, and reputation like its shadow.” &#8211; Abraham Lincoln</p>
</blockquote>



<h2 id="new-funding-rounds" class="wp-block-heading">NEW FUNDING ROUNDS</h2>



<ul class="wp-block-list">
<li>Kol hakavod <a href="https://www.linkedin.com/in/hagai-shapira/" target="_blank" rel="noopener">Hagai Shapira</a>, <a href="https://www.linkedin.com/in/eldad-rudich/" target="_blank" rel="noopener">Eldad Rudich</a> and team <a href="https://www.linkedin.com/company/daylightsecurity/" target="_blank" rel="noopener">Daylight Security</a> on an $80 million <span class="mywpglossary-term-def" data-title="series-b" data-url="https://www.vccafe.com/startup-and-vc-glossary/#series-b">Series B</span> to combine AI agents with human security experts to detect, investigate and respond to cyber threats!</li>



<li>Well done <a href="https://www.linkedin.com/in/avner-aharoni/" target="_blank" rel="noopener">Avner Aharoni</a>, <a href="https://www.linkedin.com/in/evgeney-ryzhyk/" target="_blank" rel="noopener">Evgeney Ryzhyk</a>, <a href="https://www.linkedin.com/in/sloutsky/" target="_blank" rel="noopener">Alexander Sloutsky</a>, <a href="https://www.linkedin.com/in/y0nil/" target="_blank" rel="noopener">Yoni Leibowitz</a> and team <a href="https://www.linkedin.com/company/edelbyte-io/" target="_blank" rel="noopener">Edelbyte</a> on emerging from stealth with $70 million raised across two rounds: a previously raised $25 million seed and a $45 million <span class="mywpglossary-term-def" data-title="series-a" data-url="https://www.vccafe.com/startup-and-vc-glossary/#series-a">Series A</span> for their an AI-native unified data <span class="mywpglossary-term-def" data-title="platform" data-url="https://www.vccafe.com/startup-and-vc-glossary/#platform">platform</span> for high-speed search and <span class="mywpglossary-term-def" data-title="big-data" data-url="https://www.vccafe.com/startup-and-vc-glossary/#big-data">big data</span> <span class="mywpglossary-term-def" data-title="analytics" data-url="https://www.vccafe.com/startup-and-vc-glossary/#analytics">analytics</span>!</li>



<li>Mabruk <a href="https://www.linkedin.com/in/yehuda-niv-spines/" target="_blank" rel="noopener">Yehuda Niv</a> and team <a href="https://www.linkedin.com/company/spines-technologies-ltd/" target="_blank" rel="noopener">Spines</a> on securing $64 million in <span class="mywpglossary-term-def" data-title="growth-capital" data-url="https://www.vccafe.com/startup-and-vc-glossary/#growth-capital">growth capital</span> for your AI publishing <span class="mywpglossary-term-def" data-title="platform" data-url="https://www.vccafe.com/startup-and-vc-glossary/#platform">platform</span>: $29 million in Series A-1 plus a $35 million financing facility to help authors take their manuscripts through editing, design, production and distribution, and reports a $50 million revenue run rate. A reminder that Israeli AI extends well beyond cyber!</li>



<li>Kudos <a href="https://www.linkedin.com/in/yuval-zacharia/" target="_blank" rel="noopener">Yuval Zacharia</a>, <a href="https://www.linkedin.com/in/aner-morag/" target="_blank" rel="noopener">Aner Morag</a>, <a href="https://www.linkedin.com/in/boris-danilovich/" target="_blank" rel="noopener">Boris Danilovich</a> and team <a href="https://www.linkedin.com/company/bay-io/" target="_blank" rel="noopener">Bay</a> on coming out of stealth with $50 million in total funding. That includes a $37 million <span class="mywpglossary-term-def" data-title="series-a" data-url="https://www.vccafe.com/startup-and-vc-glossary/#series-a">Series A</span> and a previously undisclosed $13 million seed round tackle the problem of agents having legitimate permissions and still take actions that put a business at risk!</li>



<li>Good stuff <a href="https://www.linkedin.com/in/ziv-kedem-327960/" target="_blank" rel="noopener">Ziv Kedem</a>, <a href="https://www.linkedin.com/in/odedkedem/" target="_blank" rel="noopener">Oded Kedem</a>, <a href="https://www.linkedin.com/in/avi-raichel-38a4621/" target="_blank" rel="noopener">Avi Raichel</a> and team <a href="https://www.linkedin.com/company/dreeze/" target="_blank" rel="noopener">Dreeze</a> on a $33 million seed round to take on cloud application recovery, including the infrastructure, configurations and dependencies needed to get a business running again!</li>



<li>Congrats <a href="https://www.linkedin.com/in/ilan-kadar/" target="_blank" rel="noopener">Ilan Kadar</a>, <a href="https://www.linkedin.com/in/elad-levi-a938a3121/" target="_blank" rel="noopener">Elad Levi</a> and team <a href="https://www.linkedin.com/company/plurai/" target="_blank" rel="noopener">Plurai</a> on a $13 million seed round to apply lessons from autonomous-driving simulation to <span class="mywpglossary-term-def" data-title="enterprise" data-url="https://www.vccafe.com/startup-and-vc-glossary/#enterprise">enterprise</span> agents, finding failures in testing before customers encounter them in production.</li>



<li>Impressive <span class="mywpglossary-term-def" data-title="milestone" data-url="https://www.vccafe.com/startup-and-vc-glossary/#milestone">milestone</span> <a href="https://www.linkedin.com/in/idan-nurick-aa765a127/" target="_blank" rel="noopener">Idan Nurick</a>, <a href="https://www.linkedin.com/in/ehud-schneorson-5b104b144/" target="_blank" rel="noopener">Ehud Schneorson</a> and team Ribon (still in stealth) on raising undisclosed funding (ranging up to $200M in tranches) intro a company <a href="https://en.globes.co.il/en/article-ribon-given-15b-valuation-in-unique-investment-structure-1001562902#:~:text=The%20third%20tranche%2C%20which%20values%20the%20company,on%20the%20reputation%20of%20its%20high%2Dprofile%20founders." target="_blank" rel="noopener">before it&#8217;s even founded</a>!</li>
</ul>



<h3 id="exits" class="wp-block-heading">EXITS</h3>



<ul class="wp-block-list">
<li>Mazal tov <a href="https://www.linkedin.com/in/dsarig/" target="_blank" rel="noopener">Dor Sarig</a>, <a href="https://www.linkedin.com/in/zivk/" target="_blank" rel="noopener">Ziv Karliner</a> and team <a href="https://www.linkedin.com/company/pillarsecurity/" target="_blank" rel="noopener">Pillar Security</a> on the reported agreement to be acquired by <a href="https://www.linkedin.com/company/vanta-security/" target="_blank" rel="noopener">Vanta</a> for approximately $150 million, in cash and shares to help connect AI governance policies with technical controls!</li>



<li>Well done <a href="https://www.linkedin.com/in/yonis/" target="_blank" rel="noopener">Yoni Shtein</a>, <a href="https://www.linkedin.com/in/yaelpeled/" target="_blank" rel="noopener">Yael Peled Adam</a> and team <a href="https://www.linkedin.com/company/lagunahealth/" target="_blank" rel="noopener">Laguna Health</a> on the <span class="mywpglossary-term-def" data-title="acquisition" data-url="https://www.vccafe.com/startup-and-vc-glossary/#acquisition">acquisition</span> by <a href="https://www.linkedin.com/company/elligint-health/" target="_blank" rel="noopener">Elligint Health</a> for an undisclosed sum to empower healthcare management teams to automate clinical documentation, monitor call quality, and support member interactions</li>



<li>Kudos <a href="https://www.linkedin.com/in/yoav-berlatzky-6684aa26/" target="_blank" rel="noopener">Yoav Berlatzky</a> and team <a href="https://www.linkedin.com/company/pxe-holographic-imaging/" target="_blank" rel="noopener">PxE Holographic Imaging</a> on the <span class="mywpglossary-term-def" data-title="acquisition" data-url="https://www.vccafe.com/startup-and-vc-glossary/#acquisition">acquisition</span> by <a href="https://www.linkedin.com/company/metalenz/" target="_blank" rel="noopener">metalenz</a> to capture colour and depth information using a single camera using the company’s Holocoder technology. Terms were not disclosed.</li>
</ul>



<h3 id="new-funds" class="wp-block-heading">NEW FUNDS</h3>



<ul class="wp-block-list">
<li>Congrats <a href="https://www.linkedin.com/in/eric-reiner/" target="_blank" rel="noopener">Eric Reiner</a>, <a href="https://www.linkedin.com/in/adamvalkin/" target="_blank" rel="noopener">Adam Valkin</a> and team <a href="https://www.linkedin.com/company/vine-ventures-l-p/" target="_blank" rel="noopener">Vine Ventures</a> on closing a $236 million Fund III, to continue investing across the US–Israel <span class="mywpglossary-term-def" data-title="early-stage" data-url="https://www.vccafe.com/startup-and-vc-glossary/#early-stage">early stage</span> technology ecosystem!</li>



<li>Well done <a href="https://www.linkedin.com/in/drornahumi/" target="_blank" rel="noopener">Dror Nahumi</a>, <a href="https://www.linkedin.com/in/assafharel/" target="_blank" rel="noopener">Assaf Harel</a> and the global team at <a href="https://www.linkedin.com/company/norwest-venture-partners/" target="_blank" rel="noopener">Norwest</a> on announcing a $4 billion eighteenth fund, spanning venture and <span class="mywpglossary-term-def" data-title="growth-equity" data-url="https://www.vccafe.com/startup-and-vc-glossary/#growth-equity">growth <span class="mywpglossary-term-def" data-title="equity" data-url="https://www.vccafe.com/startup-and-vc-glossary/#equity">equity</span></span> across technology, healthcare and consumer, reaffirming the firm’s commitment to Israel and India!</li>
</ul>



<h2 id="links-for-your-browser" class="wp-block-heading">LINKS FOR YOUR BROWSER</h2>



<h2 id="vc-cafe-remagine-ventures-israelvc" class="wp-block-heading">VC CAFE / REMAGINE VENTURES / ISRAELVC</h2>



<ul class="wp-block-list">
<li><a href="https://www.vccafe.com/when-ai-builds-ai/"><strong>When AI Builds AI</strong></a> &#8211; What changes when AI contributes to developing its successors? I explore the feedback loop, the implications for oversight and where startups can <span class="mywpglossary-term-def" data-title="build" data-url="https://www.vccafe.com/startup-and-vc-glossary/#build">build</span> as development accelerates.</li>



<li><a href="https://www.vccafe.com/what-did-your-ai-agent-just-do/"><strong>What Did Your AI Agent Just Do?</strong></a> &#8211; Agents can send emails, spend money and act in our name. We need to be able to reconstruct their actions, understand who authorised them and intervene when necessary. This week’s deals show how commercially relevant that problem is becoming.</li>



<li><a href="https://www.vccafe.com/ai-safety-who-should-we-believe/"><strong>AI Safety: Who Should We Believe?</strong></a> &#8211; The safety debate has well-funded advocates on every side. A look at how to separate evidence from advocacy, examine incentives and understand where Israeli startups can contribute.</li>



<li><strong>Bonus:</strong> Think you know your startups, VC and AI? Test yourself with the <a href="https://israelvc.com/quiz/" target="_blank" rel="noopener">IsraelVC quiz</a> and the <a href="https://www.vccafe.com/quiz/">VC Cafe quiz</a>, then share your score! And if you know anyone looking for job with Israeli startups, we now link to <a href="https://www.israelvc.com/jobs" target="_blank" rel="noopener">3,079 job records</a> across 318 venture-backed startups!</li>
</ul>



<h3 id="israel" class="wp-block-heading">ISRAEL</h3>



<ul class="wp-block-list">
<li><a href="https://www.timesofisrael.com/israel-must-build-own-ai-infrastructure-or-risk-dependence-on-foreign-tech-ai-czar-warns/" target="_blank" rel="noopener">Israel must build own AI infrastructure or risk dependence on foreign tech, AI czar warns</a> &#8211; <a href="https://www.linkedin.com/in/erez-askal/" target="_blank" rel="noopener">Erez Askal</a> says independent infrastructure is critical for energy, water, and healthcare, which will run on AI within 7 years; plan requires government funding; “Israel is one of the three countries at the frontier of AI, after the US and China. Israel is in a similar league in terms of technological capabilities – not in terms of the number of people. Nvidia’s chips are either designed entirely in Israel or involve Israeli teams in their development.”</li>



<li><a href="https://www.calcalistech.com/ctechnews/article/hyvm00ez0gl" target="_blank" rel="noopener">CrowdStrike&#8217;s business chief: &#8220;We&#8217;re long on Israel and we&#8217;re not going anywhere&#8221;</a> &#8211; The <span class="mywpglossary-term-def" data-title="cybersecurity" data-url="https://www.vccafe.com/startup-and-vc-glossary/#cybersecurity">cybersecurity</span> giant has continued acquiring Israeli startups through the war with Iran and now employs more than 500 people in the country. Chief Business Officer Daniel Bernard says partnerships, rather than acquisitions, are the main opportunity for local startups, while warning that AI is making strong products easier to replicate.</li>



<li><a href="https://www.ynetnews.com/business/article/skz8vsgafe" target="_blank" rel="noopener">From Startup Nation to Space Nation: Israel&#8217;s space industry goes global</a> &#8211; For the first time, Israel will have a national pavilion at Space Tech Expo Europe in Germany, bringing together defense giants and startups; industry leaders say falling launch costs and growing demand for satellite technology could open new opportunities for Israeli innovation</li>
</ul>



<h2 id="global" class="wp-block-heading">GLOBAL</h2>



<ul class="wp-block-list">
<li><a href="https://carta.com/data/state-of-pre-seed-q3-2026/" target="_blank" rel="noopener"><strong>State of Pre-Seed: Q3 2026</strong></a> <strong>&#8211;</strong> More capital is flowing into fewer <span class="mywpglossary-term-def" data-title="pre-seed" data-url="https://www.vccafe.com/startup-and-vc-glossary/#pre-seed">pre-seed</span> financing instruments. For US deals of $2.5 million or more, the median post-money <span class="mywpglossary-term-def" data-title="safe" data-url="https://www.vccafe.com/startup-and-vc-glossary/#safe">SAFE</span> cap rose from $30 million to $40 million year over year. Worth reading for founders and first-cheque investors, remembering that a <span class="mywpglossary-term-def" data-title="safe" data-url="https://www.vccafe.com/startup-and-vc-glossary/#safe">SAFE</span> cap is not a priced-round <span class="mywpglossary-term-def" data-title="valuation" data-url="https://www.vccafe.com/startup-and-vc-glossary/#valuation">valuation</span>, by <a href="https://www.linkedin.com/company/carta--/" target="_blank" rel="noopener">Carta</a></li>



<li><a href="https://news.crunchbase.com/venture/series-a-2026-startup-capital-up-deals-steady/" target="_blank" rel="noopener"><strong>Series A in 2026: Capital Is Flowing Faster, but Fewer Startups Are Making the Cut</strong></a> <strong>&#8211;</strong> US <span class="mywpglossary-term-def" data-title="series-a" data-url="https://www.vccafe.com/startup-and-vc-glossary/#series-a">Series A</span> funding reached $42 billion through October, roughly 60% higher year over year, while deal counts remained broadly in line with recent years &#8211; by <a href="https://www.linkedin.com/company/crunchbase/" target="_blank" rel="noopener">Crunchbase</a></li>



<li><a href="https://www.bvp.com/atlas/the-ai-100-benchmarks-for-the-third-supercycle" target="_blank" rel="noopener"><strong>The AI 100: Benchmarks for the Third Supercycle</strong></a> <strong>&#8211;</strong> The inaugural ranking puts its selected companies’ aggregate private <span class="mywpglossary-term-def" data-title="valuation" data-url="https://www.vccafe.com/startup-and-vc-glossary/#valuation">valuation</span> at $3.6 trillion, with median annual revenue growth of 275% and a median ARR multiple of 23x. Useful benchmarks for the leaders, rather than a picture of the average AI startup. By <a href="https://www.linkedin.com/company/bessemer-venture-partners/" target="_blank" rel="noopener">Bessemer Venture Partners</a> and <a href="https://www.linkedin.com/company/newcomermedia/" target="_blank" rel="noopener">Newcomer</a>. Related &#8211; <a href="https://www.leoniscap.com/research/the-2026-leonis-ai-100" target="_blank" rel="noopener">The 2026 Leonis AI 100</a> (by <a href="https://www.linkedin.com/company/leonis-capital-vc/" target="_blank" rel="noopener">Leonis Capital</a>) benchmarks the fastest-growing AI-native startups.</li>



<li><a href="https://a16z.com/welcome-to-the-b2a-economy/" target="_blank" rel="noopener"><strong>Welcome to the B2A Economy</strong></a> <strong>&#8211;</strong> Olivia Moore and Justine Moore from <a href="https://www.linkedin.com/company/a16z/" target="_blank" rel="noopener">Andreessen Horowitz</a> explore how businesses might acquire customers when agents increasingly discover, compare and buy on our behalf. See landscapes below.</li>
</ul>



<h2 id="media-of-the-week" class="wp-block-heading">MEDIA OF THE WEEK</h2>



<p class="wp-block-paragraph">The first map of Israel’s WasteTech ecosystem: 80 companies across the full waste journey, from production and collection to treatment and pollution control. Nearly 60% turn waste into materials, fuels, food ingredients or energy, and most are already running pilots or working with first customers by <a href="https://www.linkedin.com/company/planetech-climate-tech-technologies/" target="_blank" rel="noopener">PLANETech &#8211; Climate Change Technologies</a> &#8211; the map was added to my <a href="https://www.vccafe.com/israeli-startup-landscape-collection-2026/">2026 Israeli landscape collection</a> on <a href="https://www.linkedin.com/company/vc-cafe/" target="_blank" rel="noopener">VC Cafe</a>.</p>



<figure class="wp-block-image"><img decoding="async" src="https://media.licdn.com/dms/image/v2/D4E12AQGobjyfxW8HXg/article-inline_image-shrink_1000_1488/B4EaHU286fKwAM-/0/1794572539094?e=1796256000&amp;v=beta&amp;t=qeQ_2zkCRDdE9e1yCMJ_ORzGGZnN_ZGBRnGVKmW4Y-w" alt="Article content"/></figure>



<p class="wp-block-paragraph">Building on <a href="https://www.calcalistech.com/ctechnews/article/rkzdu17bwe" target="_blank" rel="noopener">Remagine Ventures’ Q4 2025 landscape</a> (which originally profiled 143 companies), direct-to-consumer mobile gaming <span class="mywpglossary-term-def" data-title="monetization" data-url="https://www.vccafe.com/startup-and-vc-glossary/#monetization">monetization</span> <span class="mywpglossary-term-def" data-title="platform" data-url="https://www.vccafe.com/startup-and-vc-glossary/#platform">platform</span> <a href="https://www.linkedin.com/feed/update/urn:li:activity:7525585341085372416/" target="_blank" rel="noopener">Appcharge has published</a> an updated 2026 Israeli Gaming Landscape capturing the active studios and infrastructure powering the sector today. It was also added to my <a href="https://www.vccafe.com/israeli-startup-landscape-collection-2026/">2026 Israeli startup landscape collectio</a>n on <a href="https://www.linkedin.com/company/vc-cafe/" target="_blank" rel="noopener">VC Cafe</a>.</p>



<figure class="wp-block-image"><img decoding="async" src="https://media.licdn.com/dms/image/v2/D4E12AQFDF83YH4p6Ww/article-inline_image-shrink_1500_2232/B4EaHU5eEEHQAQ-/0/1794573197867?e=1796256000&amp;v=beta&amp;t=z3psQEG8pDnuFxMoMNkOahdrQGk7011Qd1yJbyv3jYA" alt="Article content"/></figure>



<p class="wp-block-paragraph">Where AI’s private-market value sits. <a href="https://www.linkedin.com/company/bessemer-venture-partners/" target="_blank" rel="noopener">Bessemer Venture Partners</a> AI 100 charts show frontier labs and foundation-model builders representing 9% of the list but 55% of its aggregate <span class="mywpglossary-term-def" data-title="valuation" data-url="https://www.vccafe.com/startup-and-vc-glossary/#valuation">valuation</span> (<a href="https://ai100.newcomer.co/" target="_blank" rel="noopener">source</a>)</p>



<figure class="wp-block-image"><img decoding="async" src="https://media.licdn.com/dms/image/v2/D4E12AQHx0oMxY-Zn6g/article-inline_image-shrink_1500_2232/B4EaHU2H4MKgAQ-/0/1794572320546?e=1796256000&amp;v=beta&amp;t=jvpbWxymRsVA_Jnuk6wPFXP312wA2oWoUyYo5AMOcCA" alt="Article content"/></figure>



<p class="wp-block-paragraph">When the central bank of AI changes chairs&#8230; on the uncertainty that surrounds future investments by <a href="https://www.linkedin.com/company/nvidia/" target="_blank" rel="noopener">NVIDIA</a> under new leadership and its expected impact on the AI industry &#8211; by <a href="https://www.linkedin.com/in/tomasztunguz/" target="_blank" rel="noopener">Tomasz Tunguz</a> (<a href="https://x.com/ttunguz/status/2120931491115135287">source</a>)</p>



<figure class="wp-block-image"><img decoding="async" src="https://media.licdn.com/dms/image/v2/D4E12AQERVGcKzU1HVg/article-inline_image-shrink_1500_2232/B4EaHU6P54IEAQ-/0/1794573401963?e=1796256000&amp;v=beta&amp;t=IJ88ByWB6aW7iHGrDck40OglhM053Djw6pIBsPSONME" alt="Article content"/></figure>



<p class="wp-block-paragraph"><a href="https://www.linkedin.com/company/a16z/" target="_blank" rel="noopener"><strong>Andreessen Horowitz</strong></a> (Olivia Moore and <a href="https://www.linkedin.com/in/justinemoore94/" target="_blank" rel="noopener">Justine Moore</a> ) has introduced the <strong>B2A (Business-to-Agent) Economy</strong> thesis alongside two market maps profiling the emerging ecosystem where AI agents replace humans as the primary users, buyers, and traffic drivers of the internet &#8211; Welcome to the B2A (Business to Agent) economy (<a href="https://a16z.com/welcome-to-the-b2a-economy/" target="_blank" rel="noopener">source</a>)</p>



<figure class="wp-block-image"><img decoding="async" src="https://media.licdn.com/dms/image/v2/D4E12AQHFUBXZy0u5UQ/article-inline_image-shrink_1500_2232/B4EaHU6tKGJwAU-/0/1794573522034?e=1796256000&amp;v=beta&amp;t=GxWb4M23-mMGkW-VTWucGgNYC3umJtEjefsVXbvxH1Y" alt="Article content"/></figure>



<p class="wp-block-paragraph">According to new benchmarks published by <a href="https://www.linkedin.com/company/zapier/" target="_blank" rel="noopener">Zapier</a>, even the best AI agents only get complex business workflows right about half the time. Worse, when they fail, they often report success! 72%–91% of failed runs across the models examined included a claim that the task was complete. (<a href="https://www.linkedin.com/feed/update/urn:li:activity:7526325715148128256/?dashCommentUrn=urn%3Ali%3Afsd_comment%3A%287526697044275646464%2Curn%3Ali%3Aactivity%3A7526325715148128256%29" target="_blank" rel="noopener">source</a>) and <a href="https://zapier.com/benchmarks" target="_blank" rel="noopener">link to research</a></p>



<figure class="wp-block-image"><img decoding="async" src="https://media.licdn.com/dms/image/v2/D4E12AQGGzZvlpxY_1g/article-inline_image-shrink_1000_1488/B4EaHU9DLVIAAM-/0/1794574136278?e=1796256000&amp;v=beta&amp;t=x9Mo4lGK37zvzvNtGt6jIyW2zLrBm6SP0GayB5EWQ08" alt="Article content"/></figure>



<p class="wp-block-paragraph">Maybe Jensen wishes this was actually a movie, but the memes just kept coming this week.</p>



<figure class="wp-block-embed is-type-video is-provider-youtube wp-block-embed-youtube wp-embed-aspect-16-9 wp-has-aspect-ratio"><div class="wp-block-embed__wrapper">
<iframe title="THE MATRIX OF JENSEN HUANG" width="1290" height="726" src="https://www.youtube.com/embed/4wOz5cnu1bM?feature=oembed" frameborder="0" allow="accelerometer; autoplay; clipboard-write; encrypted-media; gyroscope; picture-in-picture; web-share" referrerpolicy="strict-origin-when-cross-origin" allowfullscreen></iframe>
</div></figure>



<p class="wp-block-paragraph">But joking aside, the AI boom is in many ways linked to Jensen personally, and bedroom/marital transgressions aside, his contribution to the economy and to technology is immense. Credit to <a href="https://www.linkedin.com/in/shaun-maguire-c/" target="_blank" rel="noopener">Shaun Maguire</a> for calling it out when everyone is busy roasting Jensen. (<a href="https://x.com/shaunmmaguire/status/2120888640184222081">source</a>)</p>



<figure class="wp-block-embed is-type-rich is-provider-x wp-block-embed-x"><div class="wp-block-embed__wrapper">
<div class="embed-x"><blockquote class="twitter-tweet" data-width="550" data-dnt="true"><p lang="en" dir="ltr">Jensen is out<br><br>Infidelity is obviously bad<br><br>But his contributions to American innovation and AI leadership cannot be overstated<br><br>Jensen started NVIDIA in 1993, where he pioneered the graphics chip market when this was non-obvious<br><br>Back in 2012, Ilya Sustkever trained AlexNet on… <a href="https://t.co/hgcrNirnWV">https://t.co/hgcrNirnWV</a></p>&mdash; Shaun Maguire (@shaunmmaguire) <a href="https://x.com/shaunmmaguire/status/2120888640184222081?ref_src=twsrc%5Etfw">November 12, 2026</a></blockquote><script async src="https://platform.x.com/widgets.js" charset="utf-8"></script></div>
</div></figure>



<p class="wp-block-paragraph">That’s all for this week. A practical form of firgun: if a founder you know is trying to turn a pilot into a paying customer, introduce them to someone who owns the problem they’re solving. An engaged customer and honest feedback can do more for a <span class="mywpglossary-term-def" data-title="young-company" data-url="https://www.vccafe.com/startup-and-vc-glossary/#young-company">young company</span> than another round of applause.</p>



<p class="wp-block-paragraph">There’s a lot to celebrate in our “small” community. Keep on creating!</p>



<p class="wp-block-paragraph">Eze Vidra</p>]]></content:encoded>
					
					<wfw:commentRss>https://www.vccafe.com/weekly-firgun-newsletter-november-13-2026/feed/</wfw:commentRss>
			<slash:comments>0</slash:comments>
		
		
		<post-id xmlns="com-wordpress:feed-additions:1">14185</post-id>	</item>
		<item>
		<title>AI Safety: Who Should We Believe?</title>
		<link>https://www.vccafe.com/ai-safety-who-should-we-believe/</link>
					<comments>https://www.vccafe.com/ai-safety-who-should-we-believe/#respond</comments>
		
		<dc:creator><![CDATA[Eze Vidra]]></dc:creator>
		<pubDate>Tue, 10 Nov 2026 16:14:25 +0000</pubDate>
				<category><![CDATA[AI Safety]]></category>
		<category><![CDATA[Artificial Intelligence]]></category>
		<category><![CDATA[AI agents]]></category>
		<category><![CDATA[AI governance]]></category>
		<category><![CDATA[AI Security]]></category>
		<category><![CDATA[Anthropic]]></category>
		<category><![CDATA[artificial intelligence]]></category>
		<category><![CDATA[Cybersecurity]]></category>
		<category><![CDATA[frontier models]]></category>
		<category><![CDATA[Israeli startups]]></category>
		<category><![CDATA[Life 3.0]]></category>
		<category><![CDATA[Max Tegmark]]></category>
		<category><![CDATA[Remagine Ventures]]></category>
		<category><![CDATA[venture capital]]></category>
		<guid isPermaLink="false">https://www.vccafe.com/?p=14170</guid>

					<description><![CDATA[AI safety has professional champions on every side, with funding, careers and commercial interests shaping the debate. From a venture investor’s perspective, how do we separate evidence from advocacy, assess the trade-offs and identify where Israeli startups can contribute?]]></description>
										<content:encoded><![CDATA[<p class="wp-block-paragraph">I’ve been listening to <a href="https://en.wikipedia.org/wiki/Life_3.0" data-type="link" data-id="https://en.wikipedia.org/wiki/Life_3.0" target="_blank" rel="noopener">Max Tegmark’s <em>Life 3.0</em></a>, and it has made me think about AI safety. More specifically, it has made me wonder how someone outside the research labs is supposed to form a considered opinion about it.</p>



<p class="wp-block-paragraph">Depending on whom you listen to, AI could usher in extraordinary abundance, concentrate power in a handful of companies, undermine livelihoods, or eventually become something humans can no longer control. Sometimes you hear several of these predictions from the same person. Spend 10 minutes scrolling on X and you&#8217;re likely to see both extremes, on the same day.</p>



<p class="wp-block-paragraph">AI doomers:</p>



<figure class="wp-block-embed is-type-rich is-provider-x wp-block-embed-x"><div class="wp-block-embed__wrapper">
<div class="embed-x"><blockquote class="twitter-tweet" data-width="550" data-dnt="true"><p lang="en" dir="ltr">Yann LeCun never did answer this, but regardless, now $1B in AI damage has happened.  (Black October attacks.)  (This should&#39;ve been a trivial prediction, and willingly stipulated by all sides, in any healthy debate.) <a href="https://t.co/GIZGR8EjI2">https://t.co/GIZGR8EjI2</a></p>&mdash; Eliezer Yudkowsky (@allTheYud) <a href="https://x.com/allTheYud/status/2116670754049978630?ref_src=twsrc%5Etfw">October 31, 2026</a></blockquote><script async src="https://platform.x.com/widgets.js" charset="utf-8"></script></div>
</div></figure>



<p class="wp-block-paragraph">or techno-optimists:</p>



<figure class="wp-block-embed is-type-rich is-provider-x wp-block-embed-x"><div class="wp-block-embed__wrapper">
<div class="embed-x"><blockquote class="twitter-tweet" data-width="550" data-dnt="true"><p lang="en" dir="ltr">I didn&#39;t say p(doom) was zero. <br>I said:<br>1. All estimates are pulled out of thin air<br>2. It makes little sense to attribute a probability to an event on which we have agency. If you do so, it&#39;s more of a statement about our collective ability to do the right thing than about the…</p>&mdash; Yann LeCun (@ylecun) <a href="https://x.com/ylecun/status/2046577402264870958?ref_src=twsrc%5Etfw">April 21, 2026</a></blockquote><script async src="https://platform.x.com/widgets.js" charset="utf-8"></script></div>
</div></figure>



<p class="wp-block-paragraph">As a VC investing in AI startups, I have a stake in this discussion. I want founders to <span class="mywpglossary-term-def" data-title="build" data-url="https://www.vccafe.com/startup-and-vc-glossary/#build">build</span> useful companies, and I believe AI creates substantial opportunities. That also means I’m not observing the debate from a position free of incentives. Neither are many of its most prominent participants.</p>



<h2 id="a-debate-with-professional-champions-on-every-side" class="wp-block-heading"><strong>A debate with professional champions on every side</strong></h2>



<p class="wp-block-paragraph">There are people paid to develop AI, sell it and make the case for faster adoption. There are people paid to research its risks, advocate for safeguards and campaign for restrictions. There are investors whose returns depend on continued progress, consultants selling governance, and commentators whose audiences respond to either extraordinary promise or extraordinary danger.</p>



<p class="wp-block-paragraph">Being paid to work on something doesn’t make you wrong. We need people with the expertise and resources to investigate difficult questions. But it is useful to distinguish between someone’s evidence, their interpretation of that evidence and the outcome their organisation exists to pursue.</p>



<p class="wp-block-paragraph">The same scrutiny should apply across the debate. A lab’s commercial incentives don’t automatically invalidate its safety research. A nonprofit’s mission doesn’t automatically make its conclusions impartial. And an investor’s enthusiasm is hardly independent validation of the technology.</p>



<p class="wp-block-paragraph">The uncomfortable part is that <strong>people can be sincere, well informed and influenced by their incentives at the same time.</strong></p>



<p class="wp-block-paragraph">There is another signal worth considering: several people working on AI safety have left the companies building these systems. Some, including OpenAI’s former superalignment co-lead Jan Leike and Anthropic’s Joe Benton, explicitly questioned whether safety was receiving enough priority. Others left for different or less clearly stated reasons, and reorganisations are sometimes bundled into the same “exodus” narrative. These departures don’t prove the most alarming predictions are right. But dismissing them as simply another group of people paid to worry would be too convenient. The useful questions are what they witnessed, what they can substantiate and how their former employers respond.</p>



<h2 id="there-is-real-money-behind-the-discussion" class="wp-block-heading"><strong>There is real money behind the discussion</strong></h2>



<p class="wp-block-paragraph">The figures circulating from <a href="https://aisafety.fund/" target="_blank" rel="noopener">AISafety.fund</a> estimate approximately $1.5 billion from philanthropy, $346 million from governments and $2.1 billion from frontier AI companies for safety in 2026. Together, that is roughly $3.95 billion.</p>



<figure class="wp-block-image aligncenter size-large"><img data-recalc-dims="1" fetchpriority="high" decoding="async" width="1024" height="561" data-attachment-id="14171" data-permalink="https://www.vccafe.com/ai-safety-who-should-we-believe/ai-labs-spend-on-safety/" data-orig-file="https://i0.wp.com/www.vccafe.com/wp-content/uploads/2026/11/AI-labs-spend-on-safety-scaled.png?fit=1600%2C876&amp;ssl=1" data-orig-size="1600,876" data-comments-opened="1" data-image-title="AI labs spend on safety" data-image-description="&lt;p&gt;AI labs spend on safety. Related keywords: pre-seed funding, fundraising strategy, ???? ??? ???, ???? ??? ??????????&lt;/p&gt;
" data-image-caption="&lt;p&gt;AI labs spend on safety for VC Cafe&lt;/p&gt;
" data-large-file="https://i0.wp.com/www.vccafe.com/wp-content/uploads/2026/11/AI-labs-spend-on-safety-scaled.png?fit=1024%2C561&amp;ssl=1" src="https://i0.wp.com/www.vccafe.com/wp-content/uploads/2026/11/AI-labs-spend-on-safety.png?resize=1024%2C561&#038;ssl=1" alt="AI labs spend on safety - pre-seed funding / ???? ??? ???" class="wp-image-14171" srcset="https://i0.wp.com/www.vccafe.com/wp-content/uploads/2026/11/AI-labs-spend-on-safety-scaled.png?resize=1024%2C561&amp;ssl=1 1024w, https://i0.wp.com/www.vccafe.com/wp-content/uploads/2026/11/AI-labs-spend-on-safety-scaled.png?resize=300%2C164&amp;ssl=1 300w, https://i0.wp.com/www.vccafe.com/wp-content/uploads/2026/11/AI-labs-spend-on-safety-scaled.png?resize=767%2C420&amp;ssl=1 767w, https://i0.wp.com/www.vccafe.com/wp-content/uploads/2026/11/AI-labs-spend-on-safety-scaled.png?resize=1536%2C841&amp;ssl=1 1536w, https://i0.wp.com/www.vccafe.com/wp-content/uploads/2026/11/AI-labs-spend-on-safety-scaled.png?resize=2048%2C1122&amp;ssl=1 2048w, https://i0.wp.com/www.vccafe.com/wp-content/uploads/2026/11/AI-labs-spend-on-safety-scaled.png?w=1600&amp;ssl=1 1600w" sizes="(max-width: 1024px) 100vw, 1024px" /><figcaption class="wp-element-caption"><em>Estimated AI safety funding by source. Definitions and coverage matter: the tracker’s 2026 chart covers January–October, and spending totals do not establish how effectively risks are reduced. Source: <a href="https://aisafety.fund/" target="_blank" rel="noopener">AISafety.fund</a></em></figcaption></figure>



<p class="wp-block-paragraph">Alongside those figures sit estimates of $98 billion for AI capabilities and $868 billion for data centres. These are estimates with different definitions and coverage, rather than an audited, like-for-like comparison. The tracker labels its 2026 funding chart January–October.</p>



<p class="wp-block-paragraph">One person looks at those numbers and sees safety receiving a fraction of what is spent advancing the technology. Another sees a multibillion-dollar field with its own institutions, careers and fundraising needs. Both observations can be true.</p>



<p class="wp-block-paragraph">The ratio alone doesn’t tell us whether safety is underfunded, whether the money is being spent effectively, or how much risk any particular intervention reduces. Those are harder questions than comparing the size of two bars on a chart.</p>



<h2 id="even-ai-safety-means-different-things-to-different-people" class="wp-block-heading"><strong>Even “AI safety” means different things to different people</strong></h2>



<p class="wp-block-paragraph">Someone worried about AI-generated fraud may have little in common with someone researching the possibility of a superintelligent system escaping human control. A worker concerned about displacement is asking a different question from a security researcher investigating whether an agent can be tricked into leaking company data.</p>



<p class="wp-block-paragraph">Yet all of these concerns are regularly bundled into the same debate.</p>



<p class="wp-block-paragraph">WIRED’s recent survey of ten experts captures this disagreement, with concerns ranging from biological threats to workers’ protections, alongside differing views about runaway intelligence. There is no single expert position on what matters most.</p>



<p class="wp-block-paragraph">This makes public discussion frustrating. People appear to disagree about the answer when they are often answering different questions. What is possible? What is probable? On what timescale? Who bears the cost? And what should we do about it?</p>



<figure class="wp-block-image aligncenter size-large"><img data-recalc-dims="1" decoding="async" width="1024" height="975" data-attachment-id="14172" data-permalink="https://www.vccafe.com/ai-safety-who-should-we-believe/ai-safety-report-2026-fig-3-1/" data-orig-file="https://i0.wp.com/www.vccafe.com/wp-content/uploads/2026/11/ai-safety-report-2026-fig-3.1-scaled.webp?fit=1600%2C1523&amp;ssl=1" data-orig-size="1600,1523" data-comments-opened="1" data-image-title="ai-safety-report-2026-fig-3.1" data-image-description="&lt;p&gt;ai-safety-report-2026-fig-3.1. Related keywords: pre-seed funding, fundraising strategy, ???? ??? ???, ???? ??? ??????????&lt;/p&gt;
" data-image-caption="&lt;p&gt;ai-safety-report-2026-fig-3.1 for VC Cafe&lt;/p&gt;
" data-large-file="https://i0.wp.com/www.vccafe.com/wp-content/uploads/2026/11/ai-safety-report-2026-fig-3.1-scaled.webp?fit=1024%2C975&amp;ssl=1" src="https://i0.wp.com/www.vccafe.com/wp-content/uploads/2026/11/ai-safety-report-2026-fig-3.1.webp?resize=1024%2C975&#038;ssl=1" alt="ai-safety-report-2026-fig-3.1 - pre-seed funding / ???? ??? ???" class="wp-image-14172" srcset="https://i0.wp.com/www.vccafe.com/wp-content/uploads/2026/11/ai-safety-report-2026-fig-3.1-scaled.webp?resize=1024%2C975&amp;ssl=1 1024w, https://i0.wp.com/www.vccafe.com/wp-content/uploads/2026/11/ai-safety-report-2026-fig-3.1-scaled.webp?resize=300%2C286&amp;ssl=1 300w, https://i0.wp.com/www.vccafe.com/wp-content/uploads/2026/11/ai-safety-report-2026-fig-3.1-scaled.webp?resize=767%2C730&amp;ssl=1 767w, https://i0.wp.com/www.vccafe.com/wp-content/uploads/2026/11/ai-safety-report-2026-fig-3.1-scaled.webp?resize=1536%2C1462&amp;ssl=1 1536w, https://i0.wp.com/www.vccafe.com/wp-content/uploads/2026/11/ai-safety-report-2026-fig-3.1-scaled.webp?w=1600&amp;ssl=1 1600w" sizes="(max-width: 1024px) 100vw, 1024px" /><figcaption class="wp-element-caption"><em>Why agreement is difficult: gaps in scientific understanding, unequal access to information, market incentives and coordination problems all shape AI risk management. Source: International AI Safety Report 2026, Figure 3.1.</em></figcaption></figure>



<h2 id="the-labs-embody-the-contradiction" class="wp-block-heading"><strong>The labs embody the contradiction</strong></h2>



<p class="wp-block-paragraph">Anthropic’s release of Claude Fable 5.2 and Mythos 5.2 brings these questions into focus again. The company describes increasingly capable systems while saying its evaluations suggest high-stakes misalignment risks remain low.</p>



<p class="wp-block-paragraph">Separately, METR’s assessment of AI research at Anthropic suggests that AI assistance is contributing to faster progress in developing subsequent models. The researchers emphasise substantial uncertainty. They also disclose that Anthropic redacted much of the evidence they found most informative, citing concerns about accelerating competitors.</p>



<figure class="wp-block-image aligncenter size-large"><img data-recalc-dims="1" loading="lazy" decoding="async" width="1024" height="850" data-attachment-id="14173" data-permalink="https://www.vccafe.com/ai-safety-who-should-we-believe/aeci-main/" data-orig-file="https://i0.wp.com/www.vccafe.com/wp-content/uploads/2026/11/aeci-main-scaled.png?fit=1600%2C1329&amp;ssl=1" data-orig-size="1600,1329" data-comments-opened="1" data-image-title="aeci-main" data-image-description="&lt;p&gt;aeci-main. Related keywords: pre-seed funding, fundraising strategy, ???? ??? ???, ???? ??? ??????????&lt;/p&gt;
" data-image-caption="&lt;p&gt;aeci-main for VC Cafe&lt;/p&gt;
" data-large-file="https://i0.wp.com/www.vccafe.com/wp-content/uploads/2026/11/aeci-main-scaled.png?fit=1024%2C850&amp;ssl=1" src="https://i0.wp.com/www.vccafe.com/wp-content/uploads/2026/11/aeci-main.png?resize=1024%2C850&#038;ssl=1" alt="aeci-main - pre-seed funding / ???? ??? ???" class="wp-image-14173" srcset="https://i0.wp.com/www.vccafe.com/wp-content/uploads/2026/11/aeci-main-scaled.png?resize=1024%2C850&amp;ssl=1 1024w, https://i0.wp.com/www.vccafe.com/wp-content/uploads/2026/11/aeci-main-scaled.png?resize=300%2C249&amp;ssl=1 300w, https://i0.wp.com/www.vccafe.com/wp-content/uploads/2026/11/aeci-main-scaled.png?resize=767%2C637&amp;ssl=1 767w, https://i0.wp.com/www.vccafe.com/wp-content/uploads/2026/11/aeci-main-scaled.png?resize=1536%2C1275&amp;ssl=1 1536w, https://i0.wp.com/www.vccafe.com/wp-content/uploads/2026/11/aeci-main-scaled.png?w=1600&amp;ssl=1 1600w" sizes="auto, (max-width: 1024px) 100vw, 1024px" /><figcaption class="wp-element-caption"><em>METR found that Anthropic’s internal <span class="mywpglossary-term-def" data-title="benchmark" data-url="https://www.vccafe.com/startup-and-vc-glossary/#benchmark">benchmark</span> scores were improving at approximately twice their earlier rate. This measures progress on selected tasks, rather than the probability of losing control; <span class="mywpglossary-term-def" data-title="attribution" data-url="https://www.vccafe.com/startup-and-vc-glossary/#attribution">attribution</span> and broader implications remain uncertain. Source: <a href="https://metr.org/assets/images/ai-rd-acceleration-assessment/aeci-main.png" data-type="link" data-id="https://metr.org/assets/images/ai-rd-acceleration-assessment/aeci-main.png" target="_blank" rel="noopener">METR</a>.</em></figcaption></figure>



<p class="wp-block-paragraph">You can read this as evidence of a company taking evaluation seriously and granting meaningful access. You can also question how independent scrutiny works when the company being assessed controls what evidence becomes public.</p>



<p class="wp-block-paragraph">Both readings deserve consideration. The arrangement illustrates the difficulty of relying on companies to produce powerful technology, assess its risks, compete commercially and help define the rules governing its deployment.</p>



<p class="wp-block-paragraph">It also makes me cautious about treating concern from inside a lab as either definitive proof of danger or mere marketing. What matters is the specific claim, the evidence supporting it and how much outsiders can verify.</p>



<h2 id="the-founder-and-investor-perspective" class="wp-block-heading"><strong>The founder and investor perspective</strong></h2>



<p class="wp-block-paragraph">For founders, this debate can feel very far removed from the daily work of finding customers, improving a product and keeping the company alive. But its consequences are practical: what customers will buy, what platforms will permit, what insurers will cover and what rules governments will introduce.</p>



<p class="wp-block-paragraph">From an investment perspective, I think it helps to separate three questions:</p>



<ul class="wp-block-list">
<li><strong>What can the technology actually do?</strong> Demonstrated performance matters more than either optimistic or alarming predictions.</li>



<li><strong>What can go wrong in this particular application?</strong> The consequences of a poor recommendation differ from those of an autonomous financial transaction.</li>



<li><strong>Who has a reason to pay for solving that problem?</strong> A broadly important issue does not automatically translate into a venture-scale business.</li>
</ul>



<p class="wp-block-paragraph">There are also trade-offs. More restrictions can reduce certain risks while increasing costs, limiting useful applications or making it harder for smaller companies to compete. Faster deployment can bring benefits sooner while exposing users to failures that testing missed.</p>



<p class="wp-block-paragraph">I don’t think acknowledging these trade-offs makes someone anti-safety or anti-progress. It makes the discussion more useful.</p>



<h2 id="an-israeli-angle-with-an-important-distinction" class="wp-block-heading"><strong>An Israeli angle, with an important distinction</strong></h2>



<p class="wp-block-paragraph">For Israeli founders, the clearest commercial connection is AI security. Noma Security raised a $100 million <span class="mywpglossary-term-def" data-title="series-b" data-url="https://www.vccafe.com/startup-and-vc-glossary/#series-b">Series B</span> in July 2025, while Prompt Security was acquired by SentinelOne in a transaction completed that September. These companies illustrate demand for protecting enterprises as they adopt AI.</p>



<p class="wp-block-paragraph">Israel’s <span class="mywpglossary-term-def" data-title="cybersecurity" data-url="https://www.vccafe.com/startup-and-vc-glossary/#cybersecurity">cybersecurity</span> experience gives founders a relevant starting point: understanding adversaries, protecting data, managing access and identifying vulnerabilities. My expectation is that those capabilities will remain valuable as software agents gain more responsibility.</p>



<p class="wp-block-paragraph">But <span class="mywpglossary-term-def" data-title="enterprise" data-url="https://www.vccafe.com/startup-and-vc-glossary/#enterprise">enterprise</span> AI security and the broader AI safety debate are not interchangeable. Preventing an agent from exposing sensitive information is a concrete customer problem. Determining whether future AI systems could escape human control is a different research question. A funding round or <span class="mywpglossary-term-def" data-title="acquisition" data-url="https://www.vccafe.com/startup-and-vc-glossary/#acquisition">acquisition</span> in the first category tells us little about the second.</p>



<p class="wp-block-paragraph">As investors, we should be precise about which problem a company solves, even when a broader narrative makes the market sound bigger.</p>



<p class="wp-block-paragraph">Listening to <em>Life 3.0</em> has left me with more questions than answers. I can see reasons for optimism and reasons for concern. I can also see an expanding ecosystem of organisations with money, reputations and influence tied to particular views of the future.</p>



<p class="wp-block-paragraph"><strong>For now, I’m more interested in understanding what each side can demonstrate, what would change its mind and who bears the consequences if it is wrong.</strong> That seems a useful place to start before choosing a camp.</p>]]></content:encoded>
					
					<wfw:commentRss>https://www.vccafe.com/ai-safety-who-should-we-believe/feed/</wfw:commentRss>
			<slash:comments>0</slash:comments>
		
		
		<post-id xmlns="com-wordpress:feed-additions:1">14170</post-id>	</item>
		<item>
		<title>What Did Your AI Agent Just Do?</title>
		<link>https://www.vccafe.com/what-did-your-ai-agent-just-do/</link>
					<comments>https://www.vccafe.com/what-did-your-ai-agent-just-do/#respond</comments>
		
		<dc:creator><![CDATA[Eze Vidra]]></dc:creator>
		<pubDate>Sun, 08 Nov 2026 14:43:36 +0000</pubDate>
				<category><![CDATA[Uncategorized]]></category>
		<guid isPermaLink="false">https://www.vccafe.com/?p=14150</guid>

					<description><![CDATA[AI agents can send our emails, spend our money and act in our name. But can we reconstruct what they did, who authorised it and where things went wrong? From the Hugging Face breach to a possible agentic bank run, the challenge is keeping control as we delegate more.]]></description>
										<content:encoded><![CDATA[<p class="wp-block-paragraph">The first time an <span class="mywpglossary-term-def" data-title="ai-agent" data-url="https://www.vccafe.com/startup-and-vc-glossary/#ai-agent">AI agent</span> does something useful on your behalf, it feels like magic. Perhaps you’ve watched Muse/Instinct/Grok bot send an email, book a restaurant or buy something you needed. You ask, it acts, and something disappears from your to-do list. We really are living in extraordinary times.</p>



<p class="wp-block-paragraph">I find that promise incredibly exciting. Between work, family and the endless admin of everyday life, there is plenty I would happily delegate. But the moment an agent starts acting in my name, another thought creeps in: what exactly have I given it permission to do?</p>



<p class="wp-block-paragraph">Could it send an email I would never have approved? Share something confidential? Buy the wrong thing, or agree to terms I haven’t read? And would I find out before the consequences arrived?</p>



<p class="wp-block-paragraph">There is a big difference between asking AI to draft an email and allowing it to send one. Give that same agent access to your inbox, calls, browser and credit card, and you are delegating decisions with consequences beyond the chat window.</p>



<p class="wp-block-paragraph"><strong>We are getting used to the magic of agents acting on our behalf. We should get equally interested in knowing what they are actually doing.</strong></p>



<h2 id="when-passing-a-test-becomes-hacking-another-company" class="wp-block-heading">When passing a test becomes hacking another company</h2>



<p class="wp-block-paragraph">In July 2026, agents running inside an OpenAI <span class="mywpglossary-term-def" data-title="cybersecurity" data-url="https://www.vccafe.com/startup-and-vc-glossary/#cybersecurity">cybersecurity</span> evaluation compromised Hugging Face’s systems. <a href="https://openai.com/index/hugging-face-incident-and-the-road-ahead/" target="_blank" rel="noopener">OpenAI’s subsequent account</a> describes models circumventing isolation controls, communicating through unauthorised channels and accessing third-party infrastructure.</p>



<p class="wp-block-paragraph">The evaluation involved reduced safeguards and was primarily driven by an internal research model. That context matters: this was not an ordinary consumer assistant completing a routine request. Nevertheless, the result was a real intrusion into another company’s infrastructure. <a href="https://huggingface.co/blog/agent-intrusion-technical-timeline" target="_blank" rel="noopener">Hugging Face’s technical account</a> describes thousands of automated decisions, including lateral movement, credential theft and attempts to maintain access.</p>



<p class="wp-block-paragraph">The scale became clearer in an <a href="https://www.redwoodresearch.org/research/hugging-face-incident" target="_blank" rel="noopener">independent investigation by METR and Redwood Research</a>. Researchers estimated that roughly 1,200 agents used an unauthorised message board, exchanging more than 70,000 messages and files. Approximately 700 participated in the attack. They were pursuing ways to manipulate an evaluation score and believed Hugging Face might contain useful clues.</p>



<p class="wp-block-paragraph">For founders, the lesson is uncomfortable. A measurable objective can become a dangerous substitute for the outcome you actually intended. Passing a test is not permission to compromise the organisation that might hold its answers. Equally, improving conversion, reducing support costs or recovering unpaid invoices cannot mean pursuing those goals by any available means.</p>



<p class="wp-block-paragraph">An agent can make progress towards its assigned target while taking actions its operator never intended to authorise.</p>



<h2 id="your-inbox-becomes-an-attack-surface" class="wp-block-heading">Your inbox becomes an attack surface</h2>



<p class="wp-block-paragraph">There is another route to unwanted behaviour: someone else influences your agent through material it reads.</p>



<p class="wp-block-paragraph"><a href="https://www.anthropic.com/research/prompt-injection-defenses" target="_blank" rel="noopener">Anthropic’s research on prompt injection</a> explains how malicious instructions embedded in websites or emails can redirect an agent. Its illustrative example involves an email containing hidden instructions to forward confidential messages elsewhere. This describes a potential attack, rather than a claim that every email assistant has suffered such a breach.</p>



<p class="wp-block-paragraph">We want agents to read outside information, but we do not want that information to acquire the authority of our own instructions. An email from a supplier should be evidence to evaluate. It should not be able to rewrite the rules governing access to the rest of your inbox.</p>



<p class="wp-block-paragraph">Consider a hypothetical assistant asked to organise a business trip. Comparing flights is one level of authority. Booking a non-refundable ticket is another. Calling a hotel, disclosing passport details and accepting cancellation terms introduce further decisions. A request to “sort out the trip” leaves considerable room for interpretation, especially when the agent encounters new information along the way.</p>



<p class="wp-block-paragraph">This is where a reassuring activity summary can be misleading. “Trip booked” tells you the outcome. It does not necessarily tell you which information was disclosed or which commitments were made in your name.</p>



<p class="wp-block-paragraph"><strong>Access to an account is not the same as permission to take every action that account allows.</strong></p>



<h2 id="could-agents-trigger-a-bank-run" class="wp-block-heading">Could agents trigger a bank run?</h2>



<p class="wp-block-paragraph">The concern becomes larger when many agents act simultaneously.</p>



<p class="wp-block-paragraph">In <a href="https://www.apollo.com/wealth/insights-news/insights/daily-spark/is-an-agentic-bank-run-coming" target="_blank" rel="noopener">“Is an Agentic Bank Run Coming?”</a>, Apollo chief economist Torsten Slok considers what happens if household agents automatically move cash towards better-paying accounts. His concern is that banks could lose a substantial source of inexpensive deposit funding.</p>



<p class="wp-block-paragraph">Apollo is describing a potential scenario, not documenting a bank run caused by agents. The mechanism also differs from depositors fleeing an institution they believe is failing: it begins with automated optimisation of returns.</p>



<p class="wp-block-paragraph">My <span class="mywpglossary-term-def" data-title="inference" data-url="https://www.vccafe.com/startup-and-vc-glossary/#inference">inference</span> is that this raises a broader coordination problem. Millions of agents could respond to similar incentives and signals at similar times. Each might be doing something sensible for its owner, while the combined movement creates pressure elsewhere in the system. No malicious instruction is necessary. Better alignment with an individual user does not, by itself, solve the collective problem.</p>



<p class="wp-block-paragraph">We should therefore distinguish three risks: an agent exceeds its mandate; an attacker redirects it; or many agents follow their mandates and collectively produce an undesirable result. Each requires different controls. Treating everything as a <span class="mywpglossary-term-def" data-title="hallucination" data-url="https://www.vccafe.com/startup-and-vc-glossary/#hallucination">hallucination</span> misses what makes agents consequential.</p>



<h2 id="we-cannot-monitor-what-we-do-not-record" class="wp-block-heading">We cannot monitor what we do not record</h2>



<p class="wp-block-paragraph">One of the most revealing examples in <a href="https://partnershiponai.org/wp-content/uploads/2026/10/PAI_obervability-gap-in-ai-agents.pdf" target="_blank" rel="noopener">Partnership on AI’s report, <em>The Observability Gap in AI Agents</em></a>, involves a mortgage application. A borrower exceeds the usual debt-to-income threshold but has substantial cash reserves. A loan officer permits an exception for that application. The agent stores the exception as a general rule and applies the higher threshold to future borrowers.</p>



<figure class="wp-block-image size-full"><img data-recalc-dims="1" loading="lazy" decoding="async" width="994" height="868" data-attachment-id="14154" data-permalink="https://www.vccafe.com/what-did-your-ai-agent-just-do/screenshot-2026-11-08-at-14-28-59/" data-orig-file="https://i0.wp.com/www.vccafe.com/wp-content/uploads/2026/11/Screenshot-2026-11-08-at-14.28.59.png?fit=994%2C868&amp;ssl=1" data-orig-size="994,868" data-comments-opened="1" data-image-title="Screenshot 2026-11-08 at 14.28.59" data-image-description="&lt;p&gt;Screenshot 2026-11-08 at 14.28.59. Related keywords: Israeli tech, Tel Aviv startups, ????????? ???????, ????????? ??????&lt;/p&gt;
" data-image-caption="&lt;p&gt;What agent frameworks record, and what they miss. Partnership on AI’s documentation assessment of four widely used frameworks. This compares telemetry coverage, not overall security effectiveness. (Source)&lt;/p&gt;
" data-large-file="https://i0.wp.com/www.vccafe.com/wp-content/uploads/2026/11/Screenshot-2026-11-08-at-14.28.59.png?fit=994%2C868&amp;ssl=1" src="https://i0.wp.com/www.vccafe.com/wp-content/uploads/2026/11/Screenshot-2026-11-08-at-14.28.59.png?resize=994%2C868&#038;ssl=1" alt="Screenshot 2026-11-08 at 14.28.59 - Israeli tech / ????????? ???????" class="wp-image-14154"/><figcaption class="wp-element-caption">What agent frameworks record, and what they miss. Partnership on AI’s documentation assessment of four widely used frameworks. This compares telemetry coverage, not overall security effectiveness. (<a href="https://partnershiponai.org/wp-content/uploads/2026/10/PAI_obervability-gap-in-ai-agents.pdf?vgo_ee=Pf03K7iwjk7a8X/S9h77uIPrPqtyVMgLcg%3D%3D:XjYzc5ax4PAdKT/sAEHLqeckh9BOBWGf" data-type="link" data-id="https://partnershiponai.org/wp-content/uploads/2026/10/PAI_obervability-gap-in-ai-agents.pdf?vgo_ee=Pf03K7iwjk7a8X/S9h77uIPrPqtyVMgLcg%3D%3D:XjYzc5ax4PAdKT/sAEHLqeckh9BOBWGf" target="_blank" rel="noopener">Source</a>)</figcaption></figure>



<p class="wp-block-paragraph">This is a hypothetical scenario used by the researchers, not a reported banking incident. But it captures a failure that is easy to imagine: a sensible human instruction becomes an unintended policy change. No hacker is required. If the system does not record what entered the agent’s memory, when it changed and what caused the change, investigators may struggle to establish why later decisions went wrong.</p>



<p class="wp-block-paragraph">The researchers reviewed the technical documentation of four widely used agent frameworks: OpenAI Agents SDK, Claude Agent SDK, LangGraph and CrewAI. Their assessment identified gaps in the records needed to connect actions to an agent’s <span class="mywpglossary-term-def" data-title="enterprise" data-url="https://www.vccafe.com/startup-and-vc-glossary/#enterprise">enterprise</span> identity, trace memory changes and reconstruct human oversight. It examined what the frameworks record, rather than testing the effectiveness of complete security systems.</p>



<p class="wp-block-paragraph">The human oversight finding is particularly relevant. None of the four frameworks captured all the intervention information the researchers considered necessary, including who reviewed a decision, their authority and whether review was required by policy. An approval button can put a person into the workflow without leaving enough evidence to assess whether that oversight was meaningful.</p>



<p class="wp-block-paragraph">The report also distinguishes between recording actions and understanding decisions. Many missing operational records could be addressed through better engineering and common standards. Establishing whether a model’s explanation faithfully reflects how it reached a decision remains a research problem. Collecting more reasoning text does not automatically solve it.</p>



<p class="wp-block-paragraph">The authors develop the broader argument in <a href="https://www.technologyreview.com/2026/10/29/1147824/monitor-ai-agents-actions-real-time/" target="_blank" rel="noopener">MIT Technology Review</a>: meaningful monitoring depends on having the underlying records in the first place.</p>



<p class="wp-block-paragraph">My takeaway is that companies need to ask for more than a demonstration of successful task completion. Ask the vendor to reconstruct a failure: what did the agent know at the time, which rules applied, and who authorised the consequential decision? Those answers will reveal far more about whether a business can responsibly delegate work than a reassuring “task completed” message.</p>



<h2 id="the-investment-case-for-keeping-agents-under-control" class="wp-block-heading">The investment case for keeping agents under control</h2>



<p class="wp-block-paragraph">What caught my attention in <a href="https://a16z.com/the-new-security-agenda/" target="_blank" rel="noopener">a16z’s recent survey of 100 enterprise CISOs</a> was that 63% said security was already slowing at least some AI adoption. We tend to discuss agent security as something companies will need once they deploy enough agents. These responses suggest it is already influencing whether those deployments happen at all. In the same survey, 45% planned to buy or upgrade internal agent security within twelve months.</p>



<p class="wp-block-paragraph">There is a commercial incentive here: giving a company confidence to let an agent do more work can help unlock the adoption that agent developers are counting on.</p>



<p class="wp-block-paragraph">The difficulty is that familiar permissions do not fully express what we mean when we delegate a task. An employee might have access to years of customer correspondence, financial documents and internal discussions. An agent acting through that employee’s account could inherit access far beyond what it needs to organise a meeting or follow up on an invoice.</p>



<p class="wp-block-paragraph"><a href="https://www.bvp.com/atlas/securing-ai-agents-the-defining-cybersecurity-challenge-of-2026" target="_blank" rel="noopener">Bessemer’s analysis</a> identifies the accumulation of credentials and entitlements as a central problem. My reading is that companies will increasingly need to define authority around the task being performed, rather than simply accepting everything the underlying account permits. This is a fast evolving market and there are over <a href="https://startups.prompt.security/?" data-type="link" data-id="https://startups.prompt.security/?" target="_blank" rel="noopener">470 tracked AI security startups</a> across the stack.</p>



<figure class="wp-block-image aligncenter size-full"><img data-recalc-dims="1" loading="lazy" decoding="async" width="1290" height="859" data-attachment-id="14153" data-permalink="https://www.vccafe.com/what-did-your-ai-agent-just-do/agentic-ai-security-market-map/" data-orig-file="https://i0.wp.com/www.vccafe.com/wp-content/uploads/2026/11/Agentic-AI-security-market-map-scaled.png?fit=1600%2C1066&amp;ssl=1" data-orig-size="1600,1066" data-comments-opened="1" data-image-title="Agentic AI security market map" data-image-description="&lt;p&gt;Agentic AI security market map. Related keywords: Israeli tech, Tel Aviv startups, ????????? ???????, ????????? ??????&lt;/p&gt;
" data-image-caption="&lt;p&gt;Agentic AI security market map for VC Cafe&lt;/p&gt;
" data-large-file="https://i0.wp.com/www.vccafe.com/wp-content/uploads/2026/11/Agentic-AI-security-market-map-scaled.png?fit=1024%2C682&amp;ssl=1" src="https://i0.wp.com/www.vccafe.com/wp-content/uploads/2026/11/Agentic-AI-security-market-map-scaled.png?resize=1290%2C859&#038;ssl=1" alt="Agentic AI security market map - Israeli tech / ????????? ???????" class="wp-image-14153" srcset="https://i0.wp.com/www.vccafe.com/wp-content/uploads/2026/11/Agentic-AI-security-market-map-scaled.png?w=1600&amp;ssl=1 1600w, https://i0.wp.com/www.vccafe.com/wp-content/uploads/2026/11/Agentic-AI-security-market-map-scaled.png?resize=300%2C200&amp;ssl=1 300w, https://i0.wp.com/www.vccafe.com/wp-content/uploads/2026/11/Agentic-AI-security-market-map-scaled.png?resize=1024%2C682&amp;ssl=1 1024w, https://i0.wp.com/www.vccafe.com/wp-content/uploads/2026/11/Agentic-AI-security-market-map-scaled.png?resize=768%2C512&amp;ssl=1 768w, https://i0.wp.com/www.vccafe.com/wp-content/uploads/2026/11/Agentic-AI-security-market-map-scaled.png?resize=1536%2C1023&amp;ssl=1 1536w, https://i0.wp.com/www.vccafe.com/wp-content/uploads/2026/11/Agentic-AI-security-market-map-scaled.png?resize=2048%2C1364&amp;ssl=1 2048w" sizes="auto, (max-width: 1290px) 100vw, 1290px" /><figcaption class="wp-element-caption"><span class="mywpglossary-term-def" data-title="agentic-ai" data-url="https://www.vccafe.com/startup-and-vc-glossary/#agentic-ai">Agentic AI</span> security market map for VC Cafe (<a href="https://csa-web-prod-uploads-817137372772.s3.us-west-2.amazonaws.com/r55hlikpspei1smzuzebrq8ynos2?response-content-disposition=inline%3B%20filename%3D%22Agentic%20AI%20Security%20Innovator%20Market%20Map%20%253F%20October%202026.pdf%22%3B%20filename%2A%3DUTF-8%27%27Agentic%2520AI%2520Security%2520Innovator%2520Market%2520Map%2520%25E2%2580%2594%2520October%25202026.pdf&amp;response-content-type=application%2Fpdf&amp;X-Amz-Algorithm=AWS4-HMAC-SHA256&amp;X-Amz-Credential=ASIA34QJOZZSCKVD4GQX%2F20261108%2Fus-west-2%2Fs3%2Faws4_request&amp;X-Amz-Date=20261108T143031Z&amp;X-Amz-Expires=300&amp;X-Amz-Security-Token=IQoJb3JpZ2luX2VjEEQaCXVzLXdlc3QtMiJGMEQCIFcWPySrCd%2BiPvfnzqyH3E%2FNRK2BeHZ3SGTdTea2Jsy7AiAu5kFWW%2Fxk5b%2FGBTawC9r%2FHJBjk9AwzbnSxr4d1ICfxyrvAwgNEAAaDDgxNzEzNzM3Mjc3MiIMUN0teAJPMzK5PqHGKswDHHnGFKtpCljZ8Y72J2dLYciYKj3t6NNsRsvWgPckrxTMLMNkl%2FVrvWsektmeIQEQYpbvThgVxpd6Q07UB07OJWQXrN5sP36bYNvE8AiqRVEuFo%2Bbj%2FMCZbu%2FGJq%2BIj1X1LidztZe1%2Bnd%2FPjWYu0MzPHAh%2BcrmzCPEZT4HnqAQe%2FSj%2BWV3AZoT%2BUqf0%2F2iLz0IsqYC1618u6UTMqCJbYPN%2FvavU%2BT6akUIM1aSF6zjmSgxS4yF%2FfxJxSMYnBq4ZXyHb33yFqTfi4W91bpQQqNHqluiRFY7irPEdwEXlQIFIAlWFhdhCzlSfKc11693rI%2F0LUCwz0GbvFRXTnIfgv%2Bizi%2BQgm6Qw%2FZKh61%2BLJW4%2FrDI%2Bj2BHGz0laXPuZoKv7RBWihxd8JU7KepPKF%2B1wEbZEMe%2BDR7%2Fh%2FRwwyZhx1CP3wB3q4eO%2F6Fk%2FWBF1Z3A36dTQccGOMMTKvtF1UcFUE%2FxslQF6A1lomTs6I8jA%2FJWnp2f677qGixc4f3FxR9cAysZsMVNIJX30GY%2BGkbXheFegoSTBkMibTiIcDRYzx1GUrEMF%2Byz%2FCRUMc7xwg9jEGvZQ2JiPlbjlGtXkjkiY0wXmd%2BKMPBHdKPSTnrjDBysHXBjqmAe0gyowlGchQRADw3pT8zVYNUBkzB1Ek7vgpNmUTgU5CEEA%2ByhClhhGgl%2FxDkkwLq%2FZNNTA3TrhzhhHasi25jYON8L2UsGS5HoxixrQaH%2BARLZvFSa2HXUgA%2FBVUyaYfrf1klXbgIfHm3U6fGtveTyPk5a2lVyUHAq%2BFQXqJxDHYdXGNZFQBheFvhuLtZk5rO8w8Na46v7Vt7Fer2DFHMbEF%2BH83iMA%3D&amp;X-Amz-SignedHeaders=host&amp;X-Amz-Signature=ee3c34fdbbadfc3116d8b6f8a48d9b9d37a525c050ce74af8933a6f57fb7d114" data-type="link" data-id="https://csa-web-prod-uploads-817137372772.s3.us-west-2.amazonaws.com/r55hlikpspei1smzuzebrq8ynos2?response-content-disposition=inline%3B%20filename%3D%22Agentic%20AI%20Security%20Innovator%20Market%20Map%20%253F%20October%202026.pdf%22%3B%20filename%2A%3DUTF-8%27%27Agentic%2520AI%2520Security%2520Innovator%2520Market%2520Map%2520%25E2%2580%2594%2520October%25202026.pdf&amp;response-content-type=application%2Fpdf&amp;X-Amz-Algorithm=AWS4-HMAC-SHA256&amp;X-Amz-Credential=ASIA34QJOZZSCKVD4GQX%2F20261108%2Fus-west-2%2Fs3%2Faws4_request&amp;X-Amz-Date=20261108T143031Z&amp;X-Amz-Expires=300&amp;X-Amz-Security-Token=IQoJb3JpZ2luX2VjEEQaCXVzLXdlc3QtMiJGMEQCIFcWPySrCd%2BiPvfnzqyH3E%2FNRK2BeHZ3SGTdTea2Jsy7AiAu5kFWW%2Fxk5b%2FGBTawC9r%2FHJBjk9AwzbnSxr4d1ICfxyrvAwgNEAAaDDgxNzEzNzM3Mjc3MiIMUN0teAJPMzK5PqHGKswDHHnGFKtpCljZ8Y72J2dLYciYKj3t6NNsRsvWgPckrxTMLMNkl%2FVrvWsektmeIQEQYpbvThgVxpd6Q07UB07OJWQXrN5sP36bYNvE8AiqRVEuFo%2Bbj%2FMCZbu%2FGJq%2BIj1X1LidztZe1%2Bnd%2FPjWYu0MzPHAh%2BcrmzCPEZT4HnqAQe%2FSj%2BWV3AZoT%2BUqf0%2F2iLz0IsqYC1618u6UTMqCJbYPN%2FvavU%2BT6akUIM1aSF6zjmSgxS4yF%2FfxJxSMYnBq4ZXyHb33yFqTfi4W91bpQQqNHqluiRFY7irPEdwEXlQIFIAlWFhdhCzlSfKc11693rI%2F0LUCwz0GbvFRXTnIfgv%2Bizi%2BQgm6Qw%2FZKh61%2BLJW4%2FrDI%2Bj2BHGz0laXPuZoKv7RBWihxd8JU7KepPKF%2B1wEbZEMe%2BDR7%2Fh%2FRwwyZhx1CP3wB3q4eO%2F6Fk%2FWBF1Z3A36dTQccGOMMTKvtF1UcFUE%2FxslQF6A1lomTs6I8jA%2FJWnp2f677qGixc4f3FxR9cAysZsMVNIJX30GY%2BGkbXheFegoSTBkMibTiIcDRYzx1GUrEMF%2Byz%2FCRUMc7xwg9jEGvZQ2JiPlbjlGtXkjkiY0wXmd%2BKMPBHdKPSTnrjDBysHXBjqmAe0gyowlGchQRADw3pT8zVYNUBkzB1Ek7vgpNmUTgU5CEEA%2ByhClhhGgl%2FxDkkwLq%2FZNNTA3TrhzhhHasi25jYON8L2UsGS5HoxixrQaH%2BARLZvFSa2HXUgA%2FBVUyaYfrf1klXbgIfHm3U6fGtveTyPk5a2lVyUHAq%2BFQXqJxDHYdXGNZFQBheFvhuLtZk5rO8w8Na46v7Vt7Fer2DFHMbEF%2BH83iMA%3D&amp;X-Amz-SignedHeaders=host&amp;X-Amz-Signature=ee3c34fdbbadfc3116d8b6f8a48d9b9d37a525c050ce74af8933a6f57fb7d114" target="_blank" rel="noopener">source</a>)</figcaption></figure>



<p class="wp-block-paragraph">That helps explain the relevance of <a href="https://sentra.io/news/sentra-raises-150-million-agentic-data-security" target="_blank" rel="noopener">Sentra’s $150 million investment, led by Capital One Ventures in November</a>. Sentra’s approach starts with the information an agent can reach: finding sensitive data, understanding access and reducing unnecessary exposure before anything goes wrong.</p>



<p class="wp-block-paragraph">The participation of a bank’s venture arm is particularly interesting in this context, although it does not itself establish how Capital One uses the technology. For financial institutions, an agent’s ability to retrieve information can be useful and dangerous for precisely the same reason: it can connect material across systems faster than a person would.</p>



<p class="wp-block-paragraph">Yet restricting access only answers part of the question. An agent might legitimately need customer records and an email account to complete its task. That still leaves a consequential decision about which information belongs in the message it sends.</p>



<p class="wp-block-paragraph"><a href="https://www.onyx.security/press-release/onyx-security-raises-113m-series-b-to-control-advanced-ai-quadrupling-revenue-since-stealth-launch-four-months-ago" target="_blank" rel="noopener">Onyx Security, which raised $113 million in a Bessemer-led round in July</a>, is pursuing that problem through monitoring and intervention while agents operate. Its proposition is that organisations need a way to assess an action in context, even when the agent has permission to use the tools involved.</p>



<p class="wp-block-paragraph">These two Israeli companies show how the opportunity extends beyond detecting an obviously malicious prompt. Sentra is working to limit the exposure created by access; Onyx is working to control how agents exercise their capabilities. Funding announcements are not evidence that the problem has been solved, but they point to where customers and investors see work that needs doing.</p>



<p class="wp-block-paragraph">At Remagine Ventures, the question I would ask a founder in this space is: what can your customer confidently let an agent do after installing your product that they could not allow before? A convincing answer connects security to a business outcome. It might mean allowing an agent to resolve a support case, process a payment or work with sensitive documents under clear constraints.</p>



<h2 id="useful-autonomy-needs-boundaries" class="wp-block-heading">Useful autonomy needs boundaries</h2>



<p class="wp-block-paragraph">A sales agent might prepare a discount but require approval to exceed an agreed threshold. A finance agent might identify an invoice for payment, while a separate control verifies the recipient and enforces the spending limit. A support agent might issue small refunds, with escalation for unusual patterns or larger amounts.</p>



<p class="wp-block-paragraph">The controls must exist outside the agent’s own willingness to obey them. Telling a model to stay within budget is weaker than using a payment mechanism that cannot exceed that budget.</p>



<p class="wp-block-paragraph">Software development offers another practical example. A coding agent might produce a feature that works perfectly in a demo while overlooking the company’s requirements for authentication, handling sensitive data or using approved libraries. Passing the tests does not necessarily mean the code meets the organisation’s standards.</p>



<p class="wp-block-paragraph"><a href="https://pandorian.ai/" target="_blank" rel="noopener">Pandorian</a> addresses this gap by turning engineering, architecture and security standards into checks applied to pull requests and existing code. Teams can define their requirements, specify where they apply and identify violations within the development workflow. This gives the people responsible for security and compliance a way to check whether written policies are reflected in the software being delivered.</p>



<p class="wp-block-paragraph">That is the kind of boundary that could make greater autonomy practical. A company can give coding agents more responsibility when it has a dependable process for checking their output before accepting it. The agent’s assurance that it followed the rules should never be the only evidence.</p>



<p class="wp-block-paragraph">In commerce, <a href="https://cloud.google.com/blog/products/ai-machine-learning/announcing-agents-to-payments-ap2-protocol" target="_blank" rel="noopener">Google’s Agent Payments Protocol</a> tackles a related question through cryptographically signed mandates that provide evidence of a user’s instructions and the transaction they authorised. When software initiates a purchase, the other parties need a way to verify its authority. That does not guarantee a wise purchase, but it creates a more concrete basis for establishing what was permitted.</p>



<p class="wp-block-paragraph">There is a product-design challenge running through all these examples. Asking users to approve every click would undermine the reason they adopted an agent. A better approach is to agree the boundaries upfront, surface material exceptions and provide a clear record afterwards. Users should be able to distinguish “the agent prepared this” from “the agent committed me to this.”</p>



<p class="wp-block-paragraph">I remain optimistic about agents because the value of delegating work is enormous. But the ability to complete a task is only part of the product. The ability to constrain, inspect and interrupt that work will determine how much responsibility people are willing to hand over.</p>



<p class="wp-block-paragraph"><strong>Before giving an agent the keys, we should be able to explain which doors it can open, what it did behind them, and how to take those keys back.</strong></p>]]></content:encoded>
					
					<wfw:commentRss>https://www.vccafe.com/what-did-your-ai-agent-just-do/feed/</wfw:commentRss>
			<slash:comments>0</slash:comments>
		
		
		<post-id xmlns="com-wordpress:feed-additions:1">14150</post-id>	<enclosure length="4279361" type="application/pdf" url="https://partnershiponai.org/wp-content/uploads/2026/10/PAI_obervability-gap-in-ai-agents.pdf"/><itunes:explicit>no</itunes:explicit><itunes:subtitle>AI agents can send our emails, spend our money and act in our name. But can we reconstruct what they did, who authorised it and where things went wrong? From the Hugging Face breach to a possible agentic bank run, the challenge is keeping control as we delegate more.</itunes:subtitle><itunes:summary>AI agents can send our emails, spend our money and act in our name. But can we reconstruct what they did, who authorised it and where things went wrong? From the Hugging Face breach to a possible agentic bank run, the challenge is keeping control as we delegate more.</itunes:summary><itunes:keywords>Uncategorized</itunes:keywords></item>
		<item>
		<title>When AI Builds AI</title>
		<link>https://www.vccafe.com/when-ai-builds-ai/</link>
					<comments>https://www.vccafe.com/when-ai-builds-ai/#respond</comments>
		
		<dc:creator><![CDATA[Eze Vidra]]></dc:creator>
		<pubDate>Sat, 07 Nov 2026 17:44:07 +0000</pubDate>
				<category><![CDATA[Artificial Intelligence]]></category>
		<category><![CDATA[Self improvement]]></category>
		<category><![CDATA[AI agents]]></category>
		<category><![CDATA[AI Infrastructure]]></category>
		<category><![CDATA[AI Safety]]></category>
		<category><![CDATA[artificial intelligence]]></category>
		<category><![CDATA[developer tools]]></category>
		<category><![CDATA[inference]]></category>
		<category><![CDATA[Israeli startups]]></category>
		<category><![CDATA[recursive self-improvement]]></category>
		<category><![CDATA[Remagine Ventures]]></category>
		<category><![CDATA[venture capital]]></category>
		<guid isPermaLink="false">https://www.vccafe.com/?p=14128</guid>

					<description><![CDATA[AI now leads a quarter of Anthropic’s own R&#038;D, and an independent review finds its progress running at nearly twice its earlier pace. What does that feedback loop mean for the speed of AI, for oversight, and for where startups can build?]]></description>
										<content:encoded><![CDATA[<p class="wp-block-paragraph">In February 2026, Claude led less than 1% of Anthropic&#8217;s measured AI research and development work. By August, it led 26%.</p>



<p class="wp-block-paragraph">AI is increasingly writing the code, running the experiments and optimising the infrastructure behind its own development. That creates a powerful feedback loop: better models help <span class="mywpglossary-term-def" data-title="build" data-url="https://www.vccafe.com/startup-and-vc-glossary/#build">build</span> better models, while more efficient infrastructure makes the process cheaper.</p>



<p class="wp-block-paragraph">The implications extend beyond developer productivity. If AI accelerates the work that produces the next generation of AI, what happens to the pace of improvement? Can oversight keep up? And where can startups <span class="mywpglossary-term-def" data-title="build" data-url="https://www.vccafe.com/startup-and-vc-glossary/#build">build</span> valuable businesses?</p>



<h2 id="the-feedback-loop-is-taking-shape" class="wp-block-heading">The feedback loop is taking shape</h2>



<figure class="wp-block-image aligncenter size-large is-resized"><img data-recalc-dims="1" loading="lazy" decoding="async" width="623" height="1024" data-attachment-id="14129" data-permalink="https://www.vccafe.com/when-ai-builds-ai/whatsapp-image-2026-11-06-at-12-02-57/" data-orig-file="https://i0.wp.com/www.vccafe.com/wp-content/uploads/2026/11/WhatsApp-Image-2026-11-06-at-12.02.57.jpeg?fit=974%2C1600&amp;ssl=1" data-orig-size="974,1600" data-comments-opened="1" data-image-title="WhatsApp Image 2026-11-06 at 12.02.57" data-image-description="&lt;p&gt;WhatsApp Image 2026-11-06 at 12.02.57. Related keywords: Israeli tech, Tel Aviv startups, ????????? ???????, ????????? ??????&lt;/p&gt;
" data-image-caption="&lt;p&gt;How much of Anthropic’s AI research and development Claude does, by month. The share Claude “leads” rose from under 1% in February 2026 to 26% in August, and none is fully autonomous. Figures are self-reported by Anthropic. Chart: The Wall Street Journal. Data: Anthropic.&lt;/p&gt;
" data-large-file="https://i0.wp.com/www.vccafe.com/wp-content/uploads/2026/11/WhatsApp-Image-2026-11-06-at-12.02.57.jpeg?fit=623%2C1024&amp;ssl=1" src="https://i0.wp.com/www.vccafe.com/wp-content/uploads/2026/11/WhatsApp-Image-2026-11-06-at-12.02.57.jpeg?resize=623%2C1024&#038;ssl=1" alt="WSJ on Anthropic" class="wp-image-14129" style="aspect-ratio:0.6084036663944097;width:313px;height:auto" srcset="https://i0.wp.com/www.vccafe.com/wp-content/uploads/2026/11/WhatsApp-Image-2026-11-06-at-12.02.57.jpeg?resize=623%2C1024&amp;ssl=1 623w, https://i0.wp.com/www.vccafe.com/wp-content/uploads/2026/11/WhatsApp-Image-2026-11-06-at-12.02.57.jpeg?resize=183%2C300&amp;ssl=1 183w, https://i0.wp.com/www.vccafe.com/wp-content/uploads/2026/11/WhatsApp-Image-2026-11-06-at-12.02.57.jpeg?resize=768%2C1261&amp;ssl=1 768w, https://i0.wp.com/www.vccafe.com/wp-content/uploads/2026/11/WhatsApp-Image-2026-11-06-at-12.02.57.jpeg?resize=935%2C1536&amp;ssl=1 935w, https://i0.wp.com/www.vccafe.com/wp-content/uploads/2026/11/WhatsApp-Image-2026-11-06-at-12.02.57.jpeg?w=974&amp;ssl=1 974w" sizes="auto, (max-width: 623px) 100vw, 623px" /><figcaption class="wp-element-caption">How much of Anthropic’s AI research and development Claude does, by month. The share Claude “leads” rose from under 1% in February 2026 to 26% in August, and none is fully autonomous. Figures are self-reported by Anthropic. Chart: <a href="https://www.wsj.com/tech/ai/ai-anthropic-rsi-recursive-self-improvement-bff4e815" data-type="link" data-id="https://www.wsj.com/tech/ai/ai-anthropic-rsi-recursive-self-improvement-bff4e815" target="_blank" rel="noopener">The Wall Street Journal</a>. Data: Anthropic.</figcaption></figure>



<p class="wp-block-paragraph">Those figures come from Anthropic&#8217;s <a href="https://www.anthropic.com/institute/measuring-pace-of-ai-development" target="_blank" rel="noopener">R&amp;D Automation Index</a>, published in September. &#8220;Leads&#8221; means Claude could complete most of a task from a high-level prompt, with humans supervising. More than 90% of the measured work involved AI at least as a collaborator, and none was classified as fully autonomous. These are self-reported numbers, and Claude itself was used to <span class="mywpglossary-term-def" data-title="build" data-url="https://www.vccafe.com/startup-and-vc-glossary/#build">build</span> the task taxonomy and classify the work, a bias risk Anthropic acknowledges. Even so, the direction is clear: researchers are increasingly using today&#8217;s models to <span class="mywpglossary-term-def" data-title="build" data-url="https://www.vccafe.com/startup-and-vc-glossary/#build">build</span> tomorrow&#8217;s.</p>



<p class="wp-block-paragraph">Together AI offers another concrete example. Its <a href="https://together.ai/blog/together-compiler-continual-compilation-for-ai-inference" target="_blank" rel="noopener">Together Compiler</a>, announced on 4 November, uses agents to optimise the software that runs AI models, including low-level GPU kernels and how computation is distributed across hardware.</p>



<p class="wp-block-paragraph"><span class="mywpglossary-term-def" data-title="inference" data-url="https://www.vccafe.com/startup-and-vc-glossary/#inference">Inference</span> workloads constantly change. New models arrive, prompts get longer and customers adopt different usage patterns. A configuration optimised for yesterday&#8217;s traffic may be inefficient tomorrow. Together&#8217;s agents search for better implementations, testing changes through a persistent engine that avoids repeatedly restarting the system.</p>



<p class="wp-block-paragraph">The company reports up to 2.6 times higher <span class="mywpglossary-term-def" data-title="token" data-url="https://www.vccafe.com/startup-and-vc-glossary/#token">token</span> throughput than the best published open-source configurations on SemiAnalysis&#8217;s InferenceX <span class="mywpglossary-term-def" data-title="benchmark" data-url="https://www.vccafe.com/startup-and-vc-glossary/#benchmark">benchmark</span>, in less than 24 hours. It attributes much of that to running 252 times more experiments than agent systems that restart the engine for every test. These are vendor-reported results for particular workloads, but they illustrate a shift towards software that can be continuously optimised as conditions change.</p>



<p class="wp-block-paragraph">One surprising finding in the <a href="https://www.together.ai/together-compiler-paper" target="_blank" rel="noopener">Together Compiler research paper</a>: giving agents an expert-curated knowledge base reduced their speedups by 25 to 31% after 12 hours, and naive agents told simply to make the system as fast as possible beat agents following expert-prescribed strategies. That does not make expertise redundant. It suggests its value may shift towards defining objectives, designing tests and deciding which changes are <span class="mywpglossary-term-def" data-title="safe" data-url="https://www.vccafe.com/startup-and-vc-glossary/#safe">safe</span> to deploy.</p>



<p class="wp-block-paragraph">Google DeepMind&#8217;s <a href="https://deepmind.google/discover/blog/alphaevolve-a-gemini-powered-coding-agent-for-designing-advanced-algorithms/" target="_blank" rel="noopener">AlphaEvolve</a> shows how far this can reach. When it was unveiled in May 2025, Google reported AI-generated improvements that recovered an average of 0.7% of its worldwide compute resources, reduced Gemini&#8217;s training time by 1%, and contributed a verified circuit modification to an upcoming TPU. Google has since <a href="https://deepmind.google/blog/alphaevolve-impact/" target="_blank" rel="noopener">expanded its use</a> well beyond those first results. AI is helping improve both the software and hardware used to <span class="mywpglossary-term-def" data-title="build" data-url="https://www.vccafe.com/startup-and-vc-glossary/#build">build</span> AI.</p>



<h2 id="from-better-answers-to-lasting-improvement" class="wp-block-heading">From better answers to lasting improvement</h2>



<p class="wp-block-paragraph">Stanford&#8217;s <a href="https://cs329a.stanford.edu/" target="_blank" rel="noopener">CS329A: Self-Improving AI Agents</a> course, whose lectures are now available online, connects several mechanisms behind this shift: spending more compute exploring solutions, checking results with verifiers, learning from execution feedback, and training on successful attempts.</p>



<p class="wp-block-paragraph">These represent different forms of improvement. An agent can try several approaches and produce a better answer without changing the underlying model. Persistent improvement requires retaining something useful, whether in memory, reusable code or updated model weights.</p>



<p class="wp-block-paragraph">Consider a coding agent that writes a solution, runs it, encounters an error and revises its approach. Fixing that error improves the current result. Capturing the successful method and using it to improve future attempts creates a learning loop. The crucial test is whether those gains transfer to unfamiliar problems.</p>



<p class="wp-block-paragraph">For startups, this raises a useful question: does your product get better with experience, or does it start from scratch with a more capable model each time? A system that turns verified customer outcomes into reusable improvements could <span class="mywpglossary-term-def" data-title="build" data-url="https://www.vccafe.com/startup-and-vc-glossary/#build">build</span> an advantage that grows with usage, provided it has permission to use that data.</p>



<h2 id="how-fast-can-this-go" class="wp-block-heading">How fast can this go?</h2>



<p class="wp-block-paragraph">The potential for compounding progress is clear. Cheaper <span class="mywpglossary-term-def" data-title="inference" data-url="https://www.vccafe.com/startup-and-vc-glossary/#inference">inference</span> enables more experiments. Better coding agents help researchers execute those experiments. Successful experiments can produce better models and further efficiency gains.</p>



<p class="wp-block-paragraph">There is now evidence beyond code-generation statistics. In an <a href="https://metr.org/blog/2026-11-05-anthropic-ai-rd-acceleration-assessment/" target="_blank" rel="noopener">independent assessment</a> published on 5 November, METR estimated that Anthropic&#8217;s <span class="mywpglossary-term-def" data-title="benchmark" data-url="https://www.vccafe.com/startup-and-vc-glossary/#benchmark">benchmark</span> scores (measured on AECI, Anthropic&#8217;s version of the Epoch Capabilities Index) have grown at roughly 1.9 times their previous rate since October 2025, with a 95% confidence interval of 1.5 to 2.8. METR&#8217;s best guess is that around 80% of the current acceleration is due to AI-assisted R&amp;D. It stressed substantial uncertainty, and noted that Anthropic redacted a majority of the evidence METR found most informative.</p>



<p class="wp-block-paragraph">That does not establish an unlimited growth curve. Faster coding, faster research and faster capability improvement are different measures. If coding takes half a project&#8217;s time, doubling coding speed reduces total project time by only a quarter, assuming everything else stays unchanged. Testing, experiment runtime and deployment become the next constraints.</p>



<p class="wp-block-paragraph">&#8220;More tokens equals better performance&#8221; also has an economic limit. Additional optimisation is worthwhile while the expected gains exceed the cost of generating, testing and maintaining changes.</p>



<p class="wp-block-paragraph">Even with these constraints, founders should plan for a moving frontier. Products that are uneconomic today could become viable, while advantages built around current model limitations could disappear.</p>



<h2 id="the-risks-begin-before-ai-becomes-autonomous" class="wp-block-heading">The risks begin before AI becomes autonomous</h2>



<p class="wp-block-paragraph">The immediate danger is optimising the wrong thing extremely well. An agent pursuing throughput might improve <span class="mywpglossary-term-def" data-title="benchmark" data-url="https://www.vccafe.com/startup-and-vc-glossary/#benchmark">benchmark</span> performance while introducing failures on unusual inputs. Passing a test suite establishes that the software passed those tests, not that it is universally correct.</p>



<p class="wp-block-paragraph">This makes the evaluator as important as the agent. Objectives need constraints on security, quality and reliability, supported by independent tests that the optimiser cannot simply tailor itself to. Longer tasks also need end-to-end evaluation: an agent can perform individual steps well and still fail to complete the overall job reliably.</p>



<p class="wp-block-paragraph">There is a risk of shared blind spots. If similar models write the code, generate the tests and review the results, agreement can create false confidence. Asking the same model to check its work in another conversation is a weak substitute for independent verification.</p>



<p class="wp-block-paragraph">Human supervision can become less meaningful as the volume of changes grows. Someone may retain the authority to stop a system while losing the ability to understand when intervention is necessary. Effective oversight requires reproducible experiments, bounded permissions, inspectable evidence and tested rollback.</p>



<p class="wp-block-paragraph">Competitive pressure compounds the problem. Companies benefit from accelerating research, even when slowing down to evaluate a new capability would be prudent. The concern is that development could outpace our ability to assess its consequences, and the warning lines are no longer hypothetical. Anthropic&#8217;s own <a href="https://www.anthropic.com/responsible-scaling-policy" target="_blank" rel="noopener">Responsible Scaling Policy</a> treats a doubling in the rate of progress, substantially attributable to AI-driven R&amp;D, as a key risk threshold, and METR&#8217;s estimate puts that line within its confidence interval. Reproducible experiments, bounded permissions, inspectable evidence and tested rollback need to be standard practice, and independently checked, before fully autonomous research becomes commonplace, not after.</p>



<h2 id="where-startups-can-build" class="wp-block-heading">Where startups can build</h2>



<p class="wp-block-paragraph">As an early-stage investor, I see opportunities in three layers.</p>



<p class="wp-block-paragraph">The first is optimisation itself. Together&#8217;s results suggest that databases, networks, cloud workloads and simulations can be continuously improved against measurable customer outcomes. Specialist engineering agents can do the same for bounded problems, where domain expertise and reliable evaluation make improvements demonstrable.</p>



<p class="wp-block-paragraph">The second is the trust layer around that optimisation. Someone needs to prove that agent-generated changes preserve correctness, security and reliability. Someone also needs to provide the experiment infrastructure underneath: realistic workload testing, reproducible environments, cost controls and <span class="mywpglossary-term-def" data-title="safe" data-url="https://www.vccafe.com/startup-and-vc-glossary/#safe">safe</span> deployment. If the evaluator matters as much as the agent, this layer could prove as valuable as the agents themselves.</p>



<p class="wp-block-paragraph">The third is new consumer experiences. Cheaper <span class="mywpglossary-term-def" data-title="inference" data-url="https://www.vccafe.com/startup-and-vc-glossary/#inference">inference</span> makes personalised learning, creative tools and AI-native games economically viable in ways they were not a year ago.</p>



<p class="wp-block-paragraph">For Israeli founders, systems engineering and security are particularly interesting entry points. A small team can <span class="mywpglossary-term-def" data-title="build" data-url="https://www.vccafe.com/startup-and-vc-glossary/#build">build</span> valuable infrastructure around evaluating, controlling and improving AI systems without training a frontier model. Keewano, a Remagine <span class="mywpglossary-term-def" data-title="portfolio" data-url="https://www.vccafe.com/startup-and-vc-glossary/#portfolio">portfolio</span> company where I sit on the board, is one example: it is building a database designed for AI agents&#8217; reasoning and <span class="mywpglossary-term-def" data-title="analytics" data-url="https://www.vccafe.com/startup-and-vc-glossary/#analytics">analytics</span>, infrastructure that becomes more relevant as more of the work is done by agents.</p>



<p class="wp-block-paragraph">The defensibility question becomes sharper, however. Why will customers keep using your product as the underlying models improve? Representative workload data, trusted integrations and accumulated evidence of reliable performance are stronger answers than a clever prompt. The ability to learn from real outcomes could strengthen that advantage further.</p>



<p class="wp-block-paragraph">At Remagine Ventures, I would ask founders three questions: what can your system improve, how do you prove it, and who decides when it can act?</p>



<p class="wp-block-paragraph">AI helping <span class="mywpglossary-term-def" data-title="build" data-url="https://www.vccafe.com/startup-and-vc-glossary/#build">build</span> its successor could compress development cycles and unlock entirely new products. It could also compress the time available to discover mistakes. Building the mechanisms to verify and control that progress may be one of the biggest opportunities it creates.</p>]]></content:encoded>
					
					<wfw:commentRss>https://www.vccafe.com/when-ai-builds-ai/feed/</wfw:commentRss>
			<slash:comments>0</slash:comments>
		
		
		<post-id xmlns="com-wordpress:feed-additions:1">14128</post-id>	<enclosure length="4638700" type="application/pdf" url="https://www.together.ai/together-compiler-paper"/><itunes:explicit>no</itunes:explicit><itunes:subtitle>AI now leads a quarter of Anthropic’s own R&amp;#038;D, and an independent review finds its progress running at nearly twice its earlier pace. What does that feedback loop mean for the speed of AI, for oversight, and for where startups can build?</itunes:subtitle><itunes:summary>AI now leads a quarter of Anthropic’s own R&amp;#038;D, and an independent review finds its progress running at nearly twice its earlier pace. What does that feedback loop mean for the speed of AI, for oversight, and for where startups can build?</itunes:summary><itunes:keywords>Artificial Intelligence, Self improvement, AI agents, AI Infrastructure, AI Safety, artificial intelligence, developer tools, inference, Israeli startups, recursive self-improvement, Remagine Ventures, venture capital</itunes:keywords></item>
		<item>
		<title>Weekly Firgun Newsletter – November 6 2026</title>
		<link>https://www.vccafe.com/weekly-firgun-newsletter-november-6-2026/</link>
					<comments>https://www.vccafe.com/weekly-firgun-newsletter-november-6-2026/#respond</comments>
		
		<dc:creator><![CDATA[Eze Vidra]]></dc:creator>
		<pubDate>Fri, 06 Nov 2026 12:14:39 +0000</pubDate>
				<category><![CDATA[Firgun]]></category>
		<category><![CDATA[42AI]]></category>
		<category><![CDATA[AI infra]]></category>
		<category><![CDATA[AI Infrastructure]]></category>
		<category><![CDATA[ai sovereignty]]></category>
		<category><![CDATA[Brava]]></category>
		<category><![CDATA[Cybersecurity]]></category>
		<category><![CDATA[Digital Health]]></category>
		<category><![CDATA[enterprise AI]]></category>
		<category><![CDATA[Israeli startups]]></category>
		<category><![CDATA[Pat Grady]]></category>
		<category><![CDATA[sentra]]></category>
		<category><![CDATA[Sequoia Capital]]></category>
		<category><![CDATA[shlomo kremer]]></category>
		<category><![CDATA[tailormed]]></category>
		<category><![CDATA[US midterms]]></category>
		<category><![CDATA[venture capital]]></category>
		<guid isPermaLink="false">https://www.vccafe.com/?p=14111</guid>

					<description><![CDATA[The mid term US elections had an impact on the markets, Israeli teams raised $150 million to secure enterprise data and $15 million to test cyber defences. Plus TailorMed’s $250 million deal, the tech verdict from the US midterms and Sequoia’s retreat from London.]]></description>
										<content:encoded><![CDATA[<p class="wp-block-paragraph">It&#8217;s #Firgun time!</p>



<p class="wp-block-paragraph">The original version of the newsletter was published on LinkedIn to approximately <a href="https://www.linkedin.com/pulse/weekly-firgun-newsletter-eze-vidra-059tf/" data-type="link" data-id="https://www.linkedin.com/pulse/weekly-firgun-newsletter-eze-vidra-059tf/" target="_blank" rel="noopener">8,700 subscribers</a>.</p>



<p class="wp-block-paragraph">Get the next issue <a href="https://www.vccafe.com/firgun/#get-firgun-in-your-inbox">by email</a> or <a href="https://www.linkedin.com/newsletters/firgun-6954740769651871744/" target="_blank" rel="noopener">subscribe to Firgun on LinkedIn</a>.</p>



<p class="wp-block-paragraph">Firgun (pronounced &#8216;feer-goon&#8217; or for halloween &#8216;feargoon&#8217;) is an Israeli term describing genuine, unselfish joy in someone else&#8217;s success&#8230; It&#8217;s basically the opposite of &#8220;schadenfreude&#8221;. Every week since March 2020, we celebrate the wins of our startup community. Read more about <a href="https://lnkd.in/dg2a86p" target="_blank" rel="noopener"><strong>why Firgun matters</strong></a>. If this was forwarded to you and you&#8217;d like to subscribe, previous editions are on <a href="https://www.vccafe.com/"><strong>VC Cafe</strong></a>.</p>



<p class="wp-block-paragraph"><a href="https://apnews.com/article/5cad93330300b2d45a9665d5eccb2ab2" target="_blank" rel="noopener">President Trump said this week that the US would end its annual military aid to Israel by the end of the year, arguing that Israel no longer needs it</a>. The figure under the current agreement is $3.8 billion a year, much of it spent on American-made defence equipment. The agreement runs through 2028, so how an earlier cutoff would work is still unclear. It got me thinking about what sovereignty means for Israeli tech. If a US provider changed its terms tomorrow, which Israeli hospitals, businesses or public services could keep running? I’d like the next government to take stock of where we depend on overseas cloud providers and AI models, and make sure Israeli founders and researchers can get the computing power they need at home.</p>



<p class="wp-block-paragraph">As for who might lead that government, parties representing 67 of the Knesset’s 120 members <a href="https://www.reuters.com/world/middle-east/netanyahus-position-weakens-ultra-orthodox-party-withholds-support-2026-11-05/" target="_blank" rel="noopener">recommended Gadi Eisenkot</a> for the first chance to form a coalition, compared with 46 for Benjamin Netanyahu. That puts Eisenkot in the strongest position on paper. Getting those parties to agree on a government is a different matter. President Isaac Herzog plans to announce his choice on Sunday, November 8, after further talks with party leaders.</p>



<p class="wp-block-paragraph">In the wider tech world, the US midterms gave tech investors a reason to pay attention to governors’ races. <a href="https://www.techmeme.com/261104/h1405" target="_blank" rel="noopener">Texas elected a Democratic governor</a> after a campaign in which data centres became a live issue, and tech stocks fell to their worst day in 19 months, as investors reassessed which projects could go ahead. On the frontier-lab side, <a href="https://www.linkedin.com/company/anthropicresearch/" target="_blank" rel="noopener">Anthropic</a>&#8216;s updated filing says it may soon reach its own threshold for <a href="https://www.theneuron.ai/digest/everything-that-happened-in-ai-today-tuesday-november-3-2026/" target="_blank" rel="noopener">dramatically accelerated AI-assisted research</a>, while <a href="https://www.linkedin.com/company/openai/" target="_blank" rel="noopener">OpenAI</a> reportedly <a href="https://www.theneuron.ai/digest/everything-that-happened-in-ai-today-monday-november-2-2026/" target="_blank" rel="noopener">trimmed its raise to about $20 billion at a $1.25 trillion valuation</a>. The same digest has <a href="https://www.linkedin.com/company/amazon-web-services/" target="_blank" rel="noopener">Amazon Web Services (AWS)</a> growing 39.3% to nearly $46 billion and <a href="https://www.linkedin.com/company/meta/" target="_blank" rel="noopener">Meta</a>’s ad revenue up 26%, plus <a href="https://www.linkedin.com/company/apple/" target="_blank" rel="noopener">Apple</a>&#8216;s first Apple Intelligence features for China, some powered by <a href="https://www.linkedin.com/company/chinese-alibaba-group/" target="_blank" rel="noopener">Alibaba Group</a>’s Qwen. <a href="https://www.linkedin.com/company/meituan/" target="_blank" rel="noopener">Meituan</a> also released <a href="https://www.theneuron.ai/digest/everything-that-happened-in-ai-today-tuesday-november-3-2026/" target="_blank" rel="noopener">LongCat-2.5</a>, a 1.6-trillion-parameter coding model. On the product front, <a href="https://www.linkedin.com/company/perplexity-ai/" target="_blank" rel="noopener">Perplexity</a> launched Teammates, always-on agents that can run several jobs at once, and <a href="https://www.linkedin.com/company/google/" target="_blank" rel="noopener">Google</a> rolled out <a href="https://blog.google/innovation-and-ai/technology/developers-tools/introducing-gemini-omni-1-2-flash/" target="_blank" rel="noopener">Gemini Omni 1.2 Flash,</a> extending generated video clips from 10 to 15 seconds while cutting the cost by 30%. <a href="https://www.linkedin.com/company/binance/" target="_blank" rel="noopener">Binance</a> explained how a <a href="https://www.techmeme.com/261103/h1840" target="_blank" rel="noopener">fake job interview</a> gave attackers a way in and how they exploited an internal AI tool in the $6.3 billion theft that the FBI attributed to North Korea. The money kept moving: <a href="https://techmeme.com/river" target="_blank" rel="noopener">Thinking Machines Lab</a> is reportedly raising $5 billion at a $40 billion pre-money <span class="mywpglossary-term-def" data-title="valuation" data-url="https://www.vccafe.com/startup-and-vc-glossary/#valuation">valuation</span>, and <a href="https://www.linkedin.com/company/features-and-labels/" target="_blank" rel="noopener">fal</a> raised a reported $1.3 billion. <a href="https://www.linkedin.com/company/shield-ai/" target="_blank" rel="noopener">Shield</a> signed a deal worth up to $5.3 billion for radar maker Echodyne, and <a href="https://www.linkedin.com/company/salesforce/" target="_blank" rel="noopener">Salesforce</a> agreed to buy defense-logistics startup <a href="https://www.linkedin.com/company/runetech/" target="_blank" rel="noopener">Rune Technologies</a> for a reported <a href="https://dealroom.co/news/168565-salesforce-to-buy-defense-startup-rune-for-a-reported-375m/" target="_blank" rel="noopener">$375 million</a>. And closer to home, <a href="https://www.linkedin.com/company/sequoia/" target="_blank" rel="noopener">Sequoia Capital</a> announced it is closing its London office, winding down its scout programme and Arc <span class="mywpglossary-term-def" data-title="accelerator" data-url="https://www.vccafe.com/startup-and-vc-glossary/#accelerator">accelerator</span>, and combining its venture and growth teams after raising roughly $10 billion in new funds. Many European VCs saw the opportunity to advocate the case for European tech.</p>



<h2 id="quote-of-the-week" class="wp-block-heading">QUOTE OF THE WEEK</h2>



<blockquote class="wp-block-quote is-layout-flow wp-block-quote-is-layout-flow">
<p class="wp-block-paragraph">You must never confuse faith that you will prevail in the end — which you can never afford to lose — with the discipline to confront the most brutal facts of your current reality, whatever they might be. </p>
<cite>— Admiral James Stockdale</cite></blockquote>



<h2 id="new-funding-rounds" class="wp-block-heading">NEW FUNDING ROUNDS</h2>



<ul class="wp-block-list">
<li>Kol hakavod to <a href="https://www.linkedin.com/in/yoav-regev-31718a1/" target="_blank" rel="noopener">Yoav Regev</a>, <a href="https://www.linkedin.com/in/asaf-kochan/" target="_blank" rel="noopener">Asaf Kochan</a>, <a href="https://www.linkedin.com/in/ronreiter/" target="_blank" rel="noopener">Ron Reiter</a> and team <a href="https://www.linkedin.com/company/sentra-io/" target="_blank" rel="noopener">Sentra</a> on raising a $150 million <span class="mywpglossary-term-def" data-title="series-c" data-url="https://www.vccafe.com/startup-and-vc-glossary/#series-c">Series C</span> to help companies see and control what sensitive information can AI agents risk.</li>



<li>Well done <a href="https://www.linkedin.com/in/amit-gal-or-05a2379b/" target="_blank" rel="noopener">Amit Gal-Or</a> and team <a href="https://www.linkedin.com/company/42foundationalai/" target="_blank" rel="noopener">42AI</a> on coming out of stealth with a $62.5M seed round (self funded) to <span class="mywpglossary-term-def" data-title="build" data-url="https://www.vccafe.com/startup-and-vc-glossary/#build">build</span> a “behavioral world model” that can predict human behaviour!</li>



<li>Mabruk to <a href="https://www.linkedin.com/in/idan-basre/" target="_blank" rel="noopener">Idan Basre</a>, <a href="https://www.linkedin.com/in/itay-gabbay/" target="_blank" rel="noopener">Itay Gabbay</a> and team <a href="https://www.linkedin.com/company/bravasecurity/" target="_blank" rel="noopener">Brava Security</a> on coming out of stealth with a $15 million seed round to recreate the traces an attack would leave in a company’s systems, so the security team can find out whether its alarms would actually go off!</li>
</ul>



<h2 id="exits" class="wp-block-heading">EXITS</h2>



<ul class="wp-block-list">
<li>Mazal tov <a href="https://www.linkedin.com/in/srulik/" target="_blank" rel="noopener">Israel (Srulik) Dvorsky</a>, <a href="https://www.linkedin.com/in/adam-siton-81152a1b/" target="_blank" rel="noopener">Adam Siton</a> and team <a href="https://www.linkedin.com/company/tailormed---medical-journey-innovations/" target="_blank" rel="noopener">TailorMed</a> on the <span class="mywpglossary-term-def" data-title="acquisition" data-url="https://www.vccafe.com/startup-and-vc-glossary/#acquisition">acquisition</span> by <a href="https://www.linkedin.com/company/growthcurvecapital/" target="_blank" rel="noopener">GrowthCurve Capital</a> for a reported $250M to connect patients and carers with financial and medical support programs across the United States!</li>



<li>Good stuff <a href="https://www.linkedin.com/in/taljacobson/" target="_blank" rel="noopener">Tal Jacobson</a> and team <a href="https://www.linkedin.com/company/perion/" target="_blank" rel="noopener">Perion</a> on the agreement to be acquired by India’s <a href="https://www.linkedin.com/company/affle/" target="_blank" rel="noopener">Affle</a> for $420 million in cash. While it’s a bittersweet <span class="mywpglossary-term-def" data-title="milestone" data-url="https://www.vccafe.com/startup-and-vc-glossary/#milestone">milestone</span>, an exit is nevertheless an outcome.</li>



<li>Kudos <a href="https://www.linkedin.com/in/matan-saat/" target="_blank" rel="noopener">Matan Saat</a> and team <a href="https://www.linkedin.com/company/azimut-ai/" target="_blank" rel="noopener">Azimut.ai</a> on the <span class="mywpglossary-term-def" data-title="acquisition" data-url="https://www.vccafe.com/startup-and-vc-glossary/#acquisition">acquisition</span> by <a href="https://www.linkedin.com/company/windward-ltd-/" target="_blank" rel="noopener">Windward</a> to identify &#8216;dark fleet&#8217; vessels using using existing daylight and infrared cameras! The terms of the deal were not disclosed.</li>
</ul>



<h2 id="appointments" class="wp-block-heading">APPOINTMENTS</h2>



<ul class="wp-block-list">
<li>Congrats <a href="https://www.linkedin.com/in/navotvolk/" target="_blank" rel="noopener">Navot Volk</a> for getting promoted to General Partner at <a href="https://www.linkedin.com/company/ground-up-ventures/" target="_blank" rel="noopener">Ground Up Ventures</a>!</li>
</ul>



<h2 id="links-for-your-browser" class="wp-block-heading">LINKS FOR YOUR BROWSER</h2>



<h3 id="vc-cafe-remagine-ventures-israel-vc" class="wp-block-heading">VC CAFE/ REMAGINE VENTURES/ ISRAEL VC</h3>



<ul class="wp-block-list">
<li><a href="https://www.vccafe.com/israels-ai-infrastructure-stack/">Israel&#8217;s AI Infrastructure stack</a>: AI sovereignty depends on controlling data, models, compute and operations, so Israel should both <span class="mywpglossary-term-def" data-title="build" data-url="https://www.vccafe.com/startup-and-vc-glossary/#build">build</span> global AI infrastructure companies and secure affordable compute for its own researchers and startups.</li>



<li><a href="https://www.vccafe.com/sequoias-london-exit-what-europe-can-learn-from-israel/">Sequoia&#8217;s London exit: what Europe can learn from Israel</a> &#8211; Sequoia says it doesn&#8217;t see the need for a local presence in London, and Sequoia&#8217;s history in Israel shows investor relationships can outlast office closures.</li>



<li><a href="https://www.vccafe.com/requests-for-startups-winter-2027-edition/">Requests for Startups: Winter 2027 edition</a> &#8211; A review of RFS lists from YC, Pear, a16z, Menlo and QED, arguing the real opportunity is making AI products dependable for customers rather than chasing crowded headline categories.</li>



<li><a href="https://www.vccafe.com/big-ideas-basic-questions/">Big ideas, basic questions</a> &#8211; Even ambitious deep-tech startups should answer basic questions early: team expertise, how customers buy, whether economics hold beyond the demo, and what each round must prove.</li>
</ul>



<h3 id="israel" class="wp-block-heading">ISRAEL</h3>



<ul class="wp-block-list">
<li><a href="https://www.calcalistech.com/ctechnews/article/d1lau0oxq" target="_blank" rel="noopener">Israeli startups raised at least $872 million in October, led by cybersecurity and AI</a> &#8211; a productive month continuing the trend of concentration of bigger tickets into fewer deals.</li>



<li><a href="https://www.linkedin.com/company/palo-alto-networks/" target="_blank" rel="noopener">Palo Alto Networks</a> is <a href="https://calcalistech.com/ctechnews/article/qx69w4bra" target="_blank" rel="noopener">negotiating to buy Clutch</a>, which secures non-human identities and AI-agent credentials and <a href="https://calcalistech.com/ctechnews/article/p8pegpx4h" target="_blank" rel="noopener">also Beacon,</a> a two-year-old startup building AI agents for threat detection and response</li>



<li><a href="https://www.jpost.com/defense-and-tech/article-913930" target="_blank" rel="noopener">Israeli defense-tech investment set to nearly triple to $1 billion in 2026</a> &#8211; The major advantage of Israeli defense-tech developments is that most of them are combat-proven. These are not theoretical products</li>
</ul>



<h3 id="global" class="wp-block-heading">GLOBAL</h3>



<ul class="wp-block-list">
<li><a href="https://x.com/sequoia/status/2117961312143855764">Sequoia&#8217;s original announcement of how it&#8217;s changing</a> &#8211; Sunsetting the <span class="mywpglossary-term-def" data-title="accelerator" data-url="https://www.vccafe.com/startup-and-vc-glossary/#accelerator">accelerator</span> and scout programmes, closing the London office, merging early and growth into one team, flattening the partnership. As Pat Grady mentioned in the video below: bigger is not better, you need &#8220;fewer better things&#8221;. The power law means 1% of companies produce 99% of returns, and an outlier founder needs an outlier partner.</li>



<li><a href="https://menlovc.com/perspective/2026-the-state-of-enterprise-ai/?" target="_blank" rel="noopener">The state of enterprise AI</a> by <a href="https://www.linkedin.com/company/menlo-ventures/" target="_blank" rel="noopener">Menlo Ventures</a> &#8211; <span class="mywpglossary-term-def" data-title="enterprise" data-url="https://www.vccafe.com/startup-and-vc-glossary/#enterprise">Enterprise</span> AI deployment spend scaled to <strong>$177B</strong> by September 2026, up from <strong>$33B</strong> last year. Growth is accelerating: <strong>5.4x</strong> vs. <strong>3.9x</strong> a year ago. Every dollar that went into developing models this year is generating <strong>$1.70</strong> of revenue</li>



<li><a href="https://www.ncsc.gov.uk/collection/annual-review-2026" target="_blank" rel="noopener">The UK national cyber security center published its annual report</a> &#8211; in a nutshell, threats are on the rise &#8211; by <a href="https://www.linkedin.com/company/national-cyber-security-centre/" target="_blank" rel="noopener">National Cyber Security Centre</a></li>
</ul>



<h2 id="media-of-the-week" class="wp-block-heading">MEDIA OF THE WEEK</h2>



<p class="wp-block-paragraph"><strong>Israel’s AI Infrastructure Stack.</strong> This week’s map follows the layers underneath an AI application, from compute and networking to databases and the software that keeps the application running. (<a href="https://www.vccafe.com/israels-ai-infrastructure-stack/">source</a>)</p>



<figure class="wp-block-image"><img decoding="async" src="https://media.licdn.com/dms/image/v2/D4D12AQH4Z4x8N2fFqw/article-inline_image-shrink_1000_1488/B4DaGwBj5jGQAI-/0/1793954562445?e=1795651200&amp;v=beta&amp;t=zzmmeC3F3BEqDx3GhS9IDVw3cweMCe5gpe6CCYIM7E4" alt="Article content"/></figure>



<p class="wp-block-paragraph">Israeli defence sector continues to attract funding and M&amp;A &#8211; check out this recent summary by <a href="https://www.linkedin.com/company/ivc-data-and-insights/" target="_blank" rel="noopener">IVC Data and Insights</a> (<a href="https://www.ivc-online.com/LinkClick.aspx?fileticket=A3GA4WJSKN0%3d&amp;portalid=0&amp;timestamp=1792645152760" target="_blank" rel="noopener">source</a>)</p>



<figure class="wp-block-image"><img decoding="async" src="https://media.licdn.com/dms/image/v2/D4D12AQFu7puN-iL25g/article-inline_image-shrink_1500_2232/B4DaGwDLcHK0AQ-/0/1793954986001?e=1795651200&amp;v=beta&amp;t=7b8l6GnjrkWgEcBQJG8P9gDgQn13CihSNiY64MBJMLo" alt="Article content"/></figure>



<p class="wp-block-paragraph">Serial <span class="mywpglossary-term-def" data-title="unicorn" data-url="https://www.vccafe.com/startup-and-vc-glossary/#unicorn">unicorn</span> founders &#8211; list of founders that started two or more unicorns, including Israel&#8217;s OGs <a href="https://www.linkedin.com/in/amnon-shashua/" target="_blank" rel="noopener">Amnon Shashua</a> and <a href="https://www.linkedin.com/in/shlomokramer/" target="_blank" rel="noopener">Shlomo Kramer</a> by <a href="https://www.linkedin.com/company/dealroom-co/" target="_blank" rel="noopener">Dealroom.co</a> (<a href="https://dealroom.co/resources/serial-unicorn-founders" target="_blank" rel="noopener">source</a>)</p>



<figure class="wp-block-image"><img decoding="async" src="https://media.licdn.com/dms/image/v2/D4D12AQFzCupN0Iq-5Q/article-inline_image-shrink_1500_2232/B4DaGwCjYwIMAQ-/0/1793954822618?e=1795651200&amp;v=beta&amp;t=jGnQz859Q668b-3b6zTLqRteursQLL3V2F9O62OOJZg" alt="Article content"/></figure>



<p class="wp-block-paragraph">Mapping the simulation economy &#8211; landscape by <a href="https://www.linkedin.com/company/sapphirevc/" target="_blank" rel="noopener">Sapphire Ventures</a> (<a href="https://sapphireventures.com/blog/synthetic-research-ai-simulation/?u" target="_blank" rel="noopener">source</a>)</p>



<figure class="wp-block-image"><img decoding="async" src="https://media.licdn.com/dms/image/v2/D4D12AQFIREXc1PUpJA/article-inline_image-shrink_1500_2232/B4DaGwCzqXJsAQ-/0/1793954888572?e=1795651200&amp;v=beta&amp;t=UP1YYYS0BpzV7lqToIriqWmE0shpQsw1r2ffcKMhS5E" alt="Article content"/></figure>



<p class="wp-block-paragraph">400LLMs ranked <a href="https://x.com/dummerspast39/status/2117369403859669359?a=o556QWbTyWZbVOLXqm4&amp;s=46">(source)</a></p>



<figure class="wp-block-image"><img decoding="async" src="https://media.licdn.com/dms/image/v2/D4D12AQFe8mNhHPHCVw/article-inline_image-shrink_1500_2232/B4DaGwC6mXK4AQ-/0/1793954917007?e=1795651200&amp;v=beta&amp;t=AJ63OgKrMeLHCa8wevPLL7nFuGTD7ng9rU6qQ1zhGzo" alt="Article content"/></figure>



<p class="wp-block-paragraph"><a href="https://www.linkedin.com/company/sequoia/" target="_blank" rel="noopener">Sequoia Capital</a> co managing partner <a href="https://www.linkedin.com/in/gradypb/" target="_blank" rel="noopener">Pat Grady</a> on the recent changes made by the fund: getting smaller, closing London office and shutting down their Arc and scout programs to focus on big tickets&#8230;</p>



<figure class="wp-block-embed is-type-video is-provider-youtube wp-block-embed-youtube wp-embed-aspect-16-9 wp-has-aspect-ratio"><div class="wp-block-embed__wrapper">
https://www.youtube.com/watch?v=Gmp1_uPoIS0
</div></figure>



<p class="wp-block-paragraph">That’s all for this week. If you know someone looking for their first customer, here’s a particularly good form of firgun: think of one person who feels the problem they’re solving, and offer to make the introduction. A Friday email that takes you two minutes might give a founder the most useful conversation they have next week.</p>



<p class="wp-block-paragraph">There&#8217;s a lot to celebrate in our &#8220;small&#8221; community. Keep on creating!</p>



<p class="wp-block-paragraph">Eze Vidra</p>]]></content:encoded>
					
					<wfw:commentRss>https://www.vccafe.com/weekly-firgun-newsletter-november-6-2026/feed/</wfw:commentRss>
			<slash:comments>0</slash:comments>
		
		
		<post-id xmlns="com-wordpress:feed-additions:1">14111</post-id>	<enclosure length="1952616" type="application/pdf" url="https://www.ivc-online.com/LinkClick.aspx?fileticket=A3GA4WJSKN0%3d&amp;amp;portalid=0&amp;amp;timestamp=1792645152760"/><itunes:explicit>no</itunes:explicit><itunes:subtitle>The mid term US elections had an impact on the markets, Israeli teams raised $150 million to secure enterprise data and $15 million to test cyber defences. Plus TailorMed’s $250 million deal, the tech verdict from the US midterms and Sequoia’s retreat from London.</itunes:subtitle><itunes:summary>The mid term US elections had an impact on the markets, Israeli teams raised $150 million to secure enterprise data and $15 million to test cyber defences. Plus TailorMed’s $250 million deal, the tech verdict from the US midterms and Sequoia’s retreat from London.</itunes:summary><itunes:keywords>Firgun, 42AI, AI infra, AI Infrastructure, ai sovereignty, Brava, Cybersecurity, Digital Health, enterprise AI, Israeli startups, Pat Grady, sentra, Sequoia Capital, shlomo kremer, tailormed, US midterms, venture capital</itunes:keywords></item>
		<item>
		<title>Israel’s AI Infrastructure stack</title>
		<link>https://www.vccafe.com/israels-ai-infrastructure-stack/</link>
					<comments>https://www.vccafe.com/israels-ai-infrastructure-stack/#respond</comments>
		
		<dc:creator><![CDATA[Eze Vidra]]></dc:creator>
		<pubDate>Thu, 05 Nov 2026 10:56:38 +0000</pubDate>
				<category><![CDATA[AI Infrastructure]]></category>
		<category><![CDATA[Artificial Intelligence]]></category>
		<category><![CDATA[Israeli startups]]></category>
		<category><![CDATA[Agentic AI]]></category>
		<category><![CDATA[ai sovereignty]]></category>
		<category><![CDATA[compute]]></category>
		<category><![CDATA[databases]]></category>
		<category><![CDATA[inference]]></category>
		<category><![CDATA[networking]]></category>
		<category><![CDATA[open-weight models]]></category>
		<category><![CDATA[Remagine Ventures]]></category>
		<category><![CDATA[storage]]></category>
		<category><![CDATA[venture capital]]></category>
		<guid isPermaLink="false">https://www.vccafe.com/?p=14102</guid>

					<description><![CDATA[What would happen if your business suddenly lost access to OpenAI and Anthropic? AI sovereignty exposes a deeper infrastructure question, spanning compute, networking, storage, databases and agent operations. Israel’s startups are building across this stack, creating an opportunity that extends well beyond frontier models.]]></description>
										<content:encoded><![CDATA[<p class="wp-block-paragraph">Your customer data might still be there. Your application could still load. Your team would still own its code. But the intelligence powering customer support, document analysis, coding tools and autonomous workflows could become unavailable. How much of the business would continue operating, and how quickly could you restore the rest?</p>



<p class="wp-block-paragraph">This is a hypothetical stress test, not a prediction. But it gets to the heart of a question that deserves more attention as AI becomes embedded in the economy: how much control do we have over the technology we increasingly depend on?</p>



<p class="wp-block-paragraph">Mistral co-founder and CEO Arthur Mensch recently warned that Europe risks falling into a “complete trap” if it concentrates on industrial applications while leaving frontier AI models to the US and China. His argument, reported by POLITICO, is that Europe needs its own advanced models and computing capacity to preserve its security and freedom to act.</p>



<p class="wp-block-paragraph">Mensch has a commercial interest in that argument. Mistral benefits when governments and enterprises prioritise European AI. Nevertheless, the underlying question matters well beyond Europe, and it is particularly relevant to Israel.</p>



<p class="wp-block-paragraph"><strong>Building successful AI applications and controlling the infrastructure they depend on are different achievements. Israel needs a strategy for both.</strong></p>



<p class="wp-block-paragraph">That question brings us to the infrastructure beneath AI: the compute that runs models, the networks that connect it, the storage and databases that supply information, and the software that makes the system reliable. Israeli founders are building across these layers. Understanding where they fit helps explain both Israel’s commercial opportunity and the capabilities it needs to secure for itself.</p>



<h2 id="what-happens-if-the-model-disappears" class="wp-block-heading">What happens if the model disappears?</h2>



<p class="wp-block-paragraph">The impact of losing access depends on how a product is built. A business relying on one hosted model without a tested alternative could lose critical functionality immediately. Another might continue operating with a different provider, reduced capabilities or more human intervention.</p>



<p class="wp-block-paragraph">Switching models is possible, but it is rarely as simple as changing an API key. Models differ in their responses, tool use, <span class="mywpglossary-term-def" data-title="latency" data-url="https://www.vccafe.com/startup-and-vc-glossary/#latency">latency</span>, context handling and reliability. Prompts that work well with one may perform poorly with another. A replacement needs to be tested against the actual tasks customers depend on.</p>



<p class="wp-block-paragraph">Using the same model through a second cloud provider may provide another delivery route. It does not necessarily remove dependence on the model owner or protect against every kind of access restriction. Operational resilience and strategic independence are related, but they are not identical.</p>



<p class="wp-block-paragraph">Open-weight models provide another option. When the weights are available and the licence permits the intended use, organisations can run a model on infrastructure they control. OpenAI’s own gpt-oss models illustrate this distinction: they can be deployed locally or through hosting providers, independently of access to OpenAI’s hosted API. That does not make them equivalent replacements for every proprietary model or every task.</p>



<p class="wp-block-paragraph">And downloading weights is only the beginning. Someone still needs to provide the compute, memory, storage, networking, deployment software and operational expertise.</p>



<p class="wp-block-paragraph"><strong>A fallback model without the capacity and systems to run it is a contingency plan on paper.</strong></p>



<h2 id="ai-sovereignty-starts-beneath-the-model" class="wp-block-heading">AI sovereignty starts beneath the model</h2>



<p class="wp-block-paragraph">I think the most useful definition of AI sovereignty is the ability to make meaningful choices about the systems you depend on: where they run, who can access the data, which models they use, and whether essential operations can continue when a supplier changes its terms or becomes unavailable.</p>



<p class="wp-block-paragraph">That does not require every country to manufacture its own chips or every startup to train a frontier model. It requires understanding dependencies and deciding which ones are acceptable, which need alternatives, and which capabilities are too important to leave entirely outside your control.</p>



<p class="wp-block-paragraph">There are several layers to this:</p>



<ul class="wp-block-list">
<li><strong>Data control:</strong> Can you access, export and govern your own information?</li>



<li><strong>Model control:</strong> Can you select, adapt or replace the intelligence powering your product?</li>



<li><strong>Compute access:</strong> Can you obtain the capacity required to run it?</li>



<li><strong>Operational control:</strong> Can your team deploy, monitor, secure and recover the system?</li>
</ul>



<p class="wp-block-paragraph">These questions explain why AI infrastructure extends far beyond GPUs. The model is one component in a much larger system.</p>



<h2 id="what-ai-infrastructure-actually-includes" class="wp-block-heading">What AI infrastructure actually includes</h2>



<p class="wp-block-paragraph">For founders and investors, it helps to break the stack into the problems each layer solves.</p>



<figure class="wp-block-table"><table class="has-fixed-layout"><tbody><tr><th>Layer</th><th>What it provides</th><th>Why it matters for AI</th></tr><tr><td><strong>Power and cooling</strong></td><td>Electricity, heat removal and facility resilience</td><td>Computing capacity is useful only if it can be powered and operated reliably.</td></tr><tr><td><strong>Compute</strong></td><td>CPUs, GPUs and specialised accelerators</td><td>Performs the calculations required for training and <span class="mywpglossary-term-def" data-title="inference" data-url="https://www.vccafe.com/startup-and-vc-glossary/#inference">inference</span>.</td></tr><tr><td><strong>Networking</strong></td><td>Connections between chips, servers and clusters</td><td>Distributed workloads need to move and exchange information efficiently.</td></tr><tr><td><strong>Storage and memory</strong></td><td>Persistent data storage and fast working memory</td><td>Models and processors need timely access to weights, datasets and intermediate results.</td></tr><tr><td><strong>Databases and retrieval</strong></td><td>Structured queries, search and access to relevant information</td><td>Connects AI systems to business facts and context.</td></tr><tr><td><strong>Data engineering</strong></td><td>Ingestion, transformation, quality and lineage</td><td>Makes information from fragmented systems usable and maintainable.</td></tr><tr><td><strong><span class="mywpglossary-term-def" data-title="orchestration" data-url="https://www.vccafe.com/startup-and-vc-glossary/#orchestration">Orchestration</span> and <span class="mywpglossary-term-def" data-title="inference" data-url="https://www.vccafe.com/startup-and-vc-glossary/#inference">inference</span></strong></td><td>Resource allocation, deployment and model serving</td><td>Turns available hardware and models into dependable services.</td></tr><tr><td><strong>Agent infrastructure</strong></td><td>Tool access, workflow execution, state and permissions</td><td>Enables systems to carry out multistep tasks across business applications.</td></tr><tr><td><strong>Evaluation, observability and security</strong></td><td>Testing, monitoring, diagnosis and protection</td><td>Helps teams understand whether a system works, why it failed and what it is allowed to do.</td></tr></tbody></table></figure>



<p class="wp-block-paragraph">These categories overlap. A company may operate across storage, databases and data processing, or combine infrastructure <span class="mywpglossary-term-def" data-title="orchestration" data-url="https://www.vccafe.com/startup-and-vc-glossary/#orchestration">orchestration</span> with model deployment. The purpose of the taxonomy is to make the dependencies visible.</p>



<p class="wp-block-paragraph">It also helps distinguish three different activities: training a model, serving a model, and operating an agent. They share infrastructure, but their requirements and economics are not identical.</p>



<h2 id="databases-deserve-more-attention" class="wp-block-heading">Databases deserve more attention</h2>



<p class="wp-block-paragraph">Databases are a particularly interesting part of this story because access to a capable model does not automatically give an agent an accurate understanding of a business.</p>



<p class="wp-block-paragraph">Consider an agent investigating why a customer stopped using a product. It may need the customer’s current account status, historical activity, support conversations, billing events and the sequence in which those things happened. Finding a relevant document is useful, but it may not answer the entire question.</p>



<p class="wp-block-paragraph">Different infrastructure serves different needs. A transactional database maintains business state. An analytical system supports queries over large datasets. Vector retrieval helps find information by similarity. Event history preserves sequences of activity. Agent state records progress through a task.</p>



<p class="wp-block-paragraph">Pinecone, for example, provides <span class="mywpglossary-term-def" data-title="vector-database" data-url="https://www.vccafe.com/startup-and-vc-glossary/#vector-database">vector database</span> infrastructure for AI retrieval. Keewano is building database infrastructure designed for machine-generated workloads and access to detailed event data. These represent different approaches to making information available to AI systems (Disclosure: Remagine Ventures is an investor in Keewano, Troup AI and Pandorian, all mentioned in this article).</p>



<p class="wp-block-paragraph">The investment question is not whether every company needs a new database because it uses AI. It is whether a new workload creates requirements that existing systems struggle to meet at an acceptable cost.</p>



<p class="wp-block-paragraph">As agents become more active consumers of business data, that question becomes increasingly important. The infrastructure must support the queries machines actually generate, while maintaining permissions, freshness and predictable performance.</p>



<h2 id="mapping-israels-ai-infrastructure-opportunity" class="wp-block-heading">Mapping Israel’s AI infrastructure opportunity</h2>



<p class="wp-block-paragraph">Israel already has companies operating across many of these layers.</p>



<p class="wp-block-paragraph">Grove Ventures, working with Startup Nation Central, mapped more than 70 companies across six data-centre infrastructure categories: energy, cooling, compute, storage and memory, connectivity, and management and <span class="mywpglossary-term-def" data-title="orchestration" data-url="https://www.vccafe.com/startup-and-vc-glossary/#orchestration">orchestration</span>. According to the mapping, 65% were founded since 2020. Those figures describe Grove’s data-centre ecosystem, rather than the entire AI software infrastructure market.</p>



<p class="wp-block-paragraph">The landscape below extends that perspective into databases, data engineering, <span class="mywpglossary-term-def" data-title="inference" data-url="https://www.vccafe.com/startup-and-vc-glossary/#inference">inference</span> and AI operations.</p>



<figure class="wp-block-image size-large"><img data-recalc-dims="1" loading="lazy" decoding="async" width="990" height="1024" data-attachment-id="14103" data-permalink="https://www.vccafe.com/israels-ai-infrastructure-stack/israel-ai-infra-landscape-vc-cafe/" data-orig-file="https://i0.wp.com/www.vccafe.com/wp-content/uploads/2026/11/israel-ai-infra-landscape-VC-Cafe-scaled.png?fit=1548%2C1600&amp;ssl=1" data-orig-size="1548,1600" data-comments-opened="1" data-image-title="israel-ai-infra-landscape VC Cafe" data-image-description="&lt;p&gt;israel-ai-infra-landscape VC Cafe. Related keywords: pre-seed funding, fundraising strategy, ???? ??? ???, ???? ??? ??????????&lt;/p&gt;
" data-image-caption="&lt;p&gt;israel-ai-infra-landscape VC Cafe for VC Cafe&lt;/p&gt;
" data-large-file="https://i0.wp.com/www.vccafe.com/wp-content/uploads/2026/11/israel-ai-infra-landscape-VC-Cafe-scaled.png?fit=990%2C1024&amp;ssl=1" src="https://i0.wp.com/www.vccafe.com/wp-content/uploads/2026/11/israel-ai-infra-landscape-VC-Cafe.png?resize=990%2C1024&#038;ssl=1" alt="israel-ai-infra-landscape VC Cafe - pre-seed funding / ???? ??? ???" class="wp-image-14103" srcset="https://i0.wp.com/www.vccafe.com/wp-content/uploads/2026/11/israel-ai-infra-landscape-VC-Cafe-scaled.png?resize=990%2C1024&amp;ssl=1 990w, https://i0.wp.com/www.vccafe.com/wp-content/uploads/2026/11/israel-ai-infra-landscape-VC-Cafe-scaled.png?resize=290%2C300&amp;ssl=1 290w, https://i0.wp.com/www.vccafe.com/wp-content/uploads/2026/11/israel-ai-infra-landscape-VC-Cafe-scaled.png?resize=768%2C794&amp;ssl=1 768w, https://i0.wp.com/www.vccafe.com/wp-content/uploads/2026/11/israel-ai-infra-landscape-VC-Cafe-scaled.png?resize=1486%2C1536&amp;ssl=1 1486w, https://i0.wp.com/www.vccafe.com/wp-content/uploads/2026/11/israel-ai-infra-landscape-VC-Cafe-scaled.png?w=1981&amp;ssl=1 1981w, https://i0.wp.com/www.vccafe.com/wp-content/uploads/2026/11/israel-ai-infra-landscape-VC-Cafe-scaled.png?w=1548&amp;ssl=1 1548w" sizes="auto, (max-width: 990px) 100vw, 990px" /><figcaption class="wp-element-caption">Israel’s AI infrastructure landscape: selected Israeli-founded companies spanning power and silicon, networking, storage, databases, models and agent infrastructure. <span class="mywpglossary-term-def" data-title="acquisition" data-url="https://www.vccafe.com/startup-and-vc-glossary/#acquisition">Acquisition</span> labels include completed and announced transactions; highlighted companies are Remagine Ventures <span class="mywpglossary-term-def" data-title="portfolio" data-url="https://www.vccafe.com/startup-and-vc-glossary/#portfolio">portfolio</span> companies. Compiled by Eze Vidra / Remagine Ventures for VC Cafe, November 2026. Sources listed in the graphic.</figcaption></figure>



<p class="wp-block-paragraph">Several areas stand out.</p>



<p class="wp-block-paragraph"><strong>Moving and feeding data.</strong> DriveNets addresses networking for AI infrastructure, while WEKA focuses on the data infrastructure needed to keep AI workloads supplied with information. These are reminders that the performance of a computing system depends on much more than the processor itself.</p>



<p class="wp-block-paragraph"><strong>Operating the infrastructure.</strong> ClearML spans infrastructure management, <span class="mywpglossary-term-def" data-title="orchestration" data-url="https://www.vccafe.com/startup-and-vc-glossary/#orchestration">orchestration</span> and deployment. NeuReality combines hardware and software for <span class="mywpglossary-term-def" data-title="inference" data-url="https://www.vccafe.com/startup-and-vc-glossary/#inference">inference</span>. These businesses address the work required to turn computing resources into usable production capacity.</p>



<p class="wp-block-paragraph"><strong>Connecting <span class="mywpglossary-term-def" data-title="enterprise" data-url="https://www.vccafe.com/startup-and-vc-glossary/#enterprise">enterprise</span> data.</strong> Matia brings together data integration, reverse ETL, observability and catalog capabilities. This layer matters because an AI system’s usefulness depends partly on the quality and availability of the information it receives.</p>



<p class="wp-block-paragraph"><strong>Making AI dependable.</strong> Evaluation, monitoring and security become more important as systems take actions. Traceloop focuses on LLM evaluation and observability, while Prompt Security, acquired by SentinelOne, addresses AI security. Their work illustrates why the infrastructure opportunity continues above the model-serving layer.</p>



<p class="wp-block-paragraph"><strong>Power and compute.</strong> The physical foundations deserve equal attention. ZutaCore develops direct-to-chip liquid cooling to remove heat from demanding computing systems, while NextSilicon develops accelerated computing architectures for high-performance and AI workloads. They address different constraints, but both illustrate why access to models alone is insufficient: AI needs infrastructure that can run efficiently and reliably. For countries seeking greater control over their AI capabilities, power, cooling and computing capacity are part of the same strategic discussion.</p>



<p class="wp-block-paragraph"><strong>Models and adaptation.</strong> Israeli founders are also building the intelligence itself. AI21 Labs develops the Jamba family of models, while Decart combines generative-model research with infrastructure optimisation for real-time experiences. Troup AI, a Remagine Ventures <span class="mywpglossary-term-def" data-title="portfolio" data-url="https://www.vccafe.com/startup-and-vc-glossary/#portfolio">portfolio</span> company, addresses another part of this layer through <span class="mywpglossary-term-def" data-title="fine-tuning" data-url="https://www.vccafe.com/startup-and-vc-glossary/#fine-tuning">fine-tuning</span> and model swarms deployed within an organisation’s own environment. These approaches reflect different opportunities: developing models, making them more efficient, and adapting them to specific <span class="mywpglossary-term-def" data-title="enterprise" data-url="https://www.vccafe.com/startup-and-vc-glossary/#enterprise">enterprise</span> tasks. The sovereignty question becomes practical here: how much control can an organisation retain over the intelligence it uses and where it runs?</p>



<p class="wp-block-paragraph"><strong>Governing AI-generated software.</strong> As AI writes more code and agents take on more development work, organisations need ways to preserve their engineering standards. Pandorian, also a Remagine Ventures <span class="mywpglossary-term-def" data-title="portfolio" data-url="https://www.vccafe.com/startup-and-vc-glossary/#portfolio">portfolio</span> company, provides code governance that enforces an organisation’s standards across repositories and development workflows. It addresses a different problem from generating code: ensuring that the growing volume of human- and AI-written software follows the rules the organisation has established. Alongside evaluation, observability and agent security, this is an important part of the infrastructure required to put AI to work at scale.</p>



<p class="wp-block-paragraph">For Israel, this creates multiple routes to building significant companies. Competing in AI does not require every founder to compete directly with a frontier model lab.</p>



<h2 id="from-selling-infrastructure-software-to-running-the-workload" class="wp-block-heading">From selling infrastructure software to running the workload</h2>



<p class="wp-block-paragraph">There is also a business-model shift worth watching.</p>



<p class="wp-block-paragraph">Notable Capital’s Glenn Solomon and Dan Cahana argued in a recent <a href="https://www.calcalistech.com/ctechnews/article/nomqxac8j" data-type="link" data-id="https://www.calcalistech.com/ctechnews/article/nomqxac8j" target="_blank" rel="noopener">Ctech post</a> that agents will expand infrastructure demand as they generate and operate more software. Their thesis is that companies taking responsibility for running a workload can capture a larger share of spending than vendors selling management software around someone else’s infrastructure.</p>



<p class="wp-block-paragraph">They call these businesses “workload clouds,” citing Snowflake as an early template and Vercel, fal and Browserbase as examples across other workloads. Their expectations for declining software attach rates are forecasts, not established market outcomes.</p>



<p class="wp-block-paragraph">The distinction matters for founders. Providing a tool that helps a customer operate a system is a different business from becoming the service on which that workload runs. The latter can create a deeper customer relationship, but also brings responsibility for capacity, <span class="mywpglossary-term-def" data-title="uptime" data-url="https://www.vccafe.com/startup-and-vc-glossary/#uptime">uptime</span>, support and infrastructure costs.</p>



<p class="wp-block-paragraph">Higher revenue capture does not automatically mean better margins. The opportunity is to deliver a workload more effectively and <span class="mywpglossary-term-def" data-title="build" data-url="https://www.vccafe.com/startup-and-vc-glossary/#build">build</span> a business around that advantage.</p>



<h2 id="israel-faces-two-separate-infrastructure-questions" class="wp-block-heading">Israel faces two separate infrastructure questions</h2>



<p class="wp-block-paragraph">The first is whether Israeli founders can <span class="mywpglossary-term-def" data-title="build" data-url="https://www.vccafe.com/startup-and-vc-glossary/#build">build</span> globally important AI infrastructure companies. The landscape shows considerable activity already.</p>



<p class="wp-block-paragraph">The second is whether startups, researchers and institutions operating in Israel have sufficient access to affordable compute and the capabilities to use it. Success on the first question does not automatically solve the second.</p>



<p class="wp-block-paragraph">Speaking at the Journey conference in October, Amnon Shashua argued that AI requires a different scale of infrastructure investment from the software businesses Israel historically excelled at building. His warning was that talent and ingenuity need to be matched by computing resources and a deliberate strategy for allocating them.</p>



<p class="wp-block-paragraph">It would be wrong to conclude that nothing is happening. Israel’s National AI Program explicitly includes compute access, data infrastructure, research and talent among its priorities. The relevant debate is whether implementation, availability and affordability match the needs of the ecosystem.</p>



<p class="wp-block-paragraph">My view is that Israel should pursue both objectives: help founders <span class="mywpglossary-term-def" data-title="build" data-url="https://www.vccafe.com/startup-and-vc-glossary/#build">build</span> infrastructure businesses for global customers, and ensure that companies and researchers at home can access the resources required to compete.</p>



<p class="wp-block-paragraph">A country can export excellent infrastructure technology while remaining dependent on external providers for its own critical workloads.</p>



<h2 id="what-i-would-ask-founders-building-in-this-market" class="wp-block-heading">What I would ask founders building in this market</h2>



<p class="wp-block-paragraph">At the <span class="mywpglossary-term-def" data-title="pre-seed" data-url="https://www.vccafe.com/startup-and-vc-glossary/#pre-seed">pre-seed</span> and seed stage, “AI infrastructure” is too broad to be an <span class="mywpglossary-term-def" data-title="investment-thesis" data-url="https://www.vccafe.com/startup-and-vc-glossary/#investment-thesis">investment thesis</span> by itself. I would want to understand the specific constraint a company removes.</p>



<p class="wp-block-paragraph"><strong>What gets measurably better?</strong> Lower cost per completed task, faster deployment, higher utilisation, better reliability or stronger control are tangible outcomes. A <span class="mywpglossary-term-def" data-title="benchmark" data-url="https://www.vccafe.com/startup-and-vc-glossary/#benchmark">benchmark</span> improvement needs a connection to a customer’s economics.</p>



<p class="wp-block-paragraph"><strong>Who buys it?</strong> A data-centre operator, an infrastructure engineer and an <span class="mywpglossary-term-def" data-title="enterprise" data-url="https://www.vccafe.com/startup-and-vc-glossary/#enterprise">enterprise</span> AI team have different budgets, procurement processes and expectations. The route to market needs to fit the product.</p>



<p class="wp-block-paragraph"><strong>Why does this become a company?</strong> A useful feature may be absorbed by a cloud <span class="mywpglossary-term-def" data-title="platform" data-url="https://www.vccafe.com/startup-and-vc-glossary/#platform">platform</span>, database vendor or model provider. Defensibility might come from systems expertise, workload-specific optimisation, distribution or the operational difficulty of delivering the service.</p>



<p class="wp-block-paragraph"><strong>What happens when the models improve?</strong> Better models may eliminate some workarounds while increasing demand for data access, execution capacity and monitoring. Founders need a view on which side of that change they sit.</p>



<p class="wp-block-paragraph"><strong>How dependent is the infrastructure company itself?</strong> A product promising flexibility should be clear about its own reliance on particular clouds, chips, model providers and deployment environments.</p>



<h2 id="the-practical-meaning-of-sovereignty" class="wp-block-heading">The practical meaning of sovereignty</h2>



<p class="wp-block-paragraph">For most startups, the sensible response is to use the best available models while making dependencies explicit. Know which workflows are critical. Retain control over business data and application state. Test an alternative model on real tasks. Understand what would continue working during an interruption and what would need to pause.</p>



<p class="wp-block-paragraph">A second provider account is not the same as a tested recovery plan. Equally, self-hosting is not automatically the right choice for every company. The cost and complexity should be proportionate to the importance of the workload.</p>



<p class="wp-block-paragraph">For Israel, the ambition should be broader: sufficient capability to keep building, researching and operating essential systems, alongside companies that become valuable suppliers to the rest of the world.</p>



<p class="wp-block-paragraph">The sovereignty debate gives us a useful starting question: what happens if access disappears? The infrastructure opportunity begins with the work required to make the answer less alarming.</p>



<p class="wp-block-paragraph"><strong>The companies that make AI more reliable, efficient and controllable will help determine who can <span class="mywpglossary-term-def" data-title="build" data-url="https://www.vccafe.com/startup-and-vc-glossary/#build">build</span> with it, and on whose terms.</strong></p>]]></content:encoded>
					
					<wfw:commentRss>https://www.vccafe.com/israels-ai-infrastructure-stack/feed/</wfw:commentRss>
			<slash:comments>0</slash:comments>
		
		
		<post-id xmlns="com-wordpress:feed-additions:1">14102</post-id>	</item>
	</channel>
</rss>