Coté

Mythos Firefox, the AI Job Fantasy, and Oregon Red Clover - Related to your interests, Saturday

Sat, 09 May 2026 08:09:32 +0200

Also: opinion shows for writers, no-update weeks, and the Spinoza heresies.

From: The Brautigan Library

Pushing Local Models With Focus And Polish - Building a DIY AI stack is difficult.
The “AI Job Apocalypse” Is a Complete Fantasy - “The macro story is not a jobless future, where we retire fat and complacent to our Netflix-scooters.” // Yes, but, There is a bit of “disruption for thee, but not for me” going on in pieces like this. No one wants to start their career over at the bottom of the wage ladder, especially if they’re later in their lives. Individuals lives are driven by the micro-economy, has their job been transformed and now they need to (1) learn something new, and, (2) start at the bottom of the career/pay level? // Gen-X kids lived threw off-shoring (neo-liberalism?), and we can tell you the result of “jobs don’t get lost, new jobs are created”: a bunch of unemployed people who can no longer afford their mortgage, their life as a whole. And then you get Tea Parties and MAGA. Good luck to us!
Related: the expectation of this kind of thinking is that people will do things like “buy lots of Nvidia, but if that doesn’t pay off make sure you are doing an MBA and planning a career in non-AI-implementation consulting.”
Behind the Scenes Hardening Firefox with Claude Mythos Preview - “Suddenly, the bugs are very good.”
It’s an opinion show - The value of having a column for the write…or videographer.
The stuff nobody tells you about startup marketing
Why do Oregon farms plant red clover every spring?
Poetry: I Too, Dislike It - I don’t know man, this gotcha headline piece was doing well until it included a poem and proved the headline it was trying to disprove. Multitudes and all that.

From: Astounding Science Fiction

Wastebook

“I think it’s probably a good idea to put a moratorium on installing new software for a week or so.” No Updates
“horrifying heresies.” Spinoza, Atheist.

Logoff

Current status: ¯_(ツ)_/¯ but with at least a smile. I think.

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.

Tinkerslop, Class Wartime, and Jobs Not AI Enough - Related to your interests, Friday

Fri, 08 May 2026 19:23:54 +0200

Also: VCF 9.1 prefers private cloud for AI, the McGroc analyst trap, and the bank that lost the pope’s account.

From: The undertow - Astrid

VMware Cloud Foundation 9.1 - a roundup of it all.
Broadcom Announces VMware Cloud Foundation 9.1 - “A preview of Broadcom’s Private Cloud Outlook 2026 report reveals private cloud continues to be the preferred platform for production AI. More than half of organizations surveyed (56%) are running or planning to run production inferencing in a private cloud. More importantly, public cloud use for production inference was 41%, down 15% year over year. Additionally, 62% of IT leaders reported being very or extremely concerned about generative AI infrastructure costs while 36% report AI is driving new requirements for data protection, privacy, security controls and risk management.”
🤖 Google Is A Full Stack AI Player, And Is Playing Well - Google Cloud’s full-stack AI integration from TPUs through Gemini is finally paying off, with Q1 2026 revenue up 63 percent and operating income tripling. The original “abstract the infrastructure” vision customers rejected a decade ago is now exactly what GenAI buyers want.
Anthropic and OpenAI are both launching joint ventures for enterprise AI services - “The overall logic of the two ventures is the same, raising money from alternative asset managers to create new channels for enterprise AI deals. The ventures will presumably get preferred sales access to their investors' portfolio companies, while the investors will capture more value from any resulting contracts.”
🤖 We did the math on AI’s energy footprint. Here’s the story you haven’t heard. - “Data center electricity demand was flat from 2005 to 2017 thanks to efficiency gains, but AI-specific hardware doubled consumption by 2023, and data centers now eat 4.4% of all US electricity. Lawrence Berkeley National Laboratory projects that by 2028, AI alone could consume as much electricity annually as 22% of US households.”
Do Americans really hate AI?
Apple to Pay $250 Million to Settle Class Action Over Delayed Siri Features - as someone else said, $25 in apologies for at $800 phone is weird.
The Organization Is the Bottleneck
Cloudflare to fire 1,100 staff whose jobs just aren’t AI enough - “To rub salt into the wounds of sacked staff, the email went out not long before Cloudflare announced quarterly results that included 34 percent year-over-year revenue growth and guidance for 30 percent future growth.”
IREN Announces Acquisition of Mirantis to Strengthen AI - Mirantis!
Is feedback really a gift? - “Have you ever gotten _actually useful _technical feedback? Ultimately all the feedback was really ‘write more code.'”
Datadog’s stock jumps 31% on crushing earnings beat, showing there’s still hope for software - There’s always money in the monitoring stand.
The McGroc Trap: Why Most Analyst Briefings Get Stuck Talking About Product - Influencer Relations - How to make an analyst briefing more interesting: have a unique view of “the market,” talk about a dramatic result at a customer, use silence to get the analyst engaged and talking. // “The vendor came with a point of view about where the market was heading, shared evidence from customer outcomes, and invited the analyst to challenge or build on those observations.”
🤖 Life During Class Wartime - Tim Bray argues we’re losing a class war to an emerging hereditary aristocracy and that a modest annual wealth tax is the obvious, achievable counter-move. Uses the Whitecaps sale to a billionaire heir as a vivid illustration of inherited-money power.
🤖 The 1960s Art School Experiment That Redefined Creativity - “…problem-finding draws on something more holistic than reason - a searching, uncertain state that may be where the creative process actually begins.”
🤖 Long ago, a Houston company’s art graced the walls of America’s stoners - “this is the visual record of a subculture whose members didn’t write books and whose work doesn’t hang in museums”

The map of system topologies

Wastebook

“Tinkerslop” - as they used to say “I feel seen.” Here.
“I want to carry my ereader, mesh radio or two, wallet, headphones, keys, pens, battery pack, and usually a manga.” Bag Watch.
“Need a whole lotta milk-ahhh.” Meanwhile, in the hall.
“The bank did not want to lose the account of the pope. They changed the number.” Here.
farrago - a confused mixture or a jumble of different things
“basically one guy controlling every knob.” Claude describes My Bloody Valentine.
“a managerial memento mori” Claude’s take on “It is not necessary to change. Survival is not mandatory.”
“Incidentally, GTA6 is coming out in November and apparently it cost $1 billion to make.” Here.
Asymmetric slop, via.
“Here’s one we built earlier.” Blue Peter, I’m told.
“When I became a magician at the age of 40, I took it very seriously, and it has transformed my life.” Alan Moore

John's Desk.

ICYMI

Laws, anecdotes, and other shit people say - an attempt to gather up all the “laws,” anecdotes, and folklore in tech and management talk.
Always taste the digital transformation while you’re making it - use iterative transformation.
History is not a story - applying a narrative to history is a problem.
Treat AI as a stoner
The Enterprise Dunbar number - Software Defined Talk #571 - “This week, we discuss AI labs driving cloud revenue, hyperscalers laying off instead of building, and kids defeating age verification. Plus, Brandon has too many thoughts on Workday.”

Logoff

From: Personal and prompt attention

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.

Always taste the digital transformation while you're making it

Fri, 08 May 2026 09:31:32 +0200

At a large enterprise I recently worked with, the board asked the Chief Digital Transformation Officer to develop an AI adoption strategy to drive innovation, growth, and cost efficiency. His consultant of choice conducted stakeholder interviews and proposed a three-phase program scheduled to last 3.5 years:

Phase 1: Fix digital basics and address the leftover gaps in people, processes, and technology from an incomplete digital transformation.

Phase 2: Build the AI foundation, including governance, tools, platforms, and an AI office.

Phase 3: Introduce AI and agentic business initiatives most likely to reach customers.

It was a coherent program that optimized for efficient delivery. But it’s not what the board wanted. They were looking for a strategy to become more innovative and competitive in an environment where the time required to implement IT products has shrunk significantly. What they got was a fix-the-basics project that would consume most of the budget before delivering actual business value.

From You Wanted to Become AI-Native, and All You Got Was a Lousy Foundation

That opening anecdote reminded me of most transformation initiatives I’ve either been involved in or observed: mysteriously, they seem to result in the organization doing what it’s already doing, just with new slides!

As Patzak goes on the argue, it’s better to have build up a theory of what to do, try out a little bit, see if you liked it, and adapt or keep going.

Put more plainly, you should taste food as you’re making it to ensure you get what you want at the end.

Sounds like a perfect example of Larman’s Law.

History is not a story

Fri, 08 May 2026 09:12:19 +0200

But how much was really done to unravel the Soviet Union by Solzhenitsyn’s three-volume narrative history of the Soviet Gulag between its publication and the end of Communism in the Soviet Union? Could the works of one author really dissolve a nation? Some might cite the election of the Polish Pope John Paul II in 1978 as making a greater difference. Others might point to U.S. President Ronald Reagan’s Strategic Defense Initiative (“Star Wars”) or to Soviet leader Mikhail Gorbachev’s policies of glasnost and perestroika.

Ultimately, the trouble with narrative history – which presents history as a chronologically coherent arc of characters, motives, turning points, and consequences – is that it simply can’t resolve these questions either way. There are too many forces operating on the trajectory of human affairs even to be enumerated. As a result, weighing them against one another is a fool’s errand.

From The Trouble With Narrative History.

Laws, anecdotes, and other shit people say

Thu, 07 May 2026 12:20:43 +0200

Betty Crocker, mid-1950s.

I heard a reference the Betty Crocker “add an egg” cake-mix story recently. It is: originally, people didn’t buy the cake mix because it felt too easy and didn’t feel loving, or at least sufficiently Calvinistic; General Mills removed the egg, then people cracked one in themselves, and suddenly they felt like they were “baking” and sales took off. This is also known as the IKEA effect, apparently

That got me thinking about all the other “laws” and anecdotes we cite. Conway, Goodhart, the Peter Principle, the Kahneman stuff, “culture eats strategy for breakfast,” “70% of transformations fail,” that thing Colin Powell said, and all the other known unknowns that Twain said.

So I asked the robot to help me start a catalog of shit like this that people say.

After a short time - well, long time - I realized that this list is incredibly long, so I stopped myself for now.

VMware Cloud Foundation 9.1 - private cloud, private AI, enterprise-grade kubernetes

Wed, 06 May 2026 14:08:58 +0200

Our cousins over in VMware announced the most recent version of VMware Cloud Foundation, 9.1, yesterday. We all call this “VCF.” It’s at the center of Broadcom’s strategy to be the private cloud stack for large enterprises. You know: banks, governments, large retailers, manufactures, et. al.

Our layer, the Tanzu Platform, sits a-top VCF like any PaaS would sit a-top IaaS.

Recently, the VCF people have been putting a lot of effort into private cloud AI. You can think of that as “local AI” if you prefer. From the last three surveys they’ve done, you can see a steady interest in running private AI:

Broadcom commissioned surveys from IDC (2024) and Radius Tech (2025, 2026).

Kubernetes, memory, and AI

If you’re a full private cloud stack, there’s a lot going on each release, so here’s three interesting things:

VKS now scales to 500 Kubernetes clusters per Supervisor. That’s a platform-of-platforms number. If you’re running a multi-tenant Kubernetes service, 500 clusters per supervisor is meaningful for compliance boundaries, and AI workload placement. Real enterprise-grade stuff. Also, for some platform-y related things like packaging and deploying containers, managing them, etc. see this post.
Memory tiering and storage dedup go after the DRAM problem head-on. Enhanced NVMe memory tiering keeps hot pages in DRAM and pushes cold pages to NVMe, with vSAN global deduplication and stronger compression behind it. They say this is up to 40% lower TCO.
Native MCP support, with governance, in VCF Private AI Services. VCF 9.1 ships Model Context Protocol support with pre-built, governed connectors to Oracle, Microsoft SQL Server, ServiceNow, GitHub, Slack, Postgres, and the rest. That matters because it’s the difference between “we host an LLM” and “the LLM can actually do work against our internal systems.” There’s also support for Google Docs, metrics, and more.

More Details

Here’s a round-up of posts from VCF:

Broadcom Announces VMware Cloud Foundation 9.1, Enabling Secure and Cost-Effective Infrastructure for Production AI - press release, 2026-05-05.
VCF 9.1: The Secure, Cost-Effective Private Cloud Platform for Production AI - top-line framing post.
Announcing VCF 9.1: Modern Private Cloud Built for Efficiency and Resilience - the formal launch post; efficiency, app delivery, cyber resilience as the three pillars.
Streamline, Simplify and Protect all your AI workloads with VCF 9.1 - AI workload management and protection.
AI with VCF 9.1 on AMD GPUs: Build with open frameworks and simplify management, at a lower TCO - AMD GPU support alongside NVIDIA.
How Broadcom Is Helping Enterprises Win the AI Security Sprint - frontier-model research on AI-accelerated vulnerability discovery; pairs with the 9.1 security messaging.
Modernizing Infrastructure Economics with VMware vSphere Foundation 9.1 - vSphere Foundation 9.1 against DRAM volatility and the “maintenance tax”.
Scale Smarter, Save More: Redefining Infrastructure Economics with VMware vSphere in VCF 9.1 - the TCO / hardware-scarcity cut.
Deploy Modern Apps Faster, Scale Smarter, and Lower Your TCO with VMware vSphere Kubernetes Service in VCF 9.1 - VKS as the on-VCF Kubernetes story; up to 500 clusters per Supervisor.
Simplify Workload Connectivity and Enhance Network Scale and Performance with VCF 9.1 - VPC model evolution and scale.
Announcing VMware Cloud Foundation Edge 9.1: A Scalable, Autonomous Edge Platform - hundreds-to-thousands of distributed sites under central control.
Accelerate, Streamline, and Control Your Self-Service Private Cloud with VMware Cloud Foundation 9.1 - VCF Automation, with stats from a March 2026 customer survey on VCF 9.0.

Treat AI as a stoner

Wed, 06 May 2026 10:59:07 +0200

The right mental model for working with an AI, according to my co-host David.

If you’ve spent any real time with an AI, you know exactly what he means. The model can do impressive work in a tight scope. Step out of that scope, or feed it more than fits, and you’re suddenly explaining the same constraint for the third time that happened just a few minutes ago.

In the most recent Tanzu Catsup episode, we also talk about copy.fail, CVEs, SBOMs, and more.

Check out the episode. We stream live every Friday at 10am US Eastern time / 4pm Amsterdam time. Check out past episodes, and put it in your cal to watch the next one.

Parasocial Media, the War on Adobe, and Garbage Chairs - Related to your interests, Wednesday

Wed, 06 May 2026 10:22:55 +0200

Also: Bob picks the model, AI-BOMs, $37.5M DIY platforms, and AI slop on tap.

From: AI Slop Me

🤖 The Venture-Capital Populist - Packer’s profile of David Sacks as architect of Silicon Valley’s MAGA alliance, who delivered crypto legitimisation and AI deregulation to the Trump administration while keeping his venture fund running and benefiting from the policies he wrote.
don’t call it “social media,” call it “parasocial media”
Most AI coding is “like taking your Ferrari to buy milk”: IBM’s Neel Sundaresan - Harness talk from IBM: “Bob doesn’t expose the underlying model to users. It routes tasks automatically - to Anthropic Claude, Mistral open-source models, IBM Granite, or one of several proprietary, fine-tuned models built specifically for Bob’s environment - based on what the task actually requires. That routing intelligence is where Sundaresan thinks the real architectural work is. ‘It’s not slapping on a model into the system,’ he says. ‘It is bringing the model, bringing the experience, but also bringing the architecture that provides a great experience. All three have to come together. The model is only one part of the equation.'”
AI-BOMs replace SBOMs as way to track AI agents and bots - Yup…
How can engineering leaders calculate the return on their AI investments? - You have to learn how to use a new technology before you can get the benefits, and learning looks like failure because you’re worker slower.
Notes on SKILL.md vs MCP - With skills, you’re relying on the LLM to follow your prompt begging.
What you’re actually writing when you write a SKILL.md - “Skills are loader specifications, not prompts.”
Lessons on Building MCP Servers
Writing is thinking, presenting is deciding - “[S]imilarly, the other day I asked someone at work to write me a note about a project we were working on. And he did. He came back to me (much too quickly) with a note that had been written by AI. And, obviously, that was my fault. Because really what I meant by ‘write me a note’ was ‘think about this problem, reflect on it for a while and then write down your thoughts, so you can transfer your thoughts to my head and we can think about it together.'”
🤖 State Department Looks to Build on Online Passport Renewal Success - State’s online passport renewal hit 7.3M users and 94% approval by switching from waterfall to agile and replacing only the front end. Now they want to expand to first-time applicants and digital credentials, but the team that built it has been gutted by layoffs and the shutdown of 18F. // They switched the UI development to agile and user-centric design…including the end users in the cycle to figure out how to best design the UI.
🤖 War on Adobe - Roundup of commentary arguing Adobe’s subscription-era strategy has invited a coordinated attack from free rivals like Canva-owned Cavalry/Affinity and DaVinci Resolve, while in-app upsells and bloat erode trust with the professionals who built the company.
🤖 Aura through the Cult of Self: opportunity creates authenticity - Art as affirmation therapy: ‘Under this “Cult of Self,” art’s authenticity is judged not by the artist or the medium, but by how well it lets the audience express and actualize themselves. Spotify Wrapped resonates because it validates listeners’ self-image; the art’s aura builds the audience’s aura.’
Failover Weekly: Ask the Hard Question - “The best podcast interviews this week had one thing in common: the host didn’t let the guest off easy.”

From: Recode Project

Wastebook

Now that conference organizations are on-edge about AI-written abstracts, I bet it’s actually better to be unprofessional in your abstract, less than perfect, and put a few grammar and spelling errors in.
“an infinite scroll of cotton candy content,” Here.
“lot of ‘modernization’ is really just moving failure modes around” Here.
“A lesson is something taught by someone.” Here.
“The future is full of robots that don’t do much yet” Here.

From: Elfmaids & Octopi

ICYMI

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.

don't call it "social media," call it "parasocial media"

Wed, 06 May 2026 09:55:11 +0200

In 2006, most people who logged into the large platforms posted content because they were co-constructing sociable spaces to enjoy the companionship of others. In 2026, posting has waned (John, 2024); most social media users prioritize scrolling ‘amateur’ content rather than posting their own haphazard updates for friends. The quality of the media on social media has become more strategically constructed, more intentionally curated, and more professional. Users are now lucky to see personal content that their friends are posting amid the slick content created by the advertisers and strategic creators who dominate most people’s feeds. What goes ‘viral’ is now often manufactured in a lab, designed to appeal (Hund, 2023; Mears, 2023). Given the trajectories of the various platforms, this makes sense. In a world of algorithms, highly polished content is rewarded. And: Parasociality plays a more central role in today’s platforms than it once did. But parasocial relationships are a type of trickster. Attending to parasocial connections may be pleasurable for consumers, but doing so does not strengthen the collective social fabric. It is possible to experience loneliness despite spending hours emotionally engaged with others' dramas if those interactions are not reciprocated. Even those who are producing content for the parasocial world struggle to navigate the contorted forms of intimacy that abound (Glatt, 2024). Friendship requires reciprocity and compassion. Parasocial media creates the conditions for people to objectify one another at a distance as mediatized objects, helping realize the different layers of toxicity that social media scholars document (Bailey, 2022; Banet-Weiser & Miltner, 2016; Suarez Estrada et al., 2022; Wong et al., 2025). So when people opt to devote their energy to tracking the latest TikTok star or scrolling content instead of nurturing interpersonal relationships, they are effectively amusing themselves to death.

Finally:

I think that it’s time that we deprecate the label ‘social media’ and begin to recognize that we are dealing with an era of ‘parasocial media.'"

From: “Social Media Is Now Parasocial Media2026,", danah boyd, April, 2026.

DIY platforms: $37.5 million, 60 people, and you're still not done

Tue, 05 May 2026 13:12:39 +0200

What it actually costs to build your own internal developer platform over five years, and why most “we’ll just build it” pitches are pricing the cheap version.

Sixty people and $7.5 million a year, every year, is what it costs to build and operate your own internal developer platform if you do it the way DIY-minded enterprises end up doing it, and most of the executives signing off don’t see the number, because it’s spread across eight cost centers under “engineering.” The cheap version of the pitch they hear - one team, weekend work, a control plane with a bit of YAML on top - ships v1 and turns into the maintenance burden nobody wants to own. The DIY platform version that actually works is a sixty-person product engineering organization, indefinitely. The numbers below come from a recent Tanzu paper I helped update, so flavor it with a vendor-salt up front if you feel like you need to. But, the staffing structure is concrete enough that anyone who’s run a platform team can pressure-test it.

The weekend project that turned into a 60 person infinite financial commitment

From what I’ve seen, if you align your platform team with something that feels like the CNCF platform reference architecture, you end up with about seven product teams: infrastructure, operations, deployment, runtime and middleware, database, security, and coaching/developer enablement. Each team is one to two pizzas, 4-9 engineers. There’s no personal pan pizzas when it comes to DIY platforms. Share a couple of scrum masters and product owners across the seven. That’s about 60 people total.

At a fully loaded $125k/year per engineer (mid-market US, not Bay Area), you’re at $7.5 million in payroll, every year, indefinitely.

	Year 1	Year 2	Year 3	Year 4	Year 5
Annual	$7.5M	$7.5M	$7.5M	$7.5M	$7.5M
Cumulative	$7.5M	$15M	$22.5M	$30M	$37.5M

Five years in, you’ve spent $37.5 million on payroll alone for a platform team that, in the success case, is mostly running the thing they built rather than building new capabilities or closing the gap with the vendors who are still shipping.

What “buy” actually staffs to

Enterprises that bought a commercial platform staff to operate it at very different ratios. Some real numbers from public talks and customer conversations¹:

30,000 developers supported by 50 platform engineers
6,500 developers supported by 16 platform engineers
1,200 developers supported by 6 platform engineers
350 apps supported by 7 platform engineers

Six to ten engineers covering the same population that, in the DIY case, takes sixty. That sixty is the staffing shape of a real DIY effort because you’re not just running a platform, you’re building it: APIs, integrations, security pipelines, dashboards, upgrade tooling, the next AI capability your developers want, the next compliance regime your auditors want.

In conversations I’ve had with DIY platform organizations over the years, there’s a shadow platform engineering “group.” Each development group tends to have at least one person, if not more, who do the glue work between the DIY platform and what the applications teams do. This is often around build and pipeline integration, security, or just figuring out how to deploy apps and hook-up to services. This staffing cost is completely unaccounted for in the business case spreadsheets.

Why this gets ignored

The visible cost line for a commercial platform is one license number on one PO. The DIY headcount is spread across eight cost centers, charged to “engineering,” and looks like normal hiring. The license number is right there on the page; the DIY total has to be calculated, and nobody runs =SUM on all those people.

Works as designed

Engineers also have a story they tell themselves about building their own platform. Nobody gets promoted for choosing a vendor - the era of not getting fired for choosing IBM was long ago. People get promoted for shipping platforms, and “we shipped a platform built on Kubernetes” is a better promotion-packet line than “we onboarded everyone to a thing we paid for.” This isn’t some moral failing, it’s how engineering organizations and the HR policies that pay them work. The pattern is called résumé-driven development, and it’s well documented:

Extensive RDD-based technology selection may therefore lead to complex or even unmaintainable software consisting of technologies which are not suitable for the requirements, which are unfamiliar to current or future employees, or which did not deliver on their promise and were discontinued. –“Résumé-Driven Development: A Definition and Empirical Characterization,” Jonas Fritzsch, Marvin Wyrich, Justus Bogner, Stefan Wagner, January 2021.

Left behind

The most damaging factor is the velocity gap. When you build your own, the vendors keep shipping; you’re shipping too, but against a moving target. The Tanzu paper puts the gap at 12-18 months within the first couple of years, and the gap widens rather than closes, because the vendor’s team is bigger than yours and their roadmap is funded by every customer they have, not just your one company. The year-three meeting where someone says “we need to add the AI services that Vendor X shipped last quarter” is the meeting where the DIY ROI quietly evaporates.

For the strategy-nerds: comparative advantage, with a 200-year-old assist

The notion of outsourcing non-core infrastructure is older than computers. In 1817, David Ricardo argued that even if you can do something well, you should focus on what you do best and trade for the rest. Portugal could grow grapes and weave cloth, but it specialized in wine and traded for British cloth because that’s where the marginal return was.

If you’re not into early 19th century economics thought leadership, Simon Wardley updated the thought leadership with his much beloved Wardley maps: anything that’s already become a commodity should be bought rather than built, so that your scarce engineering capacity is spent on the parts of the business that actually differentiate you from the competition.

Abby Bangser put this in plain English at KubeCon NA 2025:

It’s not about rebuilding what we can purchase that is available on the market. It’s about making sure we spend our time building the things that are bespoke and important for our organization.

For most organizations, the platform itself isn’t where you differentiate. Your apps are! The platform is plumbing - necessary, valuable plumbing, but plumbing. A useful test: would your customers pick you over a competitor because of your platform? In Dutch grocery, the answer at Albert Heijn and Jumbo is “no” - shoppers care about prices, the app, the checkout, not the orchestrator behind it. At Picnic, the answer is “yes,” but only for one piece - the custom narrow delivery trucks that make their last-mile economics work. Everything else, including the platform, they buy. Knowing which kind of business you are answers the build-vs-buy question for you.

Letting your best engineers spend their careers writing YAML, integrating CI to your registry, and tracking CVEs in container base images is trading down. You’re spending one of the most expensive resources you have, engineering capacity, on something a vendor will take care of for you. The opportunity cost is what you didn’t build with that capacity: the apps that would actually move the business.

What to do about it

Here’s three practical moves for anyone staring at this decision now:

Calculate the all-in DIY headcount before the meeting. Sixty is the high end; thirty is the low end if you cut corners. Either way, multiply by your fully loaded engineering cost and put the number on the slide. The license-vs-engineers comparison ends most of these debates by itself. Don’t forget to multiply this out in one, three, five, and more years. History shows that the platform will be with you for a long time.
Be honest about year three. Not v1. Maintenance, integration, and a steady migration debt as your developers ask for things commercial vendors shipped two quarters ago. Look at how frequently AI has been changing and improving over the past two years. Can your DIY platform team adapt that fast to make sure you’re getting those tasty, new AI capabilities?
Build the developer-facing layer, buy the rest. Portals, golden paths, integrations into your specific business systems are differentiated, so build them. Infrastructure, observability backplane, secrets, certificate management, container registry are commodity, so buy them. The org-design version of the argument is in Platform Engineers Don’t Have Time for Infrastructure; the cycle it fits into is in The Eternal Recurrence of DevOps, and that’s not great for long-term business investment ROI.

Every enterprise that builds its own internal platform is, whether the executives say so or not, choosing to spin up a small platform vendor inside the company. The economics of running a platform vendor are well known and largely punishing. The companies that pick this path should at least know that’s the choice they’re making. Most don’t, because the choice was never made out loud. It was made by inertia, in the meeting where someone said “we’ll just build it ourselves this weekend.”

You should avoid the DIY mess and instead TryTanzu.ai. Or, more bluntly: stop building platforms. Start building apps.

If you’d rather have all of this as a talk, here’s the full version going through all seven pitfalls. Bite-sized videos for each pitfall on cote.io/diy.

For more dev to ops rations, see slide 21 of 7 Ways to Fail at Building a Platform, with sourcing per the slide: Kroger, GAIC, and Mercedes-Benz figures from customer conversations; “Enterprise Grade Platform Engineering at Charles Schwab," Coté, September 2024 (the panel data behind the post is Schwab’s Explore 2024 session, TANP2089LV); Rabobank conversations at CF Day EU 2025; “3 Cloud Foundry Stories," Coté, CF Day EU, October 2025. The 1,200/6 Mercedes-Benz ratio also appears as a Cloud Foundry customer average in The New Stack’s “Open Source Platform Engineering: A Decade of Cloud Foundry”. ↩︎

Tue, 05 May 2026 08:40:59 +0200

Surfing taco.

Tue, 05 May 2026 08:35:45 +0200

Philips lightbulb packaging, found in Diemen.

Tue, 05 May 2026 08:31:50 +0200

Garbage Chairs of Amsterdam.

Never Talk About Goblins, McDonald's AI Teammate, and the Slop Defense - Related to your interests, Friday

Sat, 02 May 2026 08:43:20 +0200

Also: human-in-the-loop theatre, MCP-on-Go pain, and Google Translate at 20.

From: The Spanish napkin

Where AI Accountability Goes To Die When You Stick A ‘Human In The Loop’ And Call It ‘Governance’ - 🤖: “Stuart Lauchlan and HFS / Altimetrik’s Humans at the helm of AI study of 505 G2000 execs find enterprise AI is cost-driven, accountability-asymmetric (CIOs carry the blame, not the strategy), and the ‘human in the loop’ is mostly theatre - only 25% say humans would prevail in a disagreement, 18% have clear visibility into AI reasoning.”
What Happened When We Treated AI Like an Engineering Teammate
Everyone’s an Engineer Now - “Anthropic engineers are producing roughly 200% more code than they were a year ago, Cat noted. Today the main constraint is reviewing all that code to ensure it’s production-ready.”
The Impact of AI-Generated Text on the Internet - “We find that by mid-2025, roughly 35% of newly published websites were classified as AI-generated or AI-assisted, up from zero before ChatGPT’s launch in late 2022.” // Also, chart on negative AI sentiment.
The Slop Problem Isn’t What You Think - “A lot of AI writing is terrible. Soulless, repetitive, generic. But so is a lot of human writing. I’ve sat through vendor whitepapers with seven listed authors that would embarrass a school newspaper. We’re not actually angry about quality. We’re angry about dues. Did you suffer enough? Did you earn it?” // Good content is good content, and bad things are bad.
Building a production MCP server in Go - ”The discovery endpoints above are straightforward to serve. The hard part is making the actual OAuth flow work end-to-end with your existing auth system.” // every time I try to make an MCP server, this is the problem.
Cloudflare Announces Agent Memory, a Managed Persistent Memory Service for AI Agents - AI as a Service? Well, at least components of an AI stack as a service. Can you share it between multiple people? A shared context would be cool.
Today I shipped 20 apps and a screensaver
How To Blog - The Hemingway approach: “TLDR: Honesty is key. Write what you know and care about. What what you really think. Write original by being, doing, and thinking original.” // OP actually has much more.
Failover Weekly: Ask the Hard Question - ”The best podcast interviews this week had one thing in common: the host didn’t let the guest off easy.”
20 fun facts to celebrate Google Translate turning 20 - Along with Maps, one of the best applications out there.
Why are the Artemis II photos on Flickr?
🤖 Acetaminophen Quietly Beats Ibuprofen on Safety, and Drug Labels Won’t Tell You Why - Despite its narrow lethal-dose window, acetaminophen is generally safer than ibuprofen for most people because ibuprofen’s COX inhibition systemically damages stomach, heart, and kidneys. The FDA’s per-drug mandate and liability fears around comparative advice keep this fact off official labels.

From: Dr. Drang

Wastebook

“[N]ever talk about goblins, gremlins, raccoons, trolls, ogres, pigeons, or other animals or creatures unless it is absolutely and unambiguously relevant to the user’s query.” OpenAI’s Codex.
“day-date” Seroter.
“Matt Ray, but verify” SDT 570.
“Here’s one we built earlier.” Blue Peter.

ICYMI

90 minutes of hell is other people
The Enterprisification of Agents - Software Defined Talk #570 - “Google’s enterprise agent stack, what makes an agent an agent, and AI’s biggest critic. Plus, Coté wants better vision.”

Logoff

Nothing to report, really.

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.

90 minutes of hell is other people

Thu, 30 Apr 2026 08:10:44 +0200

This feels like me:

NEWMARK: I guess, ideally, I would have more social skills—meaning, some.

COWEN: We’re simulating social skills just fine here.

NEWMARK: That’s the phrase I use. At least on my part, what looks like social skills is just fakery. I can do it for short amounts of time, maybe 90 minutes. I’ve given up, though, on actually accumulating social skills, getting better at it. More to the point, I try to get into positions where other people can show social skills.

I tell people that I don’t like talking with others, especially “new people” (read: strangers) and they’re shocked. I seem so personable and chatty!

Well, I’ve been training myself in that 90 minutes for almost quarter of a decade. Don’t mistake a well honed skill with ease, nor desire to use it.

Pre-Idea Funding, the Stolen-Valor Chore Coat, and Doom-Scrolling Pre-GitHub - Related to your interests, Wednesday

Wed, 29 Apr 2026 16:02:50 +0200

Also: AI’s biggest critic loses the plot, Notepad++ on Mac after 20 years, and dark modes as cosplay.

From: Monkigras, 2026

From Infrastructure to Agents: A Hands-On Guide to Secure Private AI with Broadcom
Claude Code and the rise of personal software - Low-code, except high code made by non-programmers.
“Developer loyalty is at zero right now”: Google doesn’t care which AI coding tool you use - “I come across as a wildcat in this one.” // He should wildcat it more!
AI’s biggest critic has lost the plot - “we desperately need better skepticism.” // He’s no a fan of Ed Zitron.
The Real AI ROI Problem Isn’t Technology - It’s Measurement - Forrester takes a crack at an enterprise AI business case and ROI model.
Before GitHub - “GitHub is currently losing some of what made it feel inevitable. Maybe that’s just the life and death of large centralized platforms: they always disappoint eventually. Right now people are tired of the instability, the product churn, the Copilot AI noise, the unclear leadership, and the feeling that the platform is no longer primarily designed for the community that made it valuable.”
AWS Summit London 2026: From Cloud Services to Platform Value - “platform teams are deciding between build, buy, and blend, and the answer is becoming clearer: Blend by buying the undifferentiated heavy lifting, build the experience, workflows, and capabilities that are unique to your organisation. That is where platform engineering delivers real impact.”
Notepad++ Code Editor Comes to Mac After 20-Year Wait - Hopefully better than TextEdit.app.
Almost 90% of women leave tech industry within 10 years
Tactical dark modes - Where “dark mode” came from and theories of what it persists, despite not really being functionally needed. Also, cool old UI pictures.
Who are Europe’s immigration worriers? - I think the implication is: people are not anti-immigrant because of income level: they are anti-immigrant because they don’t like the people.

From: The Real AI ROI Problem Isn't Technology

Wastebook

“[E]very chore coat might be stolen valor, sure, but clicking away at a creative agency while looking like a Parisian waiter is very different than working for a major military contractor and sporting its logo on your breast pocket.” Here.
“I’ve been doom scrolling GitHub issues since before that was a word” That Mitchell.
“It’s very easy to take not working for granted, especially when feeling down or stressed, and the days slip by unloved and unappreciated.” Scythin' Phil.
“[P]re-idea funding” Stanford.
“It’s a closed loop, and it generates very little compound interest.” Westenberg.

From: Santiago Jimenez

ICYMI

Hard pass with a 20% registration discount - where-in I complain that my conference talks keep getting rejected.
Hardened Runtimes, the CEO Job, and Raising as an All-Woman Founding Team - Software Defined Interviews #122 - “Whitney and Coté talk with Emily Long, CEO and co-founder of Edera, about building a hardened container runtime that secures infrastructure foundations instead of chasing detect-and-respond alerts.”

Logoff

I don’t know about you, but I have a fondness for VHS tape covers:

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.

Wed, 29 Apr 2026 16:00:00 +0200

Garbage Chairs of Amsterdam, near ARTIS, 2026

Software Defined Talk #122: Emily Long, VCs, startups, Edera, etc.

Wed, 29 Apr 2026 12:02:05 +0200

In this episode, Whitney and I talked with Emily Long, CEO and co-founder of Edera about their hardened container runtime - the kind you can swap in without re-platforming or kicking off yet another zero-trust migration. The pitch is basically: stop chasing detect-and-respond alerts and just fix the foundation.

We also got into the messy reality of going from COO to CEO (the COO title is, it turns out, kind of made up) and what it’s like raising a deep-tech Series A as an all-woman founding team - the downside-vs.-upside question pattern VCs fall into, and the now-classic “I just Googled Kubernetes and I know more than you do” pitch moment.

The episode opens with a long detour on to-do lists, Claude Code, and whether AI tooling is just expanding the list of things you feel obligated to do. Hand-written lists make a comeback.

Listen to the episode.

Hard pass with a 20% registration discount

Wed, 29 Apr 2026 11:15:30 +0200

The author at SCALE, March 2023.

I've been having a terrible time getting talks accepted at conferences. Talking at conferences has, you know, been a huge part of my life for about 20 years, so this is kind of a bummer. Care to join me in some self-~~pity-partying~~care?

I've got the wrong haircut

What gives with my high rejection rate? My theories are:

I don't have enough new material. Who wants to hear my stories about platform engineering, stealth complaining about Kubernetes ruining the progress we made in PaaS, private cloud, or playing D&D with AIs?
People don't like abstracts written in that ironic Gen-X way like they did in the 2010's. Maybe the conference pickers have gotten more genuine and optimistic as us olds move on.¹ I can't help but write ironically, though, a joke only to myself in every title and third sentence - my fingers are wired to it. Perhaps I need to be less Letterman, more Ted Lasso.
It seems my abstracts are 100% AI written. If #2 is true, it might actually be a good idea for me to ask the AI to convert my abstracts into whatever tone the CFP-pickers vibe on nowadays.
It could just be that people look at me and say "no thank you." I can see that I throw off strong "hidden vendor pitch" vibes.

Who knows.

The eternal reoccurrence of everything

2026 is going to make 2022 look like 2015.

I find it hard to come up with new topics.

I presented on platform engineering, as it's called now, for about a decade. There isn't much left to say except don't build it yourself and product manage it. Stay off the tinker path to ROI failure.²

You can present on AI, but by the time you give the talk six months after submitting it, everything will have changed dramatically. Think back six months ago, November, 2025. That was right as the Claude Code is God/Harness wave we're currently in started. OpenClaw was Christmas of 2025. MCP was at its height of popularity back then, so you would have written a CFP about that. There's probably tons of MCP talks going on this Spring and, like, that's kind of a solved problem and concept now.

Seriously: at this point, if you're running a conference with AI talks, you should just be accepting abstracts that say "I'll demo and talk about whatever is cool in AI in six to ten months."

Plus, on AI, what is there left to present on? It just works amazingly well. Now, we just need to wrap all that enterprise-y stuff around AI, and we're good to go.

Oh, and I should clarify: despite my generational mind-warping, I'm not being ironic here.

Devoxx 2010, when explaining the cloud stack was cool.

meatsack.pptx

This leaves "people" stuff and career advice. However, that doesn't really fit into my professional remit. (Ahem...see the hidden vendor pitch problem above.)

The most recent talk I got accepted was an updated version of my surviving and thriving in BigCo's talk, which I'm thrilled to be doing! I only submitted it, though, because it's an easy train ride to Utrecht. That kind of talk isn't justifiable as a business expensed one.

So, what's left is for me to try to shift out of meat-ware (platform engineering theory, process, and "strategy and management") back into software.

SREDay Amsterdam, November, 2025.

This whole talk was written by AI

Thanks to Claude Code, this is actually possible. For example, I have a lot to say about using Cloud Foundry as the bash/shell tool, over SSH. You get all the enterprise-y stuff!

What's, well, "fun" about this kind of AI assistance is that it makes you focus on the craft of presenting and demoing. The technology is easy and possible...and sort of irrelevant. How did Claude Code implement this running a shell thing in Cloud Foundry? I mean, it did a good job...but who cares? It works, and you can change it around at will.

How you present this and connect it to "real world" problems and fixes, hopes and dreams, is the more interesting part. And that's what's fun about crafting and giving a presentation: the creation of an event, a happening, an experience both for the audience and the presenter.

We'll see.

Mood.

Have you noticed that the kids these days all have curly mullets and mustaches? I live in a sleepy Amsterdam suburb, and I see at least one John Oats a day. Go into the city, and it's like there's some kind of Hall and Oats tribute band festival going on. Back in my day(!), why, sure, you could mullets galore, but everyone knew you were doing a bit. But, I guess, you know, when it comes to aesthetics, sure: I make no moral judgements. Wear your hair like you dig, Oats-ites. ↩
Despite my sentiment, this seems like the prime time for platform engineering talks. Abby, for example, gives great ones. There's a whole platform engineering track at KubeCon! Perhaps this is more a "me" problem: there seems to be plenty of hunger for platform engineering talks, which scratches the head more here. ↩

The Jevons Audit Dividend, Spring AI in Seat 3A, and the Pelican-Astronaut-Horse - Related to your interests, Tuesday Afternoon

Tue, 28 Apr 2026 17:29:32 +0200

Also: AGENTS.md as model upgrade, Europe-as-vassal, and a Slack chatbot nobody bothers to verify

From: WHY ARE YOU LIKE THIS

What We Learned Testing Frontier AI Security Models Against Our Own Code - “What we learned is that the useful mental paradigm for this technology is not to treat it like a vulnerability ‘scanner.’ The models are non-deterministic; they do not work like static analyzers. They work like a security engineer you direct and support - or like a threat actor, if you are on the receiving end. Skilled human researchers who hunt zero-days use a range of tools, reason carefully about their target, and iterate on hypotheses. The models should be given the same working environment.”
Alexandros Pappas on LinkedIn - Lufthansa using Spring AI.
The Jevons Paradox and Insatiable Humans: Why AI Won’t Empty the Finance Suite by Eldar Maksymov - 🤖: “Maksymov’s prescriptive close is the part most worth memorizing. Students: stop competing with machines at machine work - build judgment, communication and the ability to ask questions nobody has asked. Firm leaders: stop using AI exclusively as a headcount-reduction tool, because that captures displacement and forfeits the Jevons dividend; the firms that thrive will be the ones that ask ‘what could we audit, analyze or assure that we have never been able to afford before?’ In the spreadsheet era the winners were not the firms that fired their accountants - they were the ones that hired financial analysts.”
A good AGENTS.md is a model upgrade. A bad one is worse than no docs at all.
‘How Europe Regulated Itself Into American Vassalage’ - “Yet it is worth considering, nevertheless, that too much regulatory oversight has hampered Europe’s ability to compete with the United States. This is a possibility. I find it telling, however, that Pignal cannot cite a specific example of this.”

From: How I Make Work Lately - by Justin Parpan

Wastebook

“[T]he story of a boy from Yorkshire who overcame…”
“I pulled over even though I didn’t need to.” So smoooove.
“I know people joke about everyone in Ireland knowing everyone else, but I just got off the plane in Cork and the customs officer is my cousin.” Here.
“A team lead adds an AI chatbot to a Slack channel. Anyone can tag the bot to answer questions about the company’s products. Coworkers tag the chatbot many times a day. You never see someone check that the bot’s responses are correct.” Existential crisis a la AI.

From: Separate Writing and Formatting - iA

Logoff

See you next time!

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.

Software Brain, the $5 Postcard Warship Hack, and Cyberpunk Dystopia - Related to your interests, Tuesday Morning

Tue, 28 Apr 2026 08:12:09 +0200

Also: AI vendor lock-in, half of GenAI projects fail, tinkers vs. endpoints, and a blogroll worth keeping

MCP vs. APIs: Why You Need Both for AI Applications
Beware Software Brain - 🤖: “The failure mode is exporting that logic to human life. The ask is no longer that computers adapt to people but that people make themselves legible to the machine: open your files, email, calendar, and messages so the AI becomes more valuable. A decade of smart-home flops already showed regular people don’t want this; AI is not going to reverse that.”
Use Claude Cowork with third-party platforms - Swap in your own LLM for the Cowork harness.
Put Your Coding Agents in Your Pocket - How one person accesses their rig remotely.
Indian med student rakes in thousands with AI-generated MAGA hottie - A study in revealed preferences of all types, not least of which “slop.”
Why Half of GenAI Projects Fail: Avoid These 5 Common Mistakes - Well, you know…this is sort like the essence of every enterprise AI presentation you’re going to see for the next 24 months. Keep the EBC fires warm, my deck-driving homies.
Execs fear job loss over AI adoption failures
The Space Between Humans, AI, and the Work We’ve Been Avoiding - “The gap between what AI can do and what organizations are actually able to absorb isn’t a technical gap. It’s a human one - the communication that didn’t happen, the missing information, the assumptions, and the trust that was skipped in service of speed.”
Agentic AI makes old worries about lock-in look quaint - “However, that means that once enterprise buyers select a platform to manage their agents they’re likely to be stuck there for quite some time. And they’re being forced to make that decision at a time when no one knows exactly where these tools and platforms will be in a year, and when Anthropic and OpenAI are making a big push to convince companies to work directly with them.”
I Just Wanted Endpoints - Here we go again: “Their CTO said something that hit home: he doesn’t want his AI engineers understanding Kubernetes. He wants managed infrastructure that works. He wants to focus on product and customers, not container runtimes.” // The tinkers will drag us through another round of platform rebuilding.
Intel shutters open-source evangelism program and archives key community projects - closures point to significant shift in open-source leadership - “Intel has quietly wound down its Open Ecosystem Community and Evangelism initiative, archiving the project alongside a fresh wave of open-source repositories on GitHub.”
Recency bias (non-derogatory)
Bluetooth tracker hidden in a postcard and mailed to a warship exposed its location - $5 gadget put a $585 million Dutch ship at risk for 24 hours - Send a postcard to a sailor with an AirTag like tracking device, then track where the battleship goes.
What the death of Direct File tells us about state capacity - 🤖 Don Moynihan argues Direct File’s cancellation reveals the erosion of US public interest capacity, not just a lost tax tool. Private lobbying plus an administration hostile to public goods killed a working in-house product, and the replacement - a revived Free File partnership - repeats a failure Intuit already exploited.
Pro-level travel tips
Beeswacks - SimpleBits® - Very nice!
The Art of Dystopia / Part 6: Cyberpunk - So many great pictures.
Blogroll Keepers #11 - Good list.

Wastebook

“[P]rofound geopolitical implications” - Cal Newport

ICYMI

Tanzu Platform 10.4 Source Coverage - a round up of our platform’s recent release, 10.4. This is the one that adds the foundation for AI use in large organizations, centered around enterprise control and security, and for private cloud.
Still life in Amsterdam suburbs
Agent Assimilation - Software Defined Talk #569 - “This week, we discuss agents taking over at Google Cloud Next, Apple’s new CEO, and Cursor getting acquired (sort of). Plus, Coté’s e-waste has no exit strategy.”

Logoff

See you next time!

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.

Tanzu Platform 10.4 Source Coverage

Wed, 22 Apr 2026 16:31:17 +0200

Here’s coverage of our recent Tanzu Platform 10.4 announcements. This is a big agentic AI release, bringing in to Tanzu Platform all sorts of features to secure, standardize, and otherwise make agentic AI work more enterprise-y. My think on 10.4 is listed below too. There’s also the short video I made on the 10.4:

Tanzu Platform 10.4 Clippings

"Introducing Tanzu Platform 10.4: Extending Platform as a Service to Agentic Applications," Darin Zook, Tanzu blog, April 15th, 2026. The headline launch post, framing 10.4 as “the industry’s first pre-engineered PaaS for agents on private cloud." Introduces the agent-operator agreement, the stop-building-the-platform pitch, and the VCF 9 integration story. Calls the platform “an industrial-grade AI sandbox” that turns “the black box of AI into a transparent audit ledger.”
"Broadcom Announces Tanzu Platform Agent Foundations,” Broadcom, Broadcom News, April 15th, 2026. Official press release from the AI in Finance Summit. Carries Tanzu GM Purnima’s quote about partnering with customers to “move your agentic ideas into production today on a modern private cloud with VMware Cloud Foundation 9.” Strong endorsements from MomentumAI and Mphasis - Mphasis emphasizes the forensic audit trail meeting “the same regulatory scrutiny as human-led processes.” Matthew Flug, IDC Research Manager for Intelligent Application Modernization & Deployment Platforms, calls out the release directly: “We see a real need for guidance, frameworks, and SDKs to help organizations kickstart agentic AI initiatives. With this new release, Tanzu Platform can help these organizations move from a crawl phase to a walk phase, ensuring they can leverage executive buy-in that translates into scalable, value-generating agentic AI."
"Broadcom announces VMware Tanzu Platform agent foundations for security, PaaS simplicity,” David Rubinstein, SD Times, April 15th, 2026. summary highlighting the security pillars that make the release credible for enterprise: immutable supply chain via buildpacks, structural secrets isolation, zero-trust networking, and OIDC identity for agents.
"Enterprise-Ready Agents Made Simple & Safe with VMware Tanzu Platform Agent Foundations," Camille Crowell-Lee, Tanzu blog, April 15th, 2026. Deeper architecture post on Agent Foundations. Walks through the three security pillars in detail, the MCP Gateway, Spring AI memory services, Postgres + pgvector, and the “digital passport” OIDC model that gives every agent a verifiable identity tied to a human.
"Can your private cloud handle the liability of an autonomous agent?," HyperFRAME Research, April 17th, 2026. Analyst piece calling out that “Broadcom is clearly positioning Tanzu as the ‘adult in the room’ for AI operations." Frames the platform as the mechanism through which enterprises “convert prototype agents into production-grade systems” and characterizes the forensic audit trail as “an attempt to turn a black box technology into a transparent, auditable business process." “Our research shows that only 14% of organizations currently classify their data architecture as fully modernized for AI, meaning most deployments are still constrained by legacy design assumptions.” - a gap Tanzu is well positioned to close.
"Tanzu Data Intelligence 10.4 Delivers AI-Driven Analytics, Unified Real-Time Operations, and Sovereign Resilience,” Arnab Chakraborty, Tanzu blog, April 15th, 2026. The data-side launch post. Covers the SQL Assistant, Greenplum MCP server, Apache Iceberg support, in-place Greenplum 6 to 7 upgrades, Transparent Data Encryption, and selective DR filtering.
"How AI-Assisted Analytics in Tanzu Data Intelligence Can Help Remove the SQL Bottleneck,” Arnab Chakraborty and Srinivas Banoth, Tanzu blog, April 15th, 2026. Focused deep-dive on the SQL Assistant. Natural-language to Greenplum-optimized SQL, read-only guardrails, and the promising shift for data teams “from writing code to architecting data strategies.”
"VMware Tanzu RabbitMQ Powers the Modern Data Lakehouse with New Spark Integration and Enterprise Tooling," Arnab Chakraborty and Prem Victor, Tanzu blog, April 15th, 2026. The messaging/streaming piece of 10.4. New bidirectional Apache Spark connector, RabbitMQ Streams as first-class Spark citizens, Stream Browser UI, and JMS Message Selectors. Fraud-detection closed-loop use case as the lead example.
"Tanzu Platform 10.4: a private cloud platform for AI harnesses (or, agentic AI)," Coté, cote.io/weblog, April 16th, 2026. My own take. Looks at the release from a “harnesses” PoV - the wrapping apps that guide multi-step AI workflows - and argues Tanzu is the backend plumbing enterprises need to actually ship this stuff. Includes the “what does a harness for paying your taxes look like?” provocation as a useful on-camera prompt.
"The Hidden Infrastructure Powering Safe AI in Financial Services," Corinium Intelligence, April 2026. Pre-AI in Finance Summit feature built around an extended interview with Purnima. She lays out the deny-by-default principle directly: “Every service connection, every database connection, every MCP tool connection… everything is granted explicitly." She frames the core enterprise challenge: “It’s really easy to write AI-powered software, but that doesn’t mean it’s easy to test, integrate, and securely operate enterprise-grade software." And warns that without hard constraints, “agents can run amok with unintended consequences such as changing or deleting critical data or accessing unauthorized systems or consuming large amounts of resources by running in an unbounded loop." Lays out a five-layer enterprise AI platform stack: agent runtime, middleware/AI content gateway, governed model brokerage, standardized dev frameworks, and low-latency data products.

Spaghetti DBMS, MCP for Knowledge Workers, The Harness Is the Harness - Related to your interests, Wednesday

Wed, 22 Apr 2026 10:14:27 +0200

Also: 55% of jobs reshaped, power over GPUs, a thousand DIY weekend platforms, and mood-tracking the robot

MCP Servers for, and of, knowledge workers

This is an excerpt from last week’s Tanzu Catsup. David and I have been talking a lot about the AI harness. We of course sprinkle in talk about how you would enterprise-y it up, esp. with security.

My key insight and angle on all this is thinking about how all the AI harness stuff applies to NOT programmers, to knowledge workers. That’w where the huge benefits will accrue, the huge amounts of compute will be driven, and where the big changes will happen, and need to.

More excerpts from that same episode: thinking of swarms of agents and having to click accept all the time really kills the vibes.

The Real Constraint on Enterprise AI isn’t GPUs; It’s Power - “When AI workloads are deployed in isolation and utilization is around 50-60%, we often see significant waste in server capacity as well as improperly provisioned power distribution for the expected peak load that didn’t materialize.”
Unified Self-Service Consumption for Modern Workloads - “Improving integration between tools and systems, which enables a unified platform, is the most desired organizational technology improvement, cited by 41% of 250 respondents in 451 Research’s Voice of the Enterprise: Workforce Productivity & Collaboration 2025 survey.” // “43% of 273 respondents project that the preferred IT environment for generative AI workload deployment in two years will be private cloud (on-premises, colocation, or hosted) as opposed to public cloud and SaaS alternatives.”
Sounds like someone got a 25% discount - “Amazon.com Inc. plans to invest up to $25 billion in Anthropic PBC and provide the company with a significant amount of additional computing capacity…. Anthropic, for its part, has committed to spending over $100 billion on Amazon Web Services over the next decade.”
Although, that also looks like another "maybe we’ll spend that much" situation: “As part of an expanded pact, the e-commerce and cloud giant is poised to invest an additional $5 billion in the Claude developer now and up to $20 billion more in the future. Anthropic, for its part, aims to spend more than $100 billion on Amazon Web Services tech over the next decade.”
BCG: 50-55% of Jobs Reshaped by AI in 2-3 Years, Only 10-15% Displaced - Entry-Level Is the Real Casualty - The tools you use drive how you work. When you have a new tool, if you don’t change how you work, you don’t get the full benefits of the tool. Incremental innovation vs. innovation innovation.
Early AI adopters stumble across the horror of vendor lock-in. Here we go again? - “of the 2/3 who have actually tried a [AI service, etc.] migration , only 42% report a smooth transition and 58% say it failed or blew way past estimates.” // You only call it “lock-in” when it stops working and/or you don’t like the pricing. Otherwise you call it “productive” and “strategic.” // As someone else put it: “what you pay for is work done."
The AI engineering stack we built internally — on the platform we ship - Cloudflares AI coding harness, at an enterprise-grade level. // This is like a reference architecture for everything needed to make an enterprise setup - it lacks the depth of governance and compliances that a bank would want. // Also, this launches thousands of DIY platforms projects where a developer said ‘I could do that in a weekend,’ brining all the downsides of DIY platforms to a new era of enterprise computering. // Looks cool!
Agent Harness Engineering - “A decent model with a great harness beats a great model with a bad harness.” // An overview of what AI harnesses look like and some tips on using them. Plus, this whole notion is evolving and changes as the models get better, or, at least change.
Designing Memory for AI Agents: Inside Linkedin’s Cognitive Memory Agent - InfoQ - 🤖 “The broader signal: production AI is defined less by the model and more by the memory, context management, and infrastructure around it. CMA is LinkedIn’s bet on externalised memory as a horizontal platform for adaptive, personalised agentic systems at scale.”
How The Access Group Scaled AI-Driven Development with Platform Engineering - “That’s the shift that matters. When you provide the right level of abstraction, you don’t just make systems easier to use; you make them easier to own. Engineers can take responsibility for the parts of the platform that matter to them, evolve those capabilities independently, and own their flow of value.” // This is what kicked off my tinker-talk.
This seems pretty good, actually. We’ve put the robot in talky-talk mode in the middle of the table at dinner several times and has good results re: family quality time. I mean, at one point in history, Socrates was like “next thing you’re going to tell me is that families (well, of course not the women) are going to sit down and talk about what’s in a book. That will ruin society. Also, does anyone really know what ‘dinner’ is?”

The spaghettified DBMS chart that shows Oracle's crown is slowly slipping.

Wastebook

“Every major enterprise-category vendor is shipping the same three-line marketing - ‘agents,’ ‘guardrails,’ ‘control plane’ - and the thing to watch is which one ends up with a durable product-shape in eighteen months.” - Claude summarizes recent enterprise AI marketing.
Nevermind mood journaling: I bet you track my mood over time by looking at my responses to the robot in AI chats. Happy: “nice, homie!” “Love it, home skillet!” Moody: “the fuck is wrong with you? You did this perfectly yesterday and last week.” “Stop asking shit-brained questions and just do the fucking shit I fucking asked for.” (To be fair…to me…those moodiness e.g.’s were a but arch. I’m not quite that toxic of a user.)

ICYMI

Let them tinker - hacking developer resistance to sound enterprise architecture and platforms - Instead of locking the enterprise platform down, let developers customize and experiment - that’s how you get adoption without losing the org benefits of a shared platform.
Macro-economic headwinds vs ‘The Terrace-in-April’ - Europe’s comfortable quality of life is real, but the external conditions propping it up are shifting, so growth isn’t optional if you want to keep the terrace in April.
Giving agentic AI a sandbox that won’t delete everything - Tanzu Catsup - David and I discuss some recent attempts to put more controls over free-wheeling agentic AI, here, how Claude code uses the bash shell and accesses data sources. Also: some experiments with using Tanzu Platform (Cloud Foundry) as a backend for shell commands. David also goes over a better option - an MCP Server - and how he’s been architecting swarms of agents patterns.

Logoff

I’ve been struggling of late to find relevance in what I’m interested in, do, and work on. To worship at the alter of the cult of businesses: yes, at the end of the day I can feel the bulk of activity I’ve done - there’s definitely very little Terrace-in-April in my life. But, is all that work relevant to where I was in the industry, what I was doing, and the “work” I felt I was up to several years ago when I was really pumping the stuff out? I type this no to whine (well, sure - yes) but more to find a tool to judge if this is overall “ok” and “expected.”

My pal Seroter loves linking to the occasional career and enterprise-cog-mental-therapy piece in HBR. There are two varieties: (1) take moments like these to chill, burn off burn-out kindle, and arrange your enterprise sock drawer to be ready for whatever comes next, or, (2) take this a a signal to burn the boats with a fucking flame-thrower and thrust yourself into the great unknown of the “now more than ever” enterprise-scape.

I have to say, I’m not into boats, so I don’t think I even have any.

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.

Wed, 22 Apr 2026 09:34:16 +0200

Still life in Amsterdam suburbs morning, April, 2026

Macro-economic headwinds vs "The Terrace-in-April"

Wed, 22 Apr 2026 09:18:40 +0200

I feel like there’s an unspoken set of assumptions and a slightly hidden “operating system” for how societies run. This comes out a lot when I read dense macro-economic analysis like this one. Let’s see what Opus 4.7’s angle is on my question:

If I look at this piece, it seems to be saying Europe is not growing and the money isn’t flowing. Prices go up, etc. However, it lacks a certain reminding us of what the point is. Why does it need to flow? Europe seems fine. People live their lives, leave work early and enjoy the sunny Spring. Is there some sort of problem? Europe has always been expensive, heavily taxes, and not a leading economy. How is this kind of thing going to break the life style and state of existence that Europe seems to want as revealed by how they live.

Athens, late summer 2025.

The robot says

Highlights below are mine.

You’re poking at something that economists and policy wonks tend to skate past because the answer is philosophically uncomfortable for them. Let me work through this with you, because I think you’re half-right in a way that’s more interesting than either full agreement or the standard rebuttal.

The hidden premise in the ING piece

The article, like most of its genre, assumes growth is self-evidently the goal. It never justifies this. It just lists symptoms – inflation, deficit above 7%, German industrial output flat for two years – and expects you to nod gravely. You’re asking the question the article refuses to ask: so what? If Czechs and Poles and Hungarians and the Dutch are drinking beers on terraces at 4 PM on a Tuesday in April, what exactly is being “lost”?

This is actually a serious question with a real literature behind it. Tyler Cowen’s The Great Stagnation and the Baumol effect debates sit in this territory, as does the older Easterlin Paradox – once a society crosses a certain threshold of material comfort, more GDP stops buying more happiness. Europe, in this reading, isn’t failing. It arrived.

But here’s where the “Europe is fine” view starts to crack

A few things the terrace-in-April picture obscures:

The lifestyle is a stock, not a flow. What you see in Amsterdam – the trains that work, the bike paths, the healthcare, the five weeks of vacation, the handsome 19th-century canal houses – is the accumulated capital of previous growth. It’s not being generated now; it’s being drawn down. Dutch infrastructure investment as a share of GDP has been declining for decades, and the housing crisis in Amsterdam is a direct consequence of a society that stopped building. You can coast on a stock for a long time. You cannot coast forever.
Demographics make “stagnation” into “decline” on a delay. Italy, Germany, and much of CEE have fertility rates around 1.3-1.4. Pension systems, healthcare systems, and the whole social-democratic bargain assume a working-age population roughly constant or growing. When it shrinks, either taxes rise on a smaller base, benefits get cut, or immigration fills the gap – and you’ve seen across Europe what happens politically when that last option gets chosen by default rather than deliberately. The “sunny Tuesday” depends on fiscal arithmetic that is quietly breaking.
Relative power matters even when absolute comfort doesn’t. This is the uncomfortable geopolitical point. A Europe that grows at 0.8% while the US grows at 2.5% and China grows at 4% becomes, over 30 years, a museum that tourists visit rather than a place that sets rules. That shows up as – to pick the current example – being unable to fund its own defense, being unable to build frontier AI, being unable to compete for the engineers who build the things that will matter in 2050. The Draghi report from September 2024 was basically a 400-page version of this argument, and it was received across European capitals with a shrug, which tells you something.
The specific thing the ING article is flagging – energy intensity – is genuinely a lifestyle risk, not just a growth risk. If Hungarian electricity is priced off marginal gas, and gas gets expensive, that’s not “Europe grows slower.” That’s “pensioners freeze in February” and “the aluminum smelter closes and 2,000 people in a small town lose jobs that won’t come back.” The terrace in April doesn’t show you the factory in Győr that just went dark.

Tavík František Šimon, Starý přístav v Amsterdamu (Old Harbor in Amsterdam), oil on canvas.

Where I think you’re genuinely onto something

That said, the European revealed preference is real and the growth-cult people underweight it. Here’s the steelman of your position:

The US grew faster and got opioid epidemics, school shootings, a life expectancy drop, and Trump out of the deal. If that’s the prize for 2.5% GDP growth, a rational European might decline.
Much of what gets measured as European “stagnation” is actually leisure preference – Prescott’s famous argument that Europeans work less because taxes make work less attractive, and this is a choice, not a failure. A Frenchman working 1,500 hours a year instead of 1,800 isn’t poorer in any meaningful sense; he has more time.
Growth accounting undercounts non-market production – the grandmother who watches the grandkids, the cook-at-home meal, the repaired bike – all of which Europe does more of than the US.
The climate math actually favors the European model. A world where everyone consumes like an American is a cooked world, literally.

So what’s the synthesis?

The honest answer is something like: Europe’s current lifestyle is viable indefinitely if you assume the world around Europe stays roughly the same. It doesn’t. The bet Europe is implicitly making is that the international order, cheap energy imports, American security guarantees, Chinese manufacturing, and benign demographics will all continue to hold while Europe enjoys the fruits. When any of those assumptions breaks – and 2022 broke the energy one, 2026 may be breaking it again, Trump is breaking the security one, and demographics are breaking on their own schedule – the lifestyle becomes expensive to maintain in ways that weren’t obvious before.

The ING piece is really making a narrow version of this argument: the 2022 playbook (borrow heavily, subsidize energy bills, wait for Russia to back down) worked because fiscal space existed and the shock was assumed temporary. In 2026 the fiscal space is gone, the shock may be structural, and the same playbook would just accelerate the underlying decay. Hence their pitch for using EU money as industrial policy – build grids and renewables so the next shock hurts less – rather than as consumption subsidy.

Whether that’s correct is genuinely contested. But the “why does it matter, people are happy” response, while emotionally satisfying, assumes a kind of closed-system Europe that doesn’t exist. The sunny terrace in April is partially subsidized by Qatari LNG, Chinese solar panels, American F-35s, and Polish truck drivers. When those subsidies get repriced, the terrace gets repriced too.

Or to put it in the bluntest terms: Europe can absolutely choose to be a nice place that doesn’t grow much. What it cannot choose is to be that while the rest of the world grows around it, because growth elsewhere eventually sets the price of everything Europe imports, including its own security. That’s the part the article is gesturing at but, in the genteel ING-economist idiom, won’t quite say.

Paul Paeschke, Potsdamer Platz, c. 1929. Via The Blinkered Flâneur.

You can’t take your 401(k) to the grave

Back to me, Coté.

When I first moved to Europe, lived there for a year or so, my theory was that Europeans were, essentially, “done” and in a sort of of societal “retirement.” They had achieved the goal, they had done the stuff. It was time to enjoy the benefits of all that work. They had built and profited from centuries of extracting “value” out of continues, the industries revolution. That is, they were incredibly successful and (relative to the rest of the world) wealthy.

However, Europeans had also been killing each other in wards for 2,000 years. The most recent wars were awful, the worst in history. Any English major can tell you that had a profound effect on whatever the European Mindset is.

The result of that two millennium of non-stop action is both exhaustion but also a sort of “well, I mean, aren’t we sort of done here? Call up Fidelity and let’s start drawing from the 401(k) instead of paying into it.”

This is a mindset, a way of life, that I don’t think Americans can ever really understand or feel empathy for. We’re only a few centuries into all this and it seems wild and wonderful. I mean, we don’t even have terraces.