I think “agent” may finally have a widely enough agreed upon definition to be useful jargon now - “[AI ‘agents’ are] Tools in a loop to achieve a goal… wiring up tools to an LLM in order to achieve goals using those tools in a bounded loop.” // Also, he’s not a fan of the “autonomous” vision, which feels right. // “This category of agent remains science fiction. If your agent strategy is to replace your human staff with some fuzzily defined AI system (most likely a system prompt and a collection of tools under the hood) you’re going to end up sorely disappointed."

Atlassian acquires DX, a developer productivity platform, for $1B - That’s a good match, and a quick exit.

Does anyone really want air hand-dryers around? I mean, they’re loud and they don’t work. You might as well have a sign that says “dry your hands off on your clothes."

This a good, correct framing of the AI project failure stuff. No one really knows what will work and what they’re doing yet. As we learned in the digital transformation craze of the 2010’s, this means failure == learning. And learning is what you need to do a lot of.

More so, this kind of rapid learning, innovation, and sense making is exactly what a platform like Tanzu Platform with Cloud Foundry is excels at, and has a long, proven history of supporting.

You focus on learning new business models, features for your apps, and what works by writing and deploying code instead of building containers and configuring infrastructure.

And, as I like to say, if your organization writes a lot of their own code, there’s a good chance you already have the Tanzu Platform in place, ready to use. Just ask around.

Check it out: “How To Build Agentic AI That Ships," Brian Friedman and Jonathan Eyler-Werve, The New Stack, September 15th, 2025.

Look at this busy lil' fella!

There’s a creeping notion that Claude Code (and OpenAI Codex, I guess) are very useful for things that are not code. In the Obsidian community, where you keep your decades of notes as plain text, markdown files, some people are using Claude Code to do analysis, reformatting, etc. of their notes. I can see that you could do the same thing with the pile of Dungeons & Dragons stuff I have laying around.

This makes sense, really. Source code is just plain text as well.

Doing this kind of thing from your desktop, where you control all of the integrations and data access instead of waiting for a developer to do it for you, feels like it’d be easy. I think the Goose people are figuring out this form-factor for AI apps: focus on building out a tool belt (MCP as plugins) for a general purpose chat app rather than fit-for-purpose apps that use AI in the background. A classic suite versus point-solution situation.

Here is my current theory: the GenAI we start seeing in enterprises won’t be integrated into your existing apps (like online banking), but will show up (still) in the form-factor of chat apps in those apps (those little pop-ups in the lower-right hand corner, except now they have all your context) or tool belt apps like Goose.

That is, AI will not be integrated into existing apps, it will be a new, stand-alone app (or “workload”).

Somewhat related: I’ve been consistently disappointed by Gemini’s integration into Google Suites (we have this at work). Yesterday I asked it to do some simple reformatting of a column in Sheets (convert date formats) and it flat out said “I can’t do that.” I took a screenshot of the dates column, gave it to ChatGPT, and asked it to reformat the dates and output it in a format ready to paste into Sheets. It did it perfectly, instantly. Most of the time when I use AI integrated into an app, I really just what the standard chat interface that can pull from and push back to that app.

That feels like what Claide Code is trying to do. I’ll have to try it out in Claide Not Code style.

Anthropic Economic Index report: Uneven geographic and enterprise AI adoption - 🤖: “Enterprise deployment via Anthropic’s API exposes a different facet: businesses adopt AI programmatically to automate. 77% of API usage is automation-dominant, particularly in coding, debugging, office administration, and recruitment. Surprisingly, firms are not especially price-sensitive; higher-cost tasks see higher adoption if they deliver economic value. Yet complex, high-impact deployments are constrained by context—firms need to restructure data flows and centralize knowledge to fully unlock AI potential. Without this, sophisticated tasks remain underutilized, delaying broader productivity diffusion."

“My phone -and I realise this is mildly absurd – is a constant slight disappointment to me. But I’m from the generation that thought we’d have the world on our wrist.” W.E.

“A man who had over the course of a week absorbed, as by osmosis, the Spirit of the Internet. And that is a foul, foul thing.” a slightly embarrassed announcement