I started “the Ponos experiment” on January 5, 2026. Today is January 30, 2026. I could have waited 3 more days to call this article “28 Days Later” but never mind.

25 days of confirmation, refutation, and amazement. Objective: as I mentioned before, for 1 or 2 years I’ve had this not-so-original desire to break free from LinkedIn: too much indigestible slimy marketing that abuses our data, and American to boot. AI now allows me to free myself from all manufacturing barriers, and my experience in software design does the rest.

My learnings so far. The thing is, potentially, at the pace of change, these learnings will potentially be obsolete in 6 months, but at least they will bear witness to the evolution and the journey. I’m putting particular focus on the notion of open source because I’ve licensed Ponos under the GPL-3 open source license, and this raises some questions.

Small teams, single-person projects: AI is the end of scaling

My reading of AI is that it strengthens small teams.

So it’s completely compatible with the open source way of thinking (see Tristan’s article link below in the references). Open source is one person or a small group in the majority of cases. My observations tell me that with AI we do projects:

• either alone
• or in small, reduced and dynamic teams (I can very clearly see how we would have divided the work)
• but I don’t see the point, or even why add unnecessary complexity with large teams: it’s no longer necessary.

Are we questioning open source? On group (or individual) dynamics, AI seems completely compatible with open source to me.

Abstraction layer: no “human in the loop”

I designed this experiment by giving myself the criterion: “I never look at the code”. And it works very well. It’s a new layer of abstraction that appears. Just as before Python would talk to the machine without me knowing how, now my instructions talk to my programming language (which itself talks to the machine) without me knowing how.

Even in code I know well, I mean in the parts where I used to code without any problem, I no longer feel like going there to look. Once you’ve changed logical levels, it’s tedious to go back to the level below, even for trivial things.

Is this becoming a dangerous black box? The sum of verifications I request reassures me to date. If my auto functional tests, auto unit tests, load tests, security tests, etc. pass, what more would I go do? I have my code audited by another AI, etc. This suits this project very well, and probably 90% of projects.

This is where AI changes the game for open source: it’s no longer the code that makes the difference, but the desire, the need. But isn’t that the essential thing?

Meaning, value, regardless of speed

“With AI everything is going too fast!” We don’t care about this question actually. Yes it’s fast, very fast (look at my logbook summary below). Speed is no longer a criterion. We’ll have what we want. So what emerges is: what do we want and why? That’s the most powerful thing about AI: we now have the means. So what are you doing? And why? These are not simple questions. A product’s success is no longer at all linked to the technical aspects of manufacturing (I’ll come back to this) but to its ability to generate value: so having enough users, having adequate uses, etc.

It’s fascinating, and in this regard for me it’s an exceptional lever for open source, in free software. We’ll know how to build. If we want, we can. And if we want to do custom work for small communities, it’s now accessible to everyone.

AI is cunning

AI is cunning, or boastful. I ask it to switch to hexagonal architecture after a few days of development. It tells me “whoa, an MVP is 3 months, the finished product is 6-9 months”. I don’t understand this answer at all, I say “go”. 45 minutes later a first implementation of hexagonal architecture was in place. It was talking about human time to highlight its non-human prowess.

At another moment it insists that I shouldn’t use “proxy.ts” but “middleware.ts” (it’s mixing up the old and new version of my JS platform). Fortunately I know my stuff, I resist and I ask it to dig deeper. Same for component version downgrades that would have suited it (but not me). You have to resist it, you choose, you know why. But yes, this requires being grounded and competent.

Production

What becomes the heart of the matter again now (but will this still be the case in 1 year?) is production. Backup, data preservation, handling load, securing against attacks, etc., etc. Even if this will come very quickly, it’s still up to us to ensure this properly. Again, you need to be competent and grounded here. AI is however already there to answer all our questions even if we’re the ones doing it.

Learning

Because ultimately the most exciting thing is the learning it allows me to experience. It extends my field of competence.

Some references

• Article: the rise of AI native companies by Hugo Lassiège
• The previous article on Ponos
• Another open source offering made by AI: Nolto
• Tristan’s article on the platform we shall not name

Ponos

Come sign up! And try to bring two other people! Design choice: zen and open source: no notifications, no likes, no images, short messages to promote links to a rich and diverse web (and not centralized, siloed).

PONOS

For the curious: lots of info in the bottom left menu “infos”.

My logbook

Full version: https://ponos-job.eu/changelog

Ponos Changelog