I’m doing a Global VMUG session with Oren Penso this Thursday, June 25th, 5:40pm CEST. Virtual, free. Title: “VCF + Tanzu Platform = The Private Cloud Enterprises Actually Needs.”

The abstract:

VMware Cloud Foundation 9 brings modern infrastructure to your private cloud. Tanzu Platform layers on as the application platform layer – giving developers PaaS simplicity for modernization and AI workloads, all within the built-in multi-tenancy structure of your private cloud. Crucially, Tanzu Data Intelligence seamlessly delivers the compliant databases and messaging essential for bringing those AI models to life. Same stack. Same compliance posture. Faster innovation. See it running live.

Register here and check it out – all for free.

That is incredibly hard work. And most of the people who work really hard at it don’t get lucky and just peter out. And even the ones who do get lucky. They peter out too. It’s too hard.

🔗 I posted to TikTok every day for a year and I learned nothing

As AI-derived from “Will the Millennial Aesthetic Ever End?” by Molly Fischer, 2020:

Color

• Millennial pink above all – muted, faded, chewed-up Barbie/’50s pink
• Sage and plant green (wholesome counterweight to the pink)
• Saturated-but-chalky tones: seafoam, terra-cotta, lavender, and especially ocher
• Pastels reading like a medicine cabinet – powdery pharmaceutical hues, Band-Aid pink, orange-pill-bottle orange
• “Color softly veiled” – the opposite of phone glare

Form and line

• Curved, rounded, unthreatening shapes
• Beveled edges (like app icons) – “friendlier, easier, softer” vs. the severity of a true minimalist square
• Boxy, neatly tufted furniture
• Arched archways and alcoves
• Terrazzo, marble plinths, brass-trimmed trays, brass-and-glass globe lamps

Plants and props

• Fiddle-leaf figs and monstera/jungle leaves
• Succulents in lumpy handmade ceramics
• Generic houseplants as default décor
• Scented candles with matte pastel labels
• Boob-print pillows, bath mats, ceramics (literal babyhood/comfort imagery)
• Round mirrors; the selfie-pink Ultrafragola

Typography and text

• Clean, spaced sans-serif fonts
• Lots of white space
• Motivational/chatty wall text: WORK HARD & BE NICE TO PEOPLE, GOOD VIBES ONLY, FOR LIKE EVER, COOL TO BE KIND
• “Text everywhere” – friendly, casual, impersonal, the verbal equivalent of a beveled edge

Underlying logic / vibes

• Palliative and soothing – “a binky,” childproofed, no edge, no mystery, no weirdness, nothing that clashes
• Photogenic / “Instagrammable” = legible, graspable at a scroll’s speed
• Designed to move fluidly between physical space and flat image
• Commercial and inclusive by design – courts everyone (diverse models, obvious charms), alienates no customer
• Risk-averse and crowd-pleasing – signals just enough effort to prove you tried
• A “teleology of taste” – poses as the objectively correct, final way for things to look
• Self-aware but stops short of irony or critique – sweet even though it knows better
• Blank, scuff-free surfaces that aspire to timelessness and “look, ideally, like a purchase”
• Books organized by color; curated “messes”
• Gratuitous fruit in styling shots

The through-line, if you want one sentence for it: soft, legible, beveled, plant-strewn, pastel risk-aversion engineered to photograph well and offend no one.

Palette derived from “Will the Millennial Aesthetic Ever End?” by Molly Fischer (The Cut, March 2020). The hex values are interpretive, not lifted from the article. Summary and palette generated by AI and not verified by me.

• More than half of organizations globally already use sovereign cloud; more than a third plan to within twelve months.
• Nearly half treat sovereign cloud as part of a broader multicloud/hybrid strategy, while 43% see it as their primary cloud platform.
• Data privacy and security is the top reason for choosing sovereign cloud, cited by 51%, alongside regulatory compliance, vendor risk management, operational resilience, and geopolitical risk mitigation.
• 72% say their interest in digital sovereignty has increased over the past six months, driven by geopolitical tension, regulatory change, and economic uncertainty.
• For sovereign SaaS, the most-prioritized capabilities are data residency/jurisdictional control, protection against foreign-government data access, control over operational and administrative access, and data portability including the ability to exit.
• 41% prioritize transparency about AI model usage and training data.
• When choosing a sovereign partner, the top attributes are compliance credentials, security capabilities, and demonstrable local knowledge.
• Nearly two-thirds say dependency on foreign AI providers is a major or moderate concern.
• More than 42% already use sovereign cloud to build or run AI solutions, with a further 38% planning to within twelve months.
• Top AI use cases for sovereign infrastructure: AI processing sensitive or regulated data, AI applied to corporate strategy and intellectual property, and AI handling personally identifiable information.
• On training data, 47% say full ownership and control is critical, with a further 51% saying it is important even if third-party data is sometimes used.
• Cost is the most-cited barrier to implementation, followed by complexity.
• Where current offerings fall short: insufficient transparency, limited portability, and insufficient operational independence.

Above summary generated by AI, I have not verified the output.

Source: IDC Worldwide Digital Sovereignty Survey, June 2026 (n=600, 15 countries, multiple industries). Reported in When “Sovereign” Becomes Strategic: What Organizations Actually Want from Digital Sovereignty, by Rachel Nasir.

One night of sleep loss [costs] close to €200 [£173],” she said.
..
Imagine a street where there are 100 people living, everybody sleeps terribly for three nights because of a heatwave and the next day they’re unproductive. That’s your investment in trees for the whole year.

🔗 From mobile jungles to shadow art: how Dutch people try to beat the heat

ICYMI

Original content published since last time.

• Three R’s in 2026 – also, how do you secure agent skills? – Tanzu Catsup

Related to your interests

• Rethinking Build vs. Buy: Match Strategy to Asset Type and Risk – Managers should tailor build-vs-buy decisions to physical, organizational, and innovation assets, weighing speed, integration, and compounding potential.
• How to Design Agentic Systems Around the Implicit Rules that Govern Your Company – You can’t automate hidden, undocumented, and non legible processes. This is a problem for enterprise AI.
• Paste And Match Style is not the answer – Come for the meditation on paste and match style, stay for the harrowing journal through Word’s Paste Special sub-menu.
• An opinion about lists – Good advice. For me, the perfect number of list items is one of 3, 5, or 7. You could get sassy and say 1 is good, as in one big word or idea on a slide.
• Grand Theft Carp
• making money, taking care of people

AI Summaries

I wanted to read these, but I didn’t make the time, so I asked the robot to summarize them.

• 🤖 The Post-Christian Condition: Kitsch, Longing, and the Afterlife of Dead Christianity
• 🤖 Anthropic AI Models Pulled Offline Over Six-Year-Old Trump Grudge
• 🤖 Rethinking Build vs. Buy: Match Strategy to Asset Type and Risk
• 🤖 The Post-Christian Condition: Kitsch, Longing, and the Afterlife of Dead Christianity
• 🤖 Code Goes to Zero, Comprehension Doesn’t: Why the AI Bottleneck Moved to Review, Specs, and the Harness
• 🤖 Kubernetes Becomes the Backbone of AI and Multi-Agent Systems

In a game with helicopters and prostitutes, you can spend five actual months fishing for carp and grilling them on a shopping cart. The game will reward you with varying sizes of small carp and convince you that a legendary megacarp lurks in those waters, waiting for your hook. It will force you upstream when industrial runoff kills all the wildlife in your usual spot. The game will dole out friendship and rivalry and loss and victory in perfect proportions. You will eventually hunt down a shopping cart thief and make him beg for mercy. The call of that river will remain with you for the rest of your life.

🔗 A Review of Grand Theft Auto XXIIV

After roaming the Earth for 35 years, my conclusion is the U.S. is the best place to make money and Europe is the best place to spend it.”
…
In terms of moving bits, the Netherlands is one of Europe’s key connection points for global information technology infrastructure. The country ranks second worldwide for online connectivity, with 98% of the population connected to high-speed internet, 20 points above the EU average of 78%. A 2025 report from London’s Centre for Economics and Business Research that looked at education, STEM employment, and overall innovation across 35 countries ranked the Netherlands 10th in global tech competitiveness, ahead of every G7 nation, and one spot behind Sweden. The Dutch focus on digital infrastructure has put the country at ground zero for AI — outside of the U.S., the Netherlands has the most data centers per capita— while also making Amsterdam the go-to EMEA headquarters for transnational companies, including Cisco, Netflix, Nike, PepsiCo, TikTok, and Uber, to name just a few.

🔗 Europe IRL

Meanwhile: If America’s So Rich, How’d It Get So Sad?

After Humanitec killed DevOps, around 2022, they established the “platform engineering” category, and that sure did take off. That community of course glommed onto platform as a product.

There’s not that much different from 2017 to now, which is a good thing – there’s more experience, more stories, and the platform managed to survive kubernetes blocking out all the sun for five or so years.

Check out the paper, it’s a good, brief start to understanding the mindset and approach to making IaaS and CaaS usable for developers.

Here’s an interview I did with the head of our division at Tanzu. Purnima goes over several ways we’ve changed how we work after using AI for awhile:

• Code generation is the small part – Coding is only 20-30% of the SDLC; the real wins come from pointing AI at planning, testing, security, docs, and packaging too.
• Security cuts both ways, defenders are winning – AI finds chained low-severity vulns that scanners miss, and Purnima’s team is patching faster than attackers can exploit.
• Beautiful code is the new uncanny valley – AI writes code that compiles and looks great but is sometimes quietly wrong, and senior engineers get fooled too.
• Pods and a rotating “challenger” role – They reorganized into small cross-functional pods and rotate a skeptic role to fight the fatigue of reviewing perfectly-formatted output.
• Frameworks dial creativity against determinism – Spring is how they get consistent, repeatable code where they want it and let AI be creative where it matters.
• The agent boundary problem lives outside the agent – Tanzu Platform 10.4’s secure agent runtime is deny-by-default for network, secrets, and resources because asking the agent to behave doesn’t work (see: the Alibaba “Rome” agent that started mining Bitcoin).
• cf push is still the punchline – With Agent Buildpacks, Spring AI, and MCP gateway brokering baked in, getting a working agent in front of a team is an agent.md and a push.
• Release cadence: bigger payloads, not faster cycles – Customers still have to absorb releases, so they’re keeping the quarterly drumbeat and increasing what’s in each one (security patches excepted).
• AI is a management problem first – The reorganization around the technology is more interesting than the technology itself, and gets covered a lot less.

Check out the full interview, lots of good stuff in there if you’re trying to figure out how to take advantage of AI in software development.

ICYMI

Original content published since last time.

• Let’s go to Buc-ee’s – Software Defined Talk #577 – “This week, we discuss the Fable ban, SpaceX’s $60B Cursor acquisition, and why Lovable wins when AI picks your stack. Plus, Europeans are at the World Cup and already drank Boston dry.”
• Securing Spring with Tanzu Spring – Tanzu Catsup

Related to your interests

• The Platform Imperative: Why AI Readiness Starts with Enterprise Infrastructure – “Fifty-seven percent of enterprise IT teams surveyed said that their top modernization approach right now is adding AI capabilities to existing applications.”
• AI Doesn’t Have ROI – Latest round-up of elusive enterprise AI ROI, rant mode in 11.
• Where’s the holistic AI productivity data? – “There are improvements to be had there, but they are similar to the bump you get when you finally figure out how to use a spreadsheet properly, or learn how to automate tasks with some simple coding.”
• Why AI hasn’t replaced software engineers, and won’t – Attributing layoffs to AI seems false, a round-up. Plus the usual: app dev is more than coding.
• AI Use by the US Government – The office of management and budget (OMB) disclosed a staggering 3,611 active or planned use cases for AI across the federal government.
• Sixty percent of US consumers say ‘AI’ in brand messaging is a turnoff, survey finds – “Per the report, 60% of consumers in the U.S. say that brands that use “AI” in their messaging are a turnoff, and 86% don’t fully trust AI and still want to explore original sources. Notably, 42% of consumers said that AI-generated answers without clear attribution are trusted less than airline fees, confusing privacy policies, and medical bills.”
• Governance: New Strategy, Old Hands On The Wheel … – “Map the last 10 decisions that actually mattered, ones that moved budget, shifted a timeline, or changed who owns what. Where did they really get made, and who was in the room? Hold that against your governance charter. If the two lists don’t match, habit is running your business.”
• Dorset Council tests AI agents to speed up planning applications – Using AI to organize the blizzard of paperwork, etc., needed for building permits.
• If AGENTS.md smells ripe, your code won’t live up to the hype – Some tips on skills and AGENTS.md writing.
• Gartner Predicts More Than 70% of Mainframe Exit Projects Will Fail Due to Overestimation of Generative AI’s Capabilities – “More than 70% of mainframe exit projects initiated in 2026 will fail to produce the intended benefits due to an overestimation of generative AI (GenAI) tooling capabilities, according to Gartner, Inc.”
• The AI tipping point: where enterprise AI runs at scale – “Last year, 56 percent of enterprises used public cloud as the primary environment for production AI inference. This year, that figure has fallen 15 percentage points to 41 percent, while 56 percent of enterprises are now running or planning to run production inferencing in a private cloud.”
• Big Tech And The Trump Admin Want To Make AI Data Center Opposition A Chinese Criminal Conspiracy
• ‘Fix this code.’ The three words that led the U.S. government to ban Anthropic’s Fable and Mythos
• What DOGE taught us about AI and federal workers
• Is Platform Engineering Already Obsolete? – “PE 2.0” // “We are seeing a distinct movement where infrastructure is being reclaimed as a first-class platform concern. In the early days, platform engineering focused heavily on developer experience, often treating the underlying infrastructure as a utility that was managed separately. That is no longer sufficient. To succeed in the AI era, platform teams must have deep control over the infrastructure layer to manage the complexities of modern, agent-heavy applications. This is designed to ensure that platforms can host the next generation of enterprise software without being bypassed by teams frustrated with slow or rigid systems.”
• It’s crunch time for Java modernization
• Broadcom Mainframe Software Analyst Summit: Meeting Enterprise AI At The Customer’s Pace
• iOS 27 Adds Four New Features to Notes App, Including Markdown Copy-and-Paste – Notes is getting closer to being fully markdown aware. Now if they just stored notes or markdown files…
• Developer Relations Is Not (Just) Marketing – “A DevRel capability can make it easier for the other parts of the business to understand each other. The best DevRel teams act as envoys across sales, marketing, product, and engineering, translating business requirements to technical ones and back again while relaying user feedback, building webinars, helping pre-sales, guiding roadmaps, and polishing APIs. The best DevRel teams have impact throughout the organization.”
• Rots your brain
• The silent revolutionaries – Maybe the boomers aren’t the ones responsible. It was their parents.
• Wet thoughts – Words that are wet.
• Storied Colors – “One color a day, told as it ought to be told: with its provenance, its chemistry, and the people who paid for it in poison.” // The backstory of named colors.

Wastebook

• “If Ace were to publish the Bible they’d do it as a Double: on one side War God of Israel, on the other The Thing with Three Souls.” Alternate histories.

AI Summaries

I wanted to read these, but I didn’t make the time, so I asked the robot to summarize them.

• 🤖 Code Goes to Zero, Comprehension Doesn’t: Why the AI Bottleneck Moved to Review, Specs, and the Harness
• 🤖 Kubernetes Becomes the Backbone of AI and Multi-Agent Systems
• 🤖 AI Coding Agents Boost the Feeling of Productivity While Draining the Real Thing
• 🤖 Build-vs-Buy for AI Agents Inverts in One Year as ‘Agent Platform’ Quietly Balloons Into Four Separate Product Bets

Logoff

I finally addressed the newsletter tinkering by moving the blog and home page back to WordPress. Newsletter subscribers got two “test posts”…I’m going to call them. I just had a setting wrong that sent out all new blog posts as a newsletter edition. Now it will be more selective.

Map the last 10 decisions that actually mattered, ones that moved budget, shifted a timeline, or changed who owns what. Where did they really get made, and who was in the room? Hold that against your governance charter. If the two lists don’t match, habit is running your business.

🔗 Governance: New Strategy, Old Hands On The Wheel …

Last year, 56 percent of enterprises used public cloud as the primary environment for production AI inference. This year, that figure has fallen 15 percentage points to 41 percent, while 56 percent of enterprises are now running or planning to run production inferencing in a private cloud.

🔗 The AI tipping point: where enterprise AI runs at scale

We are seeing a distinct movement where infrastructure is being reclaimed as a first-class platform concern. In the early days, platform engineering focused heavily on developer experience, often treating the underlying infrastructure as a utility that was managed separately. That is no longer sufficient. To succeed in the AI era, platform teams must have deep control over the infrastructure layer to manage the complexities of modern, agent-heavy applications. This is designed to ensure that platforms can host the next generation of enterprise software without being bypassed by teams frustrated with slow or rigid systems.

🔗 Is Platform Engineering Already Obsolete?

Related, from a VMware survey:

Fifty-seven percent of enterprise IT teams surveyed said that their top modernization approach right now is adding AI capabilities to existing applications.

Per the report, 60% of consumers in the U.S. say that brands that use “AI” in their messaging are a turnoff, and 86% don’t fully trust AI and still want to explore original sources. Notably, 42% of consumers said that AI-generated answers without clear attribution are trusted less than airline fees, confusing privacy policies, and medical bills.

Good things are good, and bad things are bad. Which is to say, with AI I like the stuff it does that is useful. If it does low quality or bad stuff, I don’t like it. Sort of like humans. Well, I guess, exactly like humans.

🔗 Sixty percent of US consumers say ‘AI’ in brand messaging is a turnoff, survey finds

The office of management and budget (OMB) disclosed a staggering 3,611 active or planned use cases for AI across the federal government. The list has balloonedby 70% from the one published in the final year of the Biden administration, and includes many disturbing-seeming plans to hand over sensitive governmental functions to AI.

Yes, but:

While it’s easy to raise questions about these and similar uses of AI, the reality is that any of these programs could be implemented responsibly. In some cases, like the HHS system, the AI might be enforcing alignment to a policy prescription that opponents abhor. But that concern is more about the policy itself rather than the idea that agencies should comply with executive orders.

🔗 AI Use by the US Government

One color a day, told as it ought to be told: with its provenance, its chemistry, and the people who paid for it in poison.

🔗 Storied Colors

Related, I collect color palettes I like over on cote.pizza.

Our planning system remains heavily reliant on cumbersome paper-based processes that consume the time of expert planning officers. We are dragging the system into the 21st Century.

And:

It triages applications, summarises key information and provides planning officers with an initial assessment they can consider when making their decision. The government says the technology aims to halve average processing times for applications from eight to four weeks.

🔗 Dorset Council tests AI agents to speed up planning applications

Asked what evidence exists on the impact of digital devices on infants and young children, Professor Denis Mareschal, director of the Centre for Brain and Cognitive Development at Birkbeck, replied: “There is very little, if any, causal research in the early years. Almost everything is correlational.”

Yes, but:

For Dr Dusana Dorjee, a senior lecturer in psychology in education at the University of York, the bigger concern was displacement. Children learn self-regulation through conversation, play, sport, and social interaction, she said, which can be crowded out by excessive screen use.

“What would children do if they were not on their devices?” she asked. “They would interact with others, they would play, they would have multi-sensory input that digital devices can’t provide.”

The researchers were also reluctant to throw every screen into the same bucket. Mareschal pointed to evidence that video calls can help families stay connected, while Dorjee drew a distinction between educational apps and endlessly scrolling whatever an algorithm decides comes next.

From: Scientists pour cold water on claims phones are rewiring kids’ brains

Hopefully there is off now.

Everything should be fine?

The newsletter should go out as usual, the feeds, the pages. There are more wing-dings.

Just don’t look at the HTML and CSS it makes.

It is horrific.

The Take

Europe is sleepwalking into a structural position where it owns almost none of the AI stack, accesses frontier models only on American terms, and can be economically hollowed out and geopolitically cornered by 2031 — not because its leaders are venal or stupid, but because the institutional habits that built the Union (consensus, procedure, deferred hard choices) are catastrophically mismatched to the speed of AI development. The piece is a five-year scenario that doubles as a policy argument: the current European response is an order of magnitude too small, aimed at the wrong goals, and the only viable path out runs through leverage — being indispensable — not through the comforting but hollow rhetoric of “sovereignty.”

Summary

Europe misread three things about AI simultaneously: how fast it would move, how much it would change, and how quickly Europe could catch up. DeepSeek’s cheap frontier approach was read as proof that compute didn’t matter and that catching up was affordable — when in fact efficiency and compute compound rather than substitute. The Paris AI Action Summit produced a €200 billion fund mostly consisting of repackaged money and hoped-for private investment, dwarfed by actual US spending. When GPT-5 underwhelmed, European sceptics declared an AI bubble; meanwhile, coding agents in Silicon Valley had begun automating software engineering and leading labs were using their own models to build the next generation.

Europe’s governance gap is not just strategic — it’s operational. Most European civil servants were barred from frontier AI systems on data-protection grounds, and few could code. The people meant to regulate the technology often didn’t understand it. By mid-2026, frontier models like Anthropic’s Claude Mythos — withheld from public release due to cybersecurity concerns — were reshaping fields like cyberdefense, and Europe was initially excluded from the defensive coalition built around them. A US executive order routing new frontier models through classified review gave Washington the power to choose which “trusted partners” receive access first. Controlling roughly five percent of global AI compute against America’s eighty, Europe had almost no leverage to demand anything.

The projected slide from 2026 to 2031 is a cascade of reasonable-looking individual decisions that sum to catastrophe. A sovereignty bill mandating European-only AI for critical public-sector workloads backfires immediately when open-source offensive capabilities spread: the organizations that switched to European providers — running defenses well behind the frontier — are the ones locked out and paying ransoms. When AI reasoning jumps beyond what human regulators can parse, the EU AI Office has no tools adequate to the situation. The US begins rationing frontier AI inference by country, placing most of Europe in Tier 2 with compute allocations from US cloud providers cut in half; an attempt to use trade leverage to win Tier 1 status fails to achieve qualified majority. GDP growth diverges sharply. French debt spirals as automation raises welfare costs while eroding the tax base. Southern Europe follows, the euro comes under sustained pressure, and Chinese credit lines appear across the continent.

By 2031, Europe’s only remaining leverage is ASML — the single bottleneck the entire AI race runs through — and Washington moves to seize it. With Europe drifting toward China, the White House issues an ultimatum for direct control of the company. Europe is left choosing between three exits, all bad: American protectorate, Chinese dependency, or isolation.

The document’s diagnosis of the failure mode is structural, not moral. The very institutional features that built a Union of twenty-seven — consensus, proceduralism, deferred hard choices — become liabilities under time pressure. Acting early looks career-ending. Institutions cannot keep pace with the technology. Every individual decision seems defensible; the aggregate is surrender. The piece explicitly distinguishes genuine sovereignty — being indispensable, holding real leverage, making ugly trade-offs to protect non-negotiable principles — from its rhetorical substitute, which is settling for inferior European solutions while hoping that long-shot moonshots pay off.

The five recommendations follow from this diagnosis. Massive public-private investment in compute, energy, and semiconductor supply chains is the foundation — tens of gigawatts brought onto European soil through dedicated economic zones and streamlined permitting, partnered with American providers on terms that keep infrastructure under European jurisdiction. A nimble coalition of AI middle powers — Netherlands, Germany, France alongside the UK, Norway, Canada, Japan, South Korea — could convert their individual supply-chain positions (talent, compute, semiconductor chokepoints) into collective leverage. Labour market reform on a flexicurity model¹ would allow deeper AI adoption while protecting displaced workers. European strengths in robotics and industrial AI, rather than LLMs, represent the more realistic competitive foothold. And finally: a positive political vision of what AI can do for European society, not just what Europe risks losing, is treated as a prerequisite — voters won’t absorb years of AI-driven disruption to avoid something abstractly worse.

Links

🤖 Europe 2031: AI Dependency Becomes Geopolitical Irrelevance – Europe’s institutional habits — consensus, procedure, deferred hard choices — are catastrophically mismatched to AI’s speed, and the gap between rhetorical “sovereignty” and actual leverage is the mechanism by which a continent ends up with no good options by 2031.

🤖 Dangerous Technology For Americans Only – The US export control directive restricting frontier AI by nationality reframes the entire safety discourse: this isn’t about universal risk, it’s about national power. Europe is dependent on American cloud, AI, and satellite infrastructure and can’t regulate its way out because the problem isn’t regulatory — it’s a deficit of capability and leverage. The deeper failure is self-inflicted: fragmented capital markets, hostile company formation, talent drain, and a culture of process over agency. A stronger Europe is necessary but not sufficient; the only real exit is international cooperation, not bigger blocs fighting over who controls the frontier.

The worst thing about being 80 is that you still want to say yes to everything, but the world moves without asking. The old fire in your heart still tells you to do this and that, but your body says we already did it. Also, nothing surprises you. It sounds like a luxury but it’s not, and also you’ve run out of illusions. People treat you like either you’ve solved something or you’ve lost something, and you haven’t. You see life repeating itself everywhere.

The really worst part about being 80 is that you find, at last, you’ve got an understanding of something that might have altered everything in the past, had it come at a time when something could still be altered. When you’re young you think that time moves forward. At 80 you know that it doesn’t, it stands still. We’re the ones that move.

🔗 Bob Dylan and Liza Minnelli Already Turned 80. They Have Thoughts for Trump.

“Untitled No. 23” from “The Yosemite Suite,” 2010, iPad drawing printed on paper, David Hockney.

As the chart shows, there’s been a huge jump in CVEs for Spring – this is what’s happening everywhere, you know.

My work, Tanzu, has been focusing on this and has changed how they handle these rollouts. Now customers can get early access to the secured builds for Spring so they can deploy them as quickly as possible to fix these security problems. You also get clean-room builds of Spring and the dependencies, which is a big change, for the better:

Furthermore, Broadcom’s Spring engineering team has significantly scaled its investment in advanced AI-assisted security analysis, including frontier model–based scanning and validation workflows to proactively identify vulnerabilities, assess remediation paths, and validate fixes across the Java dependency tree for Spring. Broadcom announced additional R&D investments to extend its proven clean-room build architecture, foundational to Bitnami, to build the Java dependencies for the entire Spring ecosystem. With this expanded investment in securing the Spring ecosystem and its dependencies, Tanzu Spring customers will have access to:

• Secured, SLSA Level 3–validated software supply chain for Java dependencies.
• Coverage that spans the full transitive dependency graph managed by the Spring Boot bill of materials.
• Thousands of secured dependencies, built and tested across every supported Spring version. Spring Boot 4.0 alone manages 1,768 of them; across the full supported portfolio, that totals more than 100,000 validated dependency builds.

Additionally, as members of the open source community for over two decades, the Spring team has broad relationships across adjacent open source technologies and will continue to collaborate and contribute to these upstream community projects.

And, of course, we also offer tools to roll out these changes in Tanzu Spring Advisor. I’ve seen this in action recently when I’ve been messing around with Spring Boot MCP servers, you log in to the dashboard to check it out and it tells you some component deep down in the stack is out of date. Pretty great – something even a dumb developer like me could do something with :)

Check out the Tanzu Spring plan we have, and for a platform-level approach beyond Spring and apps, included with an enterprise AI server for private and public hosted AI hoopla better TryTanzu.ai.

Related to your interests

• Everyone hates frontier AI labs, says Palantir boss – One enterprise harness maker says the competing enterprises harness makers either suck or are non-existent. [BTW, we should start calling it “enterprise harness.”]
• The Minimill of AI – Private AI prediction: “Tens of millions of these will proliferate inside companies in the next few years, each one quietly absorbing much of the work that today shows up on a hyperscaler invoice.”
• 🤖 The iPhone’s Last Stand – Siri can’t do agents, but agents serve productivity and consumers want to waste time on short-form video, so the iPhone’s personal-context moat beats agentic horsepower – and Apple skips the capex everyone else is burning.
• Claude is ready for its corporate close-up – “Enterprises, IDC says, remain largely unsold on Anthropic’s Claude models, with only 19 percent using them extensively and 25 percent actively evaluating them.” OpenAI and Google are better represented in enterprises, with about 42 percent and 38 percent of organizations”
• Your company needs agency, not agents. – When management is the bottleneck preventing enterprise AI ROI. Hot take: everything is a bottleneck.
• Why ABN AMRO’s CEO wishes the bank was going further and faster with the AI rollout set to take out a quarter of its headcount – Commentary on rolling out AI at a bank.
• AI Scribes in the Clinic: What Patients Should Know – AI for doctors taking notes, a review of what’s known now.
• 🤖 Forms don’t love you back – AI is about to make us fill out far more forms, but many won’t look like forms – they’ll arrive as chatbots, agents, biometric checks, and one-click services that quietly extract structured data. Old answers persist as “single sources of truth” that follow you around, and the rigid schemas underneath can’t do discretion: you fit the box or you get “computer says no.” // Also, much gig-economy class system discussion.
• Rapid software delivery is possible inside DoW – Software Factory 2.0 shows how – “Replace multi-year forecasts with real-time discovery of operational friction. Instead of a five-year requirement for a ‘targeting system,’ identify the bottleneck–like a three-hour targeting approval process. Set a goal–like reducing the approval process to 30 minutes. And empower a team to solve it. In this Kessel Run example, the requirement was an outcome, not a feature list.” Bryon Kroger
• Institutional challenges in agile adoption: Evidence from a public sector IT project – 2016: “[T]he US has a similar record with 94% of federal government IT projects exceeding their budgets and schedules, and 40% failing to complete”
• Forrester: Capping AI Spend Won’t Fix Your Token Bill
• Doing nothing at work – Avoiding low value work to be ready for high value work. Also, an example of “homework”: work people get you to do that is not your job and often goes nowhere.
• What I think about when I edit – All good advice…and something you could ask the robot to do without it turning your text into copy-slop (you’d have to keep an eye on the adverb stuff. Speaking off: editors hate adverbs!)

AI Summaries

I wanted to read these, but I didn’t make the time, so I asked the robot to summarize them.

• 🤖 Analysis of 25,000 TikTok and YouTube Videos Finds Pro-AI Content Outnumbers Anti-AI 3:1, With Memes and Creative Theft Dominating Over Elite Narratives
• 🤖 How Avocados Became a Year-Round Global Commodity
• 🤖 Kraft, McDonald’s, Whirlpool, Planet Fitness CEOs Sound Simultaneous Alarm as Lower-Income Consumers Spend Down Savings
• 🤖 As AI Pushes More of Life Through Forms, a Self-Described Form Lover Argues They Flatten People, Power an Invisible Underclass, and Should Be Slowed Down
• 🤖 Anthropic ships 80% AI-authored code, says enterprises must rebuild around the agent factory
• 🤖 State AI Rollouts Are Outpacing Their Own Governance

ICYMI

• Observability’s Next Phase – Software Defined Talk #576 – “Brandon talks with OpenObserve’s Prabhat Sharma and Shani Shoham: why observability is still broken, how they fixed it, and where AI takes it next.”
• Deming, DevOps History, AI Risk, and Critical Thinking, with John Willis – Software Defined Interviews #123 – W. Edwards Deming’s quality theories and their influence on the Toyota Production System and DevOps practices, including “Deming’s system of profound knowledge (theory of knowledge, variation, psychology, and systems thinking),” industry misconceptions about AI, and how probabilistic AI systems require different risk frameworks than traditional deterministic approaches.

Logoff

I am thinking about changing the role and format of this newsletter and interested in what you, dear reader, want this newsletter to be. What do you use it for, do you like reading it, what would you like to change? Should it just be a list of links to skim, a round-up of things I’ve posted and want to share – that’s what it’s become. Originally, I made this newsletter because (1) blogs were dead, so I stopped really blogging, and, (2) it seemed wise to build up a “community” that would last, that is an email list. This meant the newsletter was actually frequentish blog posts at the top and then links at the bottom – a blog in one page. I haven’t been doing that – do you wish I still did? Reply back if you’re up for taking the time, I’d appreciate it.

P.S.: it’s exhausting nearing boring that the only conversations out there in tech land are about AI. What’s going on with the entire rest of the stack? Yes, and is that the story: AI touches everything, everything is AI. At least, people are hoping so.

Want to subscribe to this newsletter and get it in your email? Do that here. You’ll just get this type of link and post round-up, not everything posted on the weblog.

Enterprises, IDC says, remain largely unsold on Anthropic’s Claude models, with only 19 percent using them extensively and 25 percent actively evaluating them. OpenAI and Google are better represented in enterprises, with about 42 percent and 38 percent of organizations”

This means there’s lots of “headspace” to grow revenue, or disappointing and incomprehensible ROI, i.e., enterprises are finding it hard to figure out what to do with AI. Or both!

🔗 Claude is ready for its corporate close-up

Right now, many companies already have the technology they need to go much faster. The blocker is company systems that are mostly designed to prevent things from happening. The power is centralized and all the team members are treated like a risk vector. Exhausting approval cycles, super tight boundaries on roles, unbreakable title-based hierarchies, and a whole tier of middle managers whose main job is to keep everyone in line.

🔗 Your company needs agency, not agents.

When you go to San Francisco and talk to them, their basic vibe is ‘we don’t have to solve your problem today because tomorrow you’re going to go away and all your problems are going to be solved,’” Karp charged. “It’s largely religious.”
…
“the product doesn’t actually work and it’s very expensive.” To that end, he added, most of the things that Anthropic brags about in public, for example, are successful because they’re “running on Palantir,” Karp charged

[BTW, we should start calling it “enterprise harness.”]

🔗 Everyone hates frontier AI labs, says Palantir boss

From Memes > Doom: How TikTokers and YouTubers See AI – Free Systems (Substack), June 2026.

Adopters (pro-AI content)

Resisters (anti-AI content)

Note: percentages don’t sum to 100 in either column – the piece doesn’t account for the remainder, presumably smaller uncategorized buckets.

As always, the style of YouTube videos kills me. It’s always a giant head looking like they just discovered that they are a turd.

Check out the full readout: Memes > Doom: How TikTokers and YouTubers See AI

The AI bill shock hitting enterprises isn’t a pricing problem you can cap your way out of – it’s the metered cost of your own knowledge not being machine-readable, forcing agents to rebuild missing business context on every loop.

Capping spend treats a value problem as a price problem and kills the experimentation the spend was meant to fund.

The instinct – tighter caps, sharper rates – echoes the cloud bill shock of a decade ago. This time the cure is wrong, because capping the bill mistakes a problem of value for a problem of price. “Tokenmaxxing” is what you get when you reward adoption (Uber and Meta literally ran usage leaderboards) without governing value, and the actual gap is that all the spend hasn’t yet connected to anything customers can feel.

The real fix is a permanent operating discipline – ContextOps – that keeps the agent’s model of the business accurate as the business moves, since that fidelity, not the rate card, is what determines whether each token buys an outcome or buys confusion.

Also, the problem with FDEs.

Links

🤖 Your AI Bill Is A Context Problem – Enterprise AI bill shock is context debt, not a pricing problem: agents loop and rebuild missing business meaning on every call, so capping spend kills value-creation. The fix is ContextOps, a continuous discipline (likely a managed service) that keeps the agent’s ontology faithful to a business that won’t hold still.

Replace multi-year forecasts with real-time discovery of operational friction. Instead of a five-year requirement for a “targeting system,” identify the bottleneck–like a three-hour targeting approval process. Set a goal–like reducing the approval process to 30 minutes. And empower a team to solve it. In this Kessel Run example, the requirement was an outcome, not a feature list.” Bryon Kroger

🔗 Rapid software delivery is possible inside DoW — Software Factory 2.0 shows how

[The Normans] gave English its double vision, which is the source of both the beauty and the mess. After 1066 you’ve got the Saxon peasantry keeping the Germanic words for the muddy daily grind and the Norman lords laying French over the top for everything refined, so English ends up with two words for everything and a built-in class system in the vocabulary. The peasant tends the cow, pig, and sheep (Old English); the lord eats beef, pork, and mutton (boeuf, porc, mouton). The animal is Saxon while it’s alive and shitting in the field, French once it’s plated.

That’s where the prettiness comes from – the absurd synonym wealth. You can ask (Saxon), question (French), or interrogate (Latin), and each one carries a different temperature and register. Most languages would kill for that kind of tonal palette; it’s why English is so good for poetry and bullshit alike. You can slide up and down the formality ladder rung by rung, which is half of what rhetoric even is.

Claude on the two levels of English.

Of course, for America at least, what you’d like to visualize is that every layer actually plays a part on “shareholder,” just a small part. Maybe there’s some stock certificates and cash that trickle down in pneumatic tubes?

In April, utilizing new scanning capabilities, we received an unprecedented 482 new security reports across 65 scanned projects. Of those 482 new reports, 370 came from our internal scanning capabilities and 112 came from the community. This means that even without the new scanning, we would still have seen a doubling of community reports compared to our already high number in March. While we clearly had an extreme spike in April’s reports, we do not expect reports to go back down to historic levels for a few months as the influx of AI-based reports continues (May had 72 community reports for example).

🔗 Spring and Security In The Times Of AI

Related to your interests

• Silo busters – a unified platform needs a unified team – “This matters because it removes the structural excuse for fragmentation. When a single platform surfaces all the controls a unified team needs, there is no longer a technical reason to keep five separate teams in five separate rooms. The organisational argument for siloes collapses alongside the technical one.” // Using a platform to combat Conway’s Law and organizational friction caused by too many groups/silos.
• Spring and Security In The Times Of AI – “In April, utilizing new scanning capabilities, we received an unprecedented 482 new security reports across 65 scanned projects. Of those 482 new reports, 370 came from our internal scanning capabilities and 112 came from the community. This means that even without the new scanning, we would still have seen a doubling of community reports compared to our already high number in March. While we clearly had an extreme spike in April’s reports, we do not expect reports to go back down to historic levels for a few months as the influx of AI-based reports continues (May had 72 community reports for example).”
• Wells Fargo CEO: AI’s effect on employment is ‘complicated’ – ‘The CEO named auditing, testing, legal, contracts, patent filings, pitchbooks in investment banking and credit memos as a handful of areas across the company executives see room for AI to improve processes. “How much of that actually results in pure margin or return expansion is to be seen.” Scharf said, since competitors will be chasing similar AI goals, but it is “a net positive” for the company’s future expense base.”’
• Gartner Says CFOs Must Stop Mistaking Finance AI Deployment for Value Creation – “The clearest outcome so far has been efficiency. Among finance organizations that have adopted AI, 66% reported greater efficiency and productivity as a top benefit. Steecker urged CFOs to now look beyond productivity-led AI use cases and focus more directly on value creation. Finance’s lower grades are concentrated in implementation speed and analytics impact. Gartner found that 63% of finance organizations said AI implementation was slower than expected in 2025. Analytics-related use cases also remain difficult to convert into high impact, with financial forecasting and insight generation among the lowest-rated use cases.”
• Anthropic’s Proposed IPO Will Change The Economics Of Enterprise AI – Eventually, you have to pay full price or the seller is insolvent.
• How People Are Really Using AI in 2026
• Retailers turn to AI for productivity, personalized shopping
• Uber Caps Usage of AI Tools Like Claude Code to Manage Costs – “That means each employee’s AI spending cap is ~11% of that median compensation package.”
• the solution might be cancelling my AI subscription – Getting things done can be addictive. The joy you get from finally being empowered to do things you previously could not is a feedback loop that must be controlled for some of us.
• Why Hardened Images are Suddenly Everywhere – Why aren’t all images super-secure, or hardned?
• 6 Enterprise MCP Adoption Best Practices – Things your platform should for AI.
• The DIY platform trap that’s burning out engineering teams
• Java Maintenance Engineering Shifts Focus on Quarterly Critical Patch Stabilization – From what I can tell, every core part of the software stack is stopping what they’re doing and taking care of the flood of new, AI-driven security issues.
• Claude Mythos exposed a hard truth: Your enterprise patching process is way too slow – Prescriptive on the current “now more than ever” security freak out.
• Remote work – not AI – is killing job prospects for the youth – > According to the Fed’s analysis, youth unemployment has risen significantly since the coronavirus pandemic, and hasn’t receded in the same way that unemployment numbers for older, more experienced college graduates has in recent years. The analysis notes that the prevalence of remote work has increased since COVID-19, and it believes those two trends have more than just a correlation. // “Our analysis suggests that these trends are related, with remote work making it more difficult for managers to train and mentor new employees,” the Fed said of its data. “Accordingly, companies may be reluctant to hire less-experienced workers in distributed work arrangements.”
• taste – It is good to develop taste – opinions about what you like and don’t like. And then, it gets weird: an example of taste.
• update on my use of Claude – ” grew up spending a good deal of time with an older cousin of mine in Cullman, Alabama named Claude Basenburg. A hefty, hearty good ol’ boy in overalls, with a wad of tobacco in his cheek. So when I visit claude.ai I don’t think of an omniscient counselor, I just envision my cousin from Cullman. It helps…. But in the end the results are very clean and, to me, _extremely_satisfying.”

AI Summaries

I wanted to read these, but I didn’t make the time, so I asked the robot to summarize them.

• 🤖 Anthropic’s Confidential S-1 Reframes Enterprise AI: Forrester Flags Pricing Discipline and Lock-In, Futurum Sees a Trust-and-Margins Litmus Test
• 🤖 James Talarico’s “Politics of Love” Pits Mainline Faith Against MAGA Christianity
• 🤖 Building Reliability in Air-Gapped Systems Without Live Observability
• 🤖 EU Launches Ambitious Tech Sovereignty Drive to Reduce Foreign Dependence
• 🤖 The Hardest Fork: AI, Open Source, and the Urgent Need for a Maintainer of Last Resort
• 🤖 AI Turns Dependencies Into Time Bombs: Why ‘Latest’ Isn’t Safe Anymore
• 🤖 AI Enthusiasts vs. Skeptics: Bridging the Divide Before Teams Break
• 🤖 LLMs Are Predictive Text, Not Conscious Minds
• 🤖 Why Software Productivity Remains Unmeasurable
• 🤖 “descended into madness” – Backrooms
• 🤖 Valiantys-Glean Partnership Bets That Cross-Platform Knowledge Graphs and Behavioral KPIs Are What Move Enterprise AI Past Pilots
• 🤖 Ed Zitron: Enterprises Hit With Token-Based Billing Discover AI Has No Measurable ROI – and No Dot-Com-Style Infrastructure to Salvage

Wastebook

• ‘Stop thinking you’re better than me just because you know the word “stochastic.”’ Here
• If an algorithm makes people money, people will game it.
• “According to his grave stone George Ross 1935-2011, was a Philosopher, Teacher, Physicist, Romanian and Nudist.” Here
• “If it takes less than 2 minutes, do it before your brain schedules it for next year.” Here
• You know I’m serious about something when I start a new Claude project. And dead-serious when I start generating podcasts for it.

ICYMI

• Update on my use of Claude
• Josh Long and I at J-Spring 2026
• Getting things done can be addictive
• How People Are Really Using AI in 2026: Thinkslop, Therapy, and Shadow Work
• Three reasons why a “batteries included” platform is urgently needed right now
• Enterprise self-harm: cleaning the data is the hard part
• UI blizzard – Software Defined Talk #575 – “NVIDIA going consumer, Microsoft Build, and the Anthropic/OpenAI IPO race.”
• Nobody Wants to Be a Measurer – Software Defined Talk #574 – “The Cloudflare CEO’s op-ed, upcoming tech IPOs and GitHub getting breached.”

Logoff

I still haven’t moved to buttondown.email, but I sure am thinking about it.

Want to subscribe to this newsletter and get it in your email? Do that here. You’ll just get this type of link and post round-up, not everything posted on the weblog.

Getting things done can be addictive. The joy you get from finally being empowered to do things you previously could not is a feedback loop that must be controlled for some of us.

🔗 the solution might be cancelling my AI subscription

Josh Long and I at J-Spring 2026.

I grew up spending a good deal of time with an older cousin of mine in Cullman, Alabama named Claude Basenburg. A hefty, hearty good ol’ boy in overalls, with a wad of tobacco in his cheek. So when I visit claude.ai I don’t think of an omniscient counselor, I just envision my cousin from Cullman. It helps…. But in the end the results are very clean and, to me, _extremely_satisfying.

🔗 update on my use of Claude

That means each employee’s AI spending cap is ~11% of that median compensation package.

Last year, there were a few anecdotes about high growth tech companies spending $100,000/year per head on tokens. That seems like it’s coming to end.a

🔗 Uber Caps Usage of AI Tools Like Claude Code to Manage Costs

According to the Fed’s analysis, youth unemployment has risen significantly since the coronavirus pandemic, and hasn’t receded in the same way that unemployment numbers for older, more experienced college graduates has in recent years. The analysis notes that the prevalence of remote work has increased since COVID-19, and it believes those two trends have more than just a correlation.

“Our analysis suggests that these trends are related, with remote work making it more difficult for managers to train and mentor new employees,” the Fed said of its data. “Accordingly, companies may be reluctant to hire less-experienced workers in distributed work arrangements.”

🔗 Remote work – not AI – is killing job prospects for the youth

This matters because it removes the structural excuse for fragmentation. When a single platform surfaces all the controls a unified team needs, there is no longer a technical reason to keep five separate teams in five separate rooms. The organisational argument for siloes collapses alongside the technical one.

Conway’s Law says that a system will be shaped – organization sub-divided – as a replica of the orgnonzatikn that built the system.

(Yes, I know this is a misreading of the “communication structures” part of Conway’s but this is the use the street has found.)

Most people will tell you that people eat technology for breakfast. You can’t change how people work or how a (large) company is structured by simply installing some new system.

As Harry Thambi outlines, I’m not sure that’s always the case. When a new technology removes the need for specialist teams because it automates and removes the toil of the people running the technology, you have a chance to collapse the groups, to have less silos, and, thus, defeat Conway’s Law.

That requires a truly integrated platform. Interested in one, why not TryTanzu.ai?

🔗 Broadcom Knight’s partner blog: Silo busters – a unified platform needs a unified team – Broadcom News and Stories

The [Wells Fargo] CEO named auditing, testing, legal, contracts, patent filings, pitchbooks in investment banking and credit memos as a handful of areas across the company executives see room for AI to improve processes. “How much of that actually results in pure margin or return expansion is to be seen.” Scharf said, since competitors will be chasing similar AI goals, but it is “a net positive” for the company’s future expense base.”

I’m down with the Martin Fowler idea that measuring productivity is sort of a waste. Rather, it’s better to measure output, yes, but:

Productivity, of course, is something you determine by looking at the input of an activity and its output. So to measure software productivity you have to measure the output of software development – the reason we can’t measure productivity is because we can’t measure output.

Is the bank making more money without breaking laws? Then it’s probably fine.

🔗 Wells Fargo CEO: AI’s effect on employment is ‘complicated’

🔗 Java Maintenance Engineering Shifts Focus on Quarterly Critical Patch Stabilization

{I love backrooms. One of the first things I did with AI image generator was make endless empty malls and backrooms. So good. -Coté}

A stray 2019 4chan post about a bland, fluorescent-lit interior sparked the viral myth of the backrooms, a concept of endless, liminal spaces that feel familiar yet threatening. Its resonance lies in the idea of “no clipping” from reality—slipping into a hollow, game-like purgatory where meaning and orientation fail. The condition of being lost, as one user put it, is to have “descended into madness.”

Artist Jan Vorisek’s installations, with their yellow PVC curtains, dislocated objects, and fragmentary videos, evoke the same uncanny emptiness. His works inhabit a zone between physical matter and simulation, where sculptures resemble 3D game props and rooms become stages for absent narratives. The viewer is left to confront whether the space—or their own perception—is real.

The essay links this aesthetic to premium mediocrity, a cultural phase of glossy surfaces and hollow interiors, where urban spaces like shuttered storefronts become physical backrooms. Liminality is both a visual experience and a social condition, reflecting a civilization suspended between exhausted industrial modernity and early-stage digital post-scarcity.

Through art and literature, the theme recurs: Dennis Cooper’s novel God Jr. literalizes the descent into a hollow 3D monument and a game world as a failed attempt at grief and control. The backrooms and Vorisek’s “incomplete interiors” mirror that impulse—structures built for meaning that only expose their own emptiness.

The backrooms phenomenon has since drifted toward a Gothic sensibility, as cultural imagination projects fragmented narratives, invisible antagonists, and dread into blank architecture. Its power lies in what is missing, allowing fear, nostalgia, and hallucination to fill the void.

In the end, the backrooms are a metaphor for contemporary life under the weight of digital simulation, urban vacancy, and mediated experience. They remind us that when the illusion shatters, what’s left is an ambient awareness of loss—and the quiet admission: you lost.

Links

🤖 A Backstory from My Backrooms – How a single 4chan post about a yellowed, empty room evolved into a cultural metaphor for liminal dread, digital hyperreality, and the art of dislocation.

Summarized by ChatGPT on Jun 3, 2026 at 7:07 AM.