Your AI agent doesn't know why your project is the way it is

A talk with a live demo — and a three-person lab to take it to your own project

Jesús Pérez
AI agents write faster and break more: the telemetry measures it, and the cause is simple — the agent has no way to know what your project is, which decisions govern it, and what it must not touch. A talk with a live terminal demo, and a short lab for anyone who wants to take it to their own project. The approach is not only for software.

Your AI agent doesn’t know why your project is the way it is

And that’s why it breaks it faster than it writes it.

Recent telemetry on AI-agent development is uncomfortable: more code shipped, yes, but also far more rework, more incidents per change, and more changes merged without a single review — and strong engineering foundations do not protect you from it. The root cause isn’t a bad model. It’s that the model works blind: your project knows what is essential and what is accidental, which decision was made and against which rejected alternative, what must not be touched — and none of that lives anywhere the agent can query.

What the talk is

A session with a live terminal demo. We start from an empty project, declare an architectural decision as a checkable artifact, and watch validation reject the change that violates it — before production, not in the postmortem. No frameworks, nothing exotic to install.

It is not a recipe. There is no A→B procedure that produces coherence by being followed. What it teaches is a substrate for holding decisions without pretending they were resolved, and for letting a machine query them.

And it isn’t only a programmers’ problem

The same gap shows up in the infrastructure you inherited whose current shape nobody can explain; in an authored Work cut by hand for the book, the web and the EPUB until the versions stop matching; and in your own professional trajectory, scattered across a CV, some slides, and decisions only you remember. In all four, the same thing is missing: a queryable declaration of what this is and what governs it.

And after: a lab to take it home

For anyone who wants to move from watching to doing, there’s a short lab: one session, three people, each with a project of their own. We model live on your material — no slides. You leave with a minimal model running on a real project of yours, and with that context available to your agent too.

It is not a course or a sale: all the material is open source and you can use it without me.

In preparation. Date, venue and registration will be announced here once confirmed. To be notified, write to jpl@jesusperez.pro.

Was this useful? Rate it
Got something to add? Tell me what you think, what you'd suggest, or whether we should keep exploring this topic.
· reads

We use cookies to help this site function, understand service usage, and support marketing efforts. Cookie Policy for more info.