Here’s one last little AI post about Ponos because we’re all saturated with AI content, even though it’s perfectly normal — it’s completely upending our world.

A progress report — or maybe a final one (I have no idea) — on Ponos then. My latest thoughts. In no particular order, for myself, and for those still reading.

Mental Load

AI absorbs, devours. I have a good ability to enter a flow state, to reach a focus that puts me in an almost trance-like state of intensive concentration on a subject, especially writing. With AI this intensifies further. You open one claude in the console, then a second, a third, and while it works you jump to another project and there too, two or three consoles. Then comes cognitive overload. It’s my brain that reaches its limits. It’s smoking.

Language

AI is powerful with everything related to language: code, text, translation, etc., etc. — don’t forget that.

The End of User Stories

User stories were there to properly orchestrate the build. To support communication with the people doing the building. Now it’s AI that builds, and we really don’t need user stories anymore — or only in a test format directly to frame the build: gherkin or ATDD (acceptance test driven development).

Functional Just in Time

For that matter, I don’t think we need specifications anymore either (not to be confused with documentation). I mean “big specifications.” We always said a backlog beyond 30 items is nonsense. That’s even more true now.

Either you’re tackling a well-known domain (like: build me a dating site, build me an online betting site, build me a CRM for my sales team, build me an open-source European LinkedIn, etc.), and the territory is well-known to AI. So you simply tell it to go ahead. Then you work through iterations. If it’s regulatory, even less need to write specifications — just follow the regulation.

Or it’s an emerging product, and in that case, well, you iterate.

For me: no more specs, no more backlog. Functional “just in time.”

Software Vendors Are in for a Shakeup

From all sides, but some should be particularly worried. They know it: stock market drops are the writing on the wall. In my personal case: subscription to a Learning Management System for the Projet Winston masterclasses. €1,300 per year. It took me two weeks to build my own, better because it fits my needs perfectly. It costs me €300 per year for hosting, and let’s say occasionally 1 month of Claude Max (€100).

And now I’m seeing “AI factory” after “AI factory” popping up to replace everything. Everything. Humans. Software. Etc.

Level of Abstraction

AI is a new level of abstraction — we no longer look at the code, and that’s OK.

Too Much Anthropomorphism with Agents: A Mistake in My Opinion

I see many examples of agents built on an anthropomorphic model: the infrastructure manager agent, the quality manager agent, etc. It comes from a good intention to help explain agents, but it locks you into an error. Often a “manager,” a human, doesn’t have a single well-defined function. But I think that’s precisely what agents should have. And saying: an agent that specifically monitors all front-end performance issues is better than mapping it to a human role.

The Rejection of AI

I observe, particularly on Mastodon, a blatant denial of AI’s results — but not only on Mastodon.

I sincerely hope all these people are right. I sadly think they are wrong.

It’s like the installation of the telegraph: in 1858, people didn’t believe the news because “it arrived too fast.”

Political Action Needed

What I also don’t understand is the absence of political action on the social front. We can see a wall approaching very fast: AI will replace many jobs. And yes, probably new jobs will emerge, but in the gap there will be a shock. So two options: either the time for something like universal basic income has clearly come, or we’ll slide into a toxic master/slave relationship. Recent history (and ancient history) doesn’t make me optimistic.

Especially since AI, for now, is powerful in the hands of people with broad, systemic approaches — entrepreneurs — and not many other profiles (who have their own qualities but are in much more direct competition with the machine).

Naturally, all these points about AI will be called into question every six months.

And Ponos

Well, as expected, it’s hitting the reach wall. So like Cthulhu, it may sleep at the bottom of the seas waiting for a) an evolution of laws on connections between services to propagate accounts, b) an American-European political clash, etc. And like Cthulhu, it can wait as long as it takes.