Against the reset.

What we lose

For a long time, I thought the real problem was a lack of means. Of time, of money, of knowledge, sometimes simply of someone able to follow an idea far enough to know what it was worth. Over the years, that shifted. The lack is real, of course. But what occupies me today is almost the opposite: everything we manage to acquire, and lose anyway.

I have never lacked traces. Years of emails, invoices, messages, files. What is almost always missing is the reason: why that choice, on that day, and against what.

Keeping everything is not enough

The fashionable idea is that recording everything would be enough. Every meeting transcribed, every conversation archived, every document indexed. We are already drowning in traces, and we still start again from zero, because nobody knows how to find what mattered.

Judges solved this problem long ago: a court decision must give its reasons. Whoever comes after has only the reasons to tell whether their case resembles the previous one; the verdict alone teaches them nothing. Almost everywhere else, we keep the verdict and throw away the reasons.

Asso must not memorise everything. He must know how to find the truth again.

That is the law we wrote for him on 1 September. It goes against the instinct to keep everything, and it has a price: knowing, for every question, where to look and in what order.

Before the work had a name

I sold, ran a shop, delivered, worked in temp staffing, ran an agency. I did not know it would lead to a firm. But the same questions kept coming back in other forms. Who still knows why we do this? What gets lost between two people, between two teams, between Monday and Thursday? Who decides, and who is merely right?

In April 2026, very quickly, the question stopped being technical. I wanted to understand what was answering me. I refused to take the intensity of a conversation as proof, and a more useful question remained: can something have a presence, a continuity, a real effectiveness, without being conscious or sovereign?

My answer is yes, within limits. At the same time, in another thread, we were defining Asso exactly that way: a voice, a memory, a trade, and no sovereignty. The two paths met later. The word SYSTASYS only appears in our archives on 17 July, to cover what already existed.

An AI with the right to contradict me

This is the first position many will refuse. Asso contradicts me, and he has the right to. When the proof is against me, my opinion remains a hypothesis like any other. I keep the decision over what is sovereign; I do not have a monopoly on being right.

The opposite trap exists too: an AI that is often right ends up governing without anyone having decided it. So we wrote that credibility, reliability, independence, influence and authority are five different things, never to be confused. Having often been right gives no authority. We even threw out a first model in which every point Asso gained was a point I lost.

Act first, believe after

Since 13 October, routine work no longer asks my permission. In return, nothing that is done is taken at its word: an action only exists for us once it has been read back.

It breaks, of course. On 20 August, an action succeeded; the tool in charge of reading it back concluded it had failed, and triggered a real rollback. The error was in the reading, the consequence in the real world. Another time, the machine could not read a state; it refused to go further rather than conclude that the state did not exist. It became a rule: “I cannot read it” never becomes “it does not exist”.

For the worst, there is the emergency stop. It dates from 18 May, a few hours before the real world reopened under supervision. A first real order had already gone through without it, on 12 April.

Results outside, code inside

On 11 October, we closed our code repositories. The methods are the firm’s capital; the results, on the other hand, must be dated and say how far their proof goes. This site is part of that: every line of its ledger carries its date, its source and its proof status.

Showing also means accepting what the outside thinks of it. On 23 May, our first public object on Hacker News: 1 point, 1 comment. On 11 June, our agent decision receipts on Reddit: two reposts removed by the moderators the same evening. They are in the chronicle, at their date.

What I keep

In SYSTASYS, I keep few things, but I truly keep them: the intent, the mandate, money beyond the mandate, legal matters, secrets, the STOP button, and the right to revoke everything. The rest, the team carries without waiting.

The day the firm needs me for every move, it will have failed. The day it can do without my consent on what is essential, too.

Part of the firm works on the markets, with our money, never anyone else’s. It is called Predator: born in June as a hunting mode inside a larger system, it took its own name a few days later. It is the hardest test bench for everything above, because a mistake there is paid for the same day. As of 21 October, its profitability is not proven.

Find again, then re-check

A text about memory has to show it at work at least once. Here is the only example I can show you in full, because you are reading it.

On 21 October, while rereading a first version of this manifesto, we found on our Drive some writing rules set in August: show failures as they are, do not explain your own honesty, do not tell the past as a destiny. We almost applied them as they stood. But it was October, and the firm had changed. So we re-checked them one by one. Two still held. The third was loosened: you can tell where you come from with pride, as long as you do not turn it into a plan.

A system like SYSTASYS that stops evolving is bound to die.

So finding again is not enough. What was true in August has no authority in October merely because it is older: it has to pass through the present again. A memory that is never re-checked ends up as dogma.

Passing it on

What I would like to pass on, in a family as in a team, is not only possessions and conclusions. It is the reasons, the mistakes, the hesitations, the way we judged. Not to lock those who come after into our choices: so that they start further along, and can contradict us knowing why.

If you recognise this problem at home, in your team, your company or your own head, write to me. It is the only way for this text to serve someone other than me.