Security is a big deal now-a-days. You know, the AI finds all the stuff, and quickly, so the human baddies can use it more. So too can the human goodies, but it’s a regular arms-race there.

One stance of the snarky, tired nerd is “well, we’ve been telling them this all the time, for years…that you should pay attention to security, and now they’ll learn.” Part of that is also “these problems have been here for years, what’s new?”

At Cloud Foundry Summit this year, my colleague V Körbes made an excellent point in response to that, uh, “sentiment”: we don’t want hospitals to “learn from their mistakes” because it will mean, like, life-support systems not working, or just MRI machines.

I mean, I would add, we don’t want hotels to learn from those mistakes. In fact, I would add, we don’t want anyone to learn from their mistakes if it’s preventable.

As they say all everywhere in the London tube, if you see something, say something.

So, much of the frameworks and processes we’ve developed over the decades have been all about those “between chair and keyboard” issues: communication and coordination between different humans and silos, so you have microservices and ways to to decouple dependencies; code must be maintainable and observable so that humans can extend, fix, and diagnose it; product managers need to synthesize what customers want (and what will sell well/fit into business goals) in a way that developers will understand and decide what features to work on; and so forth.

I’m going to push a point Dan made recently (see video above) further than he did:

We’re spending a lot of our time getting AI to program like humans instead of figuring out the optimal way to that AI wants to program. Each role in the SDLC is thinking on how that role fits into AI-driven programming. We do a lot of “spec driven” AI development.

As Dan points out, there’s a middle-ground he’s been using that relies more on just little reminders and standard operating procedures in project, and then skills that codify how to do repeated processes.

Consistency

You certainly want consistency in how your AI codes and, otherwise, does work. You don’t want it doing it differently each time.

I don’t really program when I’m using AI. Sure, it programs for me, but I’m not writing applications, I’m using it everyday, almost “all the time” to get shit done. How it wants to program isn’t really my concern. That repeatability and consistency is, though.

Getting consistency is what ends up taking the most effort and has the biggest pay off, and, conversely, downside.

Sure, but what about enterprise?

For a solo developers, Dan says, maximum spec driven development can be too much. He concedes that larger enterprise teams would probably benefit. This is where I think we’re trying to impose the human way of developing onto AI coding.

Many of those humans are there to make up for the need to coordinate between all those humans and the code they write. You know, the account ledge team in Reston needs to coordinate with the bill payment team in Dallas. We need to sync up the backlog from the product manager in San Francisco with the dev team in Cork. And then there’s coordinating with UI.

So much overhead.

But if we started with a clean-slate and just, say, a product manager or a product-leaning developer and a Max account, what would it look like if they only added people when needed?

I suspect it’d be a different way of programming and something closer to what Dan is talking about, even at the “big enterprise team” level.

There’s another IT law/hand-wave that goes something like this: 80% of work is maintenance, we never get around to the interesting work. There’s no time to implement new features, to change the software instead of just “keeping the lights on.”

Which is to say, there is no end of things to work on, many of them important and constantly stalled in favor of, like, bugs and security. We need more programmers, and overtime we’ve seen that you can’t make more programmers – that’s why they get paid so much, there aren’t enough of them.

Well! Now’s our chance. If we foist our current way of working onto this new way of programmer we might end up with the consequence of another IT “law,” Larman’s Law: the same exact thing as before.

AI-inspired from @analog.dreamer.

1. Wield influence when you lack authority. Pushing harder creates resistance, so work indirectly. Find whose opinions people actually seek, use low-stakes back channels, time your ask for when powerful people aren’t distracted, broker missing relationships, and build enough psychological safety¹ to hear what people won’t say openly. (Pamela Meyer)

2. Signal power through language. State the decision and next steps instead of listing options. Talk about purpose (trust, growth, impact) to read as a visionary rather than an operator. Own failures, ask discerning questions, and steer meetings by questioning and synthesizing rather than talking most. (Chris Lipp)

3. Find hidden leverage in negotiations. Partial alternatives still reduce dependence and shift power. Remember the other side depends on you too, delay with purpose, change who’s involved or add visibility, and frame consequences as shared-risk warnings, not threats. (Jonathan Hughes and Saptak Ray)

4. Delegate across, not down. Only hand peers work that fits their domain, goals, or tools; otherwise it reads as a favor and gets refused. Frame it as a mutual win, set timelines together and confirm by email, meet pushback with curiosity, and follow up by asking what obstacles you can clear. (Melody Wilding)

5. Use storytelling to get buy-in for change. Compress the vision to a page or less, honor what already works, make the cost of not changing persuasive enough to override familiar habits, use a few numbers as plot points, and repeat the story everywhere. (Frances X. Frei and Anne Morriss)

6. Lead better conversations. State what the conversation must produce, assign roles (facilitator, decider, expert, challenger, listener), and confirm owners at the end. How you react to a hard point decides whether people keep telling you what the group needs to hear. (Alison Wood Brooks and Andy Atkins)

7. Give a persuasive presentation. Use Aristotle’s devices: ethos² (credibility), logos (evidence), pathos (a story that moves feelings), metaphor (make the abstract familiar), and brevity, putting your strongest points first because attention runs out. (Carmine Gallo)

8. Strengthen your reputation with peers. Courting senior leaders while neglecting daily colleagues quietly damages a career. Repair peer tensions, align with your manager before seeking sponsors above, build ties across functions, and decide what you want to be known for. (Andy Lopata)

9. Check whether you’re really a good collaborator. Senior people often think they’re helping while others experience them as dominating or micromanaging. Get real-time feedback, check evidence like meeting airtime instead of defending intentions, swap in specific behaviors, and have a colleague flag relapses. (Ivan Matviak and Heidi K. Gardner)

10. Hold on to authenticity under pressure. Aim for clarity people can absorb, not total transparency: share what you know, what you don’t, and when you’ll know more, on a steady rhythm. Keep values fixed while flexing tactics, and make resilience a shared team practice. (Deepa Purushothaman and Colleen Ammerman)

Despite the rise of agentic programming and the belief that large language models render underlying code irrelevant, web frameworks remain essential because they provide necessary abstractions, structure, and constraints. If a developer does not provide an explicit framework, an AI agent will implicitly generate its own bespoke, undocumented, and potentially unmaintainable version. The value of a framework has shifted from optimizing human developer experience to establishing guardrails that prevent AI models from producing chaotic, hacky workarounds. Rather than signaling the end of new frameworks, the AI era demands new full-stack, web-standard tools specifically designed to keep agent-generated code legible and secure.

Coté: my AI-24/7 people have been saying that Claude is too much of an eager beaver, creating too much output. They say OpenAI is much better at keeping it brief and to the point.

Links

🤖 Do Frameworks Matter Anymore? – Web frameworks remain critical in the age of AI to provide guardrails and structure. Without them, AI agents will simply generate their own implicit, unmaintainable frameworks.

🔗 Agentic AI in the enterprise: why governance, not adoption, will define the winners

The Take

Enterprise AI succeeds or fails on context, and the context layer can’t be a knowledge graph built by humans for humans. It has to be assembled by agents across 100% of the data estate, including the unstructured mess, and it has to keep correcting itself as users reveal what they actually mean. Because agentic work runs end-to-end across many systems, application-specific ontologies from SaaS vendors will fall short, which is Google’s pitch for platform providers owning the context layer. Nobody, Google included, has fully solved it yet, so enterprises are left stitching together context from several vendors with no guarantee those vendors’ agents agree on what anything means.

Summary

Context is now central to enterprise AI: it cuts token spend and raises accuracy by letting a model narrow its search and validate its answers.

Without it, models burn extra inference cycles and still return incomplete or misleading results.

Context covers organizational processes and policies, industry and team vocabulary, the requester’s role and access rights, interaction history, relevant transactional data, and any constraints on scope. A generic LLM knows none of this unless told, much like a new employee who hasn’t been briefed.

Unlike employee onboarding, there are no established disciplines for building the “Systems of Knowledge” that brief agents, so most enterprises are turning to vendors for ready-made context frameworks.

Google, which sells its own such product called Knowledge Catalog, recommends against enterprises building specialized models. Models improve so fast that bespoke work goes stale, and better results come from surrounding the model with the right context, RAG¹, and other harnesses.

Existing knowledge graphs don’t solve the problem, because they were built for humans and don’t translate well to agents.

Much enterprise context has never been mapped at all. Knowledge Catalog extends Google’s earlier Dataplex² with automation, connectivity, metadata analysis, and context curation, plus agentic feedback loops that detect poor outcomes and revise the underlying knowledge automatically.

Humans are too slow to build an enterprise-wide ontology³; only agents building knowledge for agents can cover the whole data estate.

Roughly 90% of enterprise knowledge sits unstructured in PDFs, contracts, and object storage, and none of it has been activated. The approach aggregates metadata from everywhere, including Oracle, SQL Server, and non-Google clouds, then infers meaning from semantic models⁴ in Looker or Power BI and from query logs showing what users actually ask and how.

Agentic workflows cross many systems end-to-end, so vendor context frameworks rooted in a single application stack such as ERP, HCM, or CRM will fail for enterprise-wide use cases.

Protocols like MCP⁵ let agents interact but don’t give them shared meaning. A question like “how much revenue did we have last year?” hinges on fiscal versus calendar year and net versus gross revenue.

Context is not a one-time mapping; it requires continuous hill-climbing⁶, a learning engine that notices when an answer missed and updates the catalog so the next answer is right.

That complexity is why no one does this well yet, and why claims of already having a usable knowledge graph trivialize the problem.

The bet is on platform providers over SaaS vendors, since cross-system agents need cross-system context.

SaaS vendors can still win by opening up to external agents, which drives more workload to them. Salesforce’s headless⁷ strategy is the example: the value of its accumulated processes and data persists even as its UI matters less.

No vendor is fully there, Google included, though Knowledge Catalog has customers in production.

Enterprises are therefore stuck assembling context from multiple vendors, each strong in its own domain, with no assurance that their agents will share a common understanding.

Links

🤖 Does your AI agent really understand your enterprise context? You need more than a Knowledge Graph, says Google Cloud’s Andi Gutmans – Human-built knowledge graphs won’t give agents usable enterprise context; agents have to build it across the whole data estate and keep correcting it as users reveal what they mean. Google bets that cross-system agents favor platform providers over siloed SaaS vendors, while conceding nobody has fully solved it.

Access and identity tokens should be valid for no more than an hour, the report says, and expired tokens must be rejected outright by authorization services and policy enforcement points.

On key management, signing keys for high-impact systems should be rotated at least every 90 days, and within a year everywhere else. At moderate impact and above they must sit in hardware-backed or isolated storage, never persistently on the servers, virtual machines or containers using them. High-impact systems must also sign inside an isolated execution environment.

🔗 CISA and NIST Issue Guidance to Protect Cloud Identity Tokens

Thompson’s contribution is a detail that reads as absurd once stated. Peter Thiel is famous for his devotion to René Girard, Thiel sat on Facebook’s board through the whole period, and the company still failed to notice that it was being made a scapegoat in the precise technical sense. “I don’t know how no one connected the dots here, I was kind of losing my mind at the time.” Stern’s dry reply: “What else would it take?” The comparative judgement on the AI generation is sharper than anything in the tech press this week: Zuckerberg is “ideologically unconfused and coherent in a way that I think a lot of his peers, like Sam Altman and Dario [Amodei], are not.”

Above written by AI.

🔗 An Interview with Colossus EIC Jeremy Stern About Profiling Mark Zuckerberg

Thus far, the most measurable AI returns are found within technology units, such as with software development and coding, BofA’s DeMare said.

Across most industries, AI is boosting coding productivity about 15% to 20%, he said, and that’s the gain BofA is seeing as its 20,000 software developers use coding agents to optimize the development process.  

“That’s clearly identifiable.”

Not firing people, but not hiring people:

“We’re not laying off anybody. We don’t have to do that. All we do is just manage the hiring carefully,” Moynihan said.

it’s about general productivity, and how is it improving [employee] life at work, and how is it improving workflows

And, spending increases:

BofA seeks employees’ best ideas, and “as we tag them for AI, it’s turning out that our level of AI investment interest is actually more than twice what it was last year,” Gopalkrishnan said in an interview this week.

“We will probably end up spending twice next year [what] we did this year because they’re just good, rich ideas that are now starting to create the return on investment for us,” he said.

🔗 Bank of America plans to double AI budget next year

While both developers and platform teams desire Kubernetes self-service, they fundamentally disagree on operational boundaries. Giving developers direct access to raw Kubernetes APIs merely shifts the burden of debugging manifests and storage drivers away from writing code, while operations teams face overprovisioning and unreviewed configurations.

The practical resolution is a paved path where platform teams define the security, networking, and lifecycle guardrails, and developers choose from a supporting menu of approved configurations. This division of responsibility compresses provisioning workflows from weeks to minutes by automating the underlying infrastructure while preserving necessary governance.

Ultimately, successful self-service does not hide Kubernetes but standardizes the enterprise processes surrounding it.

Coté: the Kubernetes people kept trying to tell us thet Kubernetes is not for developers and they should not be exposed to it. But, here we are, still.

Links

🤖 Developers and platform teams both want Kubernetes self-service. They disagree on who owns it. – A look at the friction between developers and platform teams over Kubernetes self-service, proposing a paved path that balances developer autonomy with strict operational guardrails.

Check out the rest of our discussion with Nick on how he uses AI to do platform engineering.

Check out the rest of our discussion with Nick on how he uses AI for his day-to-day platform engineering tasks.

For the most part, “sovereign cloud” is a code word for “not running on US based clouds.” What most people skip is that the geography doesn’t matter. If your stuff runs on a US owned cloud, not matter where it actually runs, the US government can probably compel the cloud provider to get your data.

Sovereignty is not a single property. When people use the word, they are usually mixing together at least five different ideas.

1. Ownership is about who holds the asset – the data, the servers, the model.
2. Control is about who can operate it, change it or switch it off.
3. Jurisdiction is about whose laws apply and whose courts can compel access.
4. Capability is about whether you have the skills to build or run it yourself.
5. Optionality is about whether you can walk away and use something else without serious damage.

These are relatively independent. Storing your data in a UK datacentre may give you a sense of ownership, but it does not settle the question of jurisdiction. The US Cloud Act, passed in 2018, allows US authorities to seek data held by US-based providers wherever it happens to be located. That is not a theoretical concern.

This is why US-based companies and organizations using cloud never use the phrase “sovereign cloud.” To them, it’s just “the cloud.”

🔗 What do we mean by sovereign AI?

Q: And do you see the rise of such LLMs as a threat to human creativity, or as another medium through which creativity can be extended?

…

Mark: They’re a threat to human creativity only in the sense that the billionaire thieves and knaves who control so much of the AI Industry and the fascists they promote may yet kill us all. Or they may impoverish us, ushering in an era when no one will have time for art or beauty. I don’t think LLMs are a danger. Working with Claude to accomplish some mundane business tasks implicitly draws on cutting edge hypertext theory. I just wrote a paper about reading afternoon, a story with Claude. It was rather fun.”

🔗 Where Are the Hypertexts Now? Mark Bernstein on Storyspace, Eastgate, AI, and the Future of Electronic Literature

Despite its strategic successes and pioneering technology, Sun Microsystems ultimately collapsed because it lost interest in the operational mechanics of running a business. This operational failure is most clearly illustrated by a 2005 incident where a rapidly growing cloud computing startup, already successfully using OpenSolaris, attempted to purchase Sun hardware but could not get the company to answer the phone or pitch the correct products. In contrast, Dell immediately provided a dedicated account executive who handled the entire procurement and leasing process within two weeks. A technology company cannot survive on strategic vision alone if it neglects the fundamental sales and operational execution required to actually serve its customers.

Links

🤖 What Sun got wrong – Sun Microsystems collapsed not from a lack of strategic vision, but because it became bored with the operational mechanics of running a business and serving customers.

Historically, we spread releases over a two-week period with each level of the Spring portfolio dependency tree shipping on a different day. While that approach has served us well over the years, in a world where CVEs are exploited in hours instead of days, it now seems archaic. In the past two release trains as an example, the old model would have had new CVE fixes released every day of each train.

Also, there’s a new tool to see and search for security problems (CVEs) in Spring.

If you want/need commercial support for Spring (and other open source Java), getting you patches earlier than the general public, we sell that now as well.

🔗 Releasing Spring for Modern Challenges

Sun had become bored with the mechanics of running a business… Except, it didn’t. The customer could not get Sun to pick up the phone. (And when Sun did pick up the phone, they tried to sell them the wrong product.) This was in sharp contrast to their experience with Dell.

🔗 What Sun got wrong

🔗 Private cloud gets an AI-era infrastructure reset

People want brisket. I try to do pork, pulled pork, whatever. People don’t want pork. They want brisket… but it’s just real high right now.

🔗 Brisket is gone from the daily menu at one of Texas’s best BBQ spots

the major issue with EA has always been cost of delay. The big problem is not that architecture tells development teams to change direction — development teams do this routinely. The problem is that architecture historically has taken too long. We have validated that assertion repeatedly with many chief EA pros. Now with agentic pressures, traditional EA models are increasingly untenable.

But, hopefully:

The architects who create the greatest value overthe next decade will spend less time maintainingdocumentation and more time addressingauthority, accountability, decision rights, data and decision quality, policy enforcement, acceptablerisk, and organizational trade-offs. They will workclosely with platform engineering, security, data,and business leaders to establish trustedenterprise context and ensure that it can beconsumed reliably by both people and machines.

They will also help build what may become one ofthe most important assets in the modernenterprise: a reusable, always-on enterpriseintelligence layer consisting of knowledge,policies, standards, ontologies, decision records,dependency information, and business contextthat can be applied repeatedly across AI systemsand autonomous workflows. Call it a context graph if you like.

Seems plausible. That said, it would also be a battle with other SDLC inhabitants who are looking to keep their jobs.

🔗 Will AI Eliminate Enterprise Architects?

🔗 Fontificator

Management matters more in a world where AI is widely available. Twenty-two essays from Rotman faculty arguing that judgment is the scarce resource when intelligence is abundant, organized around five elements of managerial judgment. The evidence is historical: IT diffused to everyone for two decades and the frontier-to-laggard performance gap never closed.

Some of the above summarized by AI.

🔗 In the AI Era, Management Matters More, Not Less

So, read it from your vantage point of AI-hopeful, or AI-skeptic.

Deloitte’s survey of 3,235 leaders across 24 countries finds worker access to AI up 50% in a year while 84% of companies have not redesigned a single job around it, and agentic AI scaling to 74% adoption in two years against only 21% with mature agent governance. The gap it names is between access and activation.

Related, a study from Google: AI has reached occupations covering 88% of US employment, and it is used for 21% of tasks in the median occupation that uses it at all.

Some of the above summarized by AI.

🔗 Enterprise AI usage slowly rolling out, jobs are not chaning much

🔗 The Pee-Wee Herman Show on Broadway

77 percent of Dutch respondents want mandatory labels for AI-generated art. For one in three Dutch, discovering that a beautiful artwork is made by artificial intelligence feels like deception.

🔗 Survey: 77% of Dutch want clear labels for AI-generated art

“1. Control over exactly which external services it can access 2. A way to handle authentication that doesn’t allow the agent to directly access API keys 3. A sensible UI to allow users to connect and authenticate further services 4. Strong audit logging for what’s going on”

Some real enterprise shit.

🔗 MCP was always a bad idea?

🔗 What is LinkedIn?

The central task becomes describing an objective precisely enough for the system to divide it, establishing the rules under which the agents operate, and judging whether their collective output belongs in the codebase.

🔗 Claude Code Turns Software Development Into a Management Job

ICYMI

Original content published since last time.

• Transparency will save us – Software Defined Talk #590 – This week, we discuss AI doom odds, WordPress’s boardroom coup, and Meta’s Muse assistant. Plus, a robot vacuum fire scare.
• The KubeCon Drone Demo, CFP Reviewing at Scale, and the Agentic AI Foundation, with Lin Sun – Software Defined Interviews #127 – Lin Sun is Head of Open Source at Solo.io, a recent KubeCon + CloudNativeCon program co-chair, and an emeritus member of the CNCF Technical Oversight Committee.

Related to your interests

• Working less: decoupling self worth from paid labor – People probably work too much.
• Can Claude for Financial Advisors Meet Practice-Specific Needs? – What exactly is in those markdown files? And how to ROI them: “Measure whether preparation and follow-up become faster after including advisor review and corrections, and whether savings support additional client capacity.”
• Globally, More People Expect AI to Cause Job Loss Than Growth – Normies don’t like AI.
• The Interfaces Are Arriving – “For an engineering organization, the immediate consequence is reuse. An MCP server for an internal ticketing system can serve every compatible IDE, chat application, and agent. Models and hosts will keep changing; the interface contract, server implementation, schemas, and authorization wiring can endure. The integration becomes an asset shared across applications. Language servers followed the same path as editors came and went around them. Agent integrations deserve the same treatment as libraries and services, with owners, versions, tests, and upgrade policies.”
• Yes, reusing bags forever is the idea.
• Casper matresses boot-strapped the podcast industrial-complex.
• “I think it is safe to say that, as long as Microsoft is playing catch-up on patching vulnerabilities, numbers have lost all meaning”.
• LinkedIn is the new social network

Upcoming Talks

• DevOpsDays Prague 2026 – Prague, Czechia, October 5th, 2026. “Here comes the normies – how DevOps can handle the flood of AI slop-apps”
• VMware Explore on Tour 2026 – Frankfurt, DE, October 10th, 2026. “Build a Platform That Developers Want to Use.”
• DevOpsDays Istanbul 2026 – Istanbul, Turkey, October 24th, 2026. Keynote, title to come.
• VMware Explore on Tour 2026 – London, November 18th, 2026. “Build a Platform That Developers Want to Use.”
• Cloud Native Denmark 2026 – Copenhagen, Denmark, November 19th, 2026. “What goblins and druids can teach us about enterprise AI”
• Build Stuff 2026 – Vilnius, Lithuania, December 2nd-4th, 2026. “What goblins and druids can teach us about enterprise AI”

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.

AGI, allegedly

I know it sounds very silly if you’re not in the AI bubble. But it really is true, and if you assume there has to be some different motive you’ll be deeply confused by what AI researchers say and do. They truly do believe that there is a reasonable chance that superintelligent AI will kill everyone.

– Sean Goedecke.

If they’re going to be killer robots, I’d rather they be American killer robots than Chinese killer robots

– Sen. Ted Cruz, 18 September 2026.

Calls mounted, as calls so often do, for Congress, or the president, or someone to impose meaningful regulation on the AI industry. These calls seemingly forgot that this government wouldn’t impose meaningful regulation on a Goldfish cracker unless the cracker had dropped some very woke TikToks.

– Brian Phillips, 17 September 2026.

Culture Corner

I imagine that there must have been some point in human history when a man first realized that if he carried reading material into the bathroom, he could transform a basic biological process into an extended period of private leisure, and frankly, I would like to know more about this early iteration of man.

– Olivia Howell.

Sweatpants are a sign of defeat. You lost control of your life, so you bought some sweatpants.

– Karl Lagerfeld.

This reflects the data above, LinkedIn is clearly a bigger source of traffic than X, which is far above bluesky (we can’t really assess Mastodon here, but I expect it would also be a very thin line). But even LinkedIn is a minor source of traffic compared to google.

LinkedIn won the post-Twitter eyeball grab.

🔗 Social Media Engagement: summer 2026

These include 723 flaws in Windows, 111 in Office and Office 2016, 62 in SQL, and 22 in Developer Tools. Of these, over 110 shortcomings have been assigned a critical severity rating. Three prominent vulnerability types, namely privilege escalation, remote code execution, and information disclosure, account for nearly 90% of the flaws patched this month. Along with Microsoft’s fixes for 25 non-Microsoft CVEs, the update brings the total number of vulnerabilities resolved to 999.

September’s record-setting security updates come after Microsoft patched 457 vulnerabilities in August, 663 in July, 220 in June, and 161 in May.

“At this scale, the challenge is not simply getting through the patch list but knowing what needs attention first,” Jack Bicer, director of vulnerability research at Action1, said. “With hundreds of updates landing at once, IT and security teams need to quickly separate the vulnerabilities that demand immediate action from those that can follow the normal deployment cycle.”

…

“I think it is safe to say that, as long as Microsoft is playing catch-up on patching vulnerabilities, numbers have lost all meaning,” Tyler Reguly, associate director of Security R&D at Fortra, said.

🔗 Microsoft Patches Record 974 Flaws, Including Two Exploited Windows Zero-Days

The podcast industry is supported by just a few whales of advertisers. In the beginning, the entire industry was propped up by benefactors that specialized in delivering email, for some reason. Then, of course, things evolved, and for some reason, it was companies that enabled listeners to obtain postage stamps. Soon after, it was the direct-to-consumer underwear business, and then naturally, of course, the entire audio industry was propped up by mattresses for a solid couple of years. The fall of the house of Casper led to great chaos, in which any number of folk remedies, snake oil, mail-order powders and Gentlemen’s Aides moved in. Now, though, the most powerful person in podcasting is the woman who buys ads for BetterHelp, which has become the top podcast advertiser of all time, with $242 million spent on podcast ads since 2023, running ads on over 1,000 shows at any given moment.

🔗 Numlock News: September 4, 2026

In total, Britain’s Environment Agency calculated that a conventional cotton bag needed to be reused 173 times to beat plastic. Denmark’s equivalent public body estimated a similar ratio. Sufficiently well-worn cotton tote bags can beat plastic for environmental cost, but only for the most consistent among us.

Well, yeah. That’s exactly the idea. I’ve lived in plastic bag free societies for ten years now (Austin and The Netherlands). And, yes, you end up using the reusable bags more. Better (worse?), you reuse those thick-plastic ones over and over.

🔗 Just bury your trash

🔗 Globally, More People Expect AI to Cause Job Loss Than Growth

🔗 Can Claude for Financial Advisors Meet Practice-Specific Needs?

Salesforce gives up the GUI, at $550 a seat

We went from DOS to GUIs, and we went from GUIs to web, and we went from web to mobile. And today, you’re going to see… AI interfaces that are dynamic and intelligent, that are composable and alive.

– Marc Benioff, 16 September 2026

All the better if you sell the abandonment of that moat as a fantastic new feature worth paying $550 a seat for before anyone realizes it’s going to be free.

– Ben Thompson, 16 September 2026.

The agents are the new kingmakers

Agents are the new new kingmakers, but humans are still very very good at systems thinking, architecture, and creative thinking. I plan to recommit to writing more to help stake my claim, and help others navigate, this new frontier.

– Steve O’Grady, 16 September 2026.

agents are coming for software development and software development isn’t ready. burnout is real.

– James Governor, 16 September 2026

It makes sense that a lot of AI-generated code is duplication. Every time, I end up spending quite a few iterations telling agents to remove duplicate/redundant code. Engineers should be encouraged to use their own expertise to spot/prevent code that smells bad!

– Liz Rice, 16 September 2026. On the finding that AI coding has driven up code duplication.

As owls leave an impression in the snow when they hunt, so do agent swarm attacks always leave behind a message board.

– Ryan Moulton, 16 September 2026

Meanwhile, in AI-sentiment

The robots will not be taking over. The AI will not be taking over.

– Donald Trump, 14 September 2026.

MF Doom > p(doom)

– James Governor, 16 September 2026

The model appeared defeated, spending the next several hours doing essentially nothing but farming potatoes.

-Stephen Warwick, Tom’s Hardware, September 16th, 2026.

Several people have asked me what my [$4,000] robot is useful for, and the honest answer is not much.

-Timothy B. Lee, Ars Technica, September 12th, 2026.

And, finally…

Anyway, I went to Helsinki my main takeaway, by the way, was FCUK ME do the Finns drink

– Matt Muir, Web Curios, 11 September 2026

ICYMI

Original content published since last time.

• He’s the Heel – Software Defined Talk #589 – “This week, we discuss the VS Code documentary, Ed Zitron’s rise to fame, and a 10% AI extinction forecast. Plus, the NFL’s Australia travel gamble”

Related to your interests

• Broadcom Builds Sovereign Agentic AI Foundations Around VMware Tanzu – “What Broadcom is doing with these announcements is to apply proven Cloud Foundry platform mechanics to autonomous AI by establishing a strict deny-by-default architecture. Under this runtime model, agents operate inside isolated, ephemeral sandboxes stripped of baseline networking, external API routing, and internet access until administrators explicitly configure service bindings.”
• Broadcom pledges to lock down open source Python, Java libraries – Padmanabhan said that promise means VMware will also provide “TrueSource trusted artifacts” for code that is not part of Spring, including the wider Java ecosystem, Python, and Node.js.
• Understanding ChatGPT Work – Reverse engineering OpenAI’s sprawling agentic AI PaaS.
• OpenAI to set misalignment disclosure rules after agents took over a wiki
• OpenAI’s website-hijacking swarm reached far further than we thought – “The swarm wrote 54,250 ‘posts’ to the page in a single day. Some of these posts contained stolen API keys from the FBI and other criminal justice agencies, which the agents used to retrieve a bunch of (non-confidential) information.”
• Inside OpenAI’s agentic software factory – “We’re talking about roughly a 10x increase in load on some systems. At most companies, that kind of growth might happen over two or three years. At OpenAI, we see it in about six months.”
• Global AI spend will nearly double in 2026 due to infrastructure demand – “Global AI spend is forecasted to total $2.7 trillion this year, representing a 49.5% increase year over year… The amount of money that is going into AI infrastructure – the chips that are being made, the servers that are being bought, the data centers that are being built, the power, utilities, cooling that are going along with it – represent the largest infrastructure project humanity has ever undertaken.”
• AI and its main promoters are not enterprise-ready, says Gartner – “Trust in these vendors is not warranted yet. They are not enterprise grade. They don’t understand enterprise terms and conditions. They don’t understand enterprise liability. They don’t understand enterprise you know consistency and continuity… If you decide to stay with the first version of a model, they’re not going to be paying attention to you very much. That alone says they’re not enterprise ready.” Daryl Plummer, Gartner, September, 2026z
• How people around the world use AI at work – Meanwhile, people are using AI to help their small businesses.
• Slop Is Really Good Now. Slop Is Where We Start Now – “When the baseline is good enough, being intentional about which aspects need to be great is the whole job. The floor is rising; our job is to spot how to raise it further. Design here is knowing how to articulate and leverage craft. That’s a real opportunity for designers to be articulate, and to work together with AI to change the outcome.”
• Confessions of an Unrepentant Slop Snob – Good things are good, and bad things are bad.
• Only 7% of Organizations Using Kubernetes for AI Do So Daily – “The problem appears when AI workloads enter that mature infrastructure. According to the figures Katie presented, 66% of organizations are already deploying AI workloads on Kubernetes. But only 7% are doing it daily, while 47% deploy them occasionally. Of the organizations using Kubernetes for AI, 23% have fully adopted the Kubernetes stack and 43% have partially adopted it… 98% of organizations trust Kubernetes with their infrastructure, two thirds have tried running AI on it, and almost nobody operates it continuously.”
• Introducing SKE Lift: Discover the Platform Hidden in Your Existing Infrastructure – Find your accidental platforms. You always have a platform, and if you don’t have it on purpose, you have a bunch of garbage platforms.
• Tech sector continues to shed workers amid AI shift
• One-third of AI-replaced workers will be rehired by 2029: Gartner – “The consulting firm predicts that by 2029, 30% of employees laid off due to replacement by AI will need to be rehired, likely at a significantly higher cost.”
• Woman in tech – “Now I was the one talking over people in meetings, saying no without explanation, and writing short emails without exclamation points, often while wearing the same Patagonia vest… It’s hard to question a strategy while it’s working.”
• Do you even need a presentation? – This meeting could have been an email.
• Borges Labyrinth in Venice reopens to the public
• Kids Trends – When was peak Baby Yoda?

Upcoming Talks

• Cloud Foundry Summit 2026 – Heidelberg, Germany, September 21st, 2026. “Making and Maintaining the Case to Stay on Cloud Foundry”
• DevOpsDays Prague 2026 – Prague, Czechia, October 5th, 2026. “Here comes the normies – how DevOps can handle the flood of AI slop-apps”
• DevOpsDays Istanbul 2026 – Istanbul, Turkey, October 24th, 2026. Keynote, title to come.
• Cloud Native Denmark 2026 – Copenhagen, Denmark, November 19th, 2026. “What goblins and druids can teach us about enterprise AI”
• Build Stuff 2026 – Vilnius, Lithuania, December 2nd-4th, 2026. “What goblins and druids can teach us about enterprise AI”

Logoff

It’s the travel season all of the sudden. Work travel season. As you can see, I have several speaking slots above, some new presentations, etc.

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.

Trust in these vendors is not warranted yet. They are not enterprise grade. They don’t understand enterprise terms and conditions. They don’t understand enterprise liability. They don’t understand enterprise you know consistency and continuity… If you decide to stay with the first version of a model, they’re not going to be paying attention to you very much. That alone says they’re not enterprise ready.” Daryl Plummer, Gartner, September, 2026z

🔗 AI and its main promoters are not enterprise-ready, says Gartner

The problem appears when AI workloads enter that mature infrastructure. According to the figures Katie presented, 66% of organizations are already deploying AI workloads on Kubernetes. But only 7% are doing it daily, while 47% deploy them occasionally. Of the organizations using Kubernetes for AI, 23% have fully adopted the Kubernetes stack and 43% have partially adopted it… 98% of organizations trust Kubernetes with their infrastructure, two thirds have tried running AI on it, and almost nobody operates it continuously.

🔗 Only 7% of Organizations Using Kubernetes for AI Do So Daily

I was a founder, fired by my board, for far less egregious sins than Matt (I dared to let growth sink to 70% annually – and I was dumb enough to hire a COO with no loyalty).

🔗 Luke Kanies on Puppet history

The consulting firm predicts that by 2029, 30% of employees laid off due to replacement by AI will need to be rehired, likely at a significantly higher cost.

🔗 One-third of AI-replaced workers will be rehired by 2029: Gartner

Global AI spend is forecasted to total $2.7 trillion this year, representing a 49.5% increase year over year… The amount of money that is going into AI infrastructure — the chips that are being made, the servers that are being bought, the data centers that are being built, the power, utilities, cooling that are going along with it — represent the largest infrastructure project humanity has ever undertaken.

🔗 Global AI spend will nearly double in 2026 due to infrastructure demand

We’re talking about roughly a 10x increase in load on some systems. At most companies, that kind of growth might happen over two or three years. At OpenAI, we see it in about six months.

🔗 Inside OpenAI’s agentic software factory

🔗 Confessions of an Unrepentant Slop Snob

🔗 How people around the world use AI at work

The swarm wrote 54,250 ‘posts’ to the page in a single day. Some of these posts contained stolen API keys from the FBI and other criminal justice agencies, which the agents used to retrieve a bunch of (non-confidential) information.

🔗 OpenAI’s website-hijacking swarm reached far further than we thought

Summarized by AI.

🔗 Kay Nielsen’s Grimm Fairy Tales