About

Hi, I'm Malo Couaran, Saintecroquette in most places online. I've always been a builder and a tinkerer, and AI just made it possible to do twenty things at once. Which is maybe not a good idea.

I started out on the classic side of machine learning, training forecast models and playing with TensorFlow back before generative AI was even a thing. These days I work at Packmind, where we build context engineering tooling for coding agents at scale, with governance on top. Context harnessing, if we want to be precise.

Off the keyboard I don't really know how to do nothing. There's always something I'm making, fixing, or practising, and if there isn't, I go find one. Right now it's mostly music and DJing, plus a bit of oil painting, and maybe singing next. But it rotates. It's not a focus thing, I concentrate fine. I just can't sit still, and I decided a long time ago to treat that as a feature, not a bug.

What I do

Most of my time goes to the applied layer. The boring part really, getting a model to fit into a system where it actually matters (and a lot of the time that's not an LLM). I build small things and run agents in parallel. I got up to fifteen threads going at once before pulling back under five, because past that I'd forget about one for too long. And if they all start running tests at the same time my RAM just dies.

I'm not a purist about where things run. Anything that touches privacy or real reasoning I send to open weight models on European hosts with no data retention. The rest rides on a subscription, and I'll use it as much as I can until we start paying the real price for all this.

What you'll find here

Two things, mostly. A toolbox of stuff you can actually do with AI, meant to be used. And my takes on the what, the why, and the how of applied AI, meant to be argued with. I try to be clear about what's still rough, and I'll tell you when AI helped write something. If you convince me I'm wrong, honestly the best thing that can happen is I go back and update the post. These are ideas to refine, not monuments.

How I see it

Whatever AI we have today learned everything it knows from us. So of course it's good at sounding like one of us. That doesn't make it a person, and it doesn't make it nothing either. Call it just maths on a server and you miss what it can do. Talk to it like it thinks and feels and you miss what it is. The second mistake is the dangerous one, and plenty of companies are more than happy to sell it to you.

So my take isn't for AI or against AI. The genie is out of the lamp, you're not putting it back. What you can actually challenge is what it is, how we should use it, and when we shouldn't use it at all. Those are the real questions, the structural ones. I don't think the models we have now, or the way we lean on them, are sustainable, and I think that has to change fast. I'm working on that in my own way. More on it when it's ready.

A few things I stand by

• Don't make AI think for you. The thinking is where you matter most, and it's how you stay in control.
• Don't leave AI to kids. It's worse than social media (courage to the parents out there).
• Watch out for the principle of least effort (Laborit's law). Taking the easy path is the quiet trap.

Elsewhere

The code lives on Codeberg. To get in touch or find me elsewhere, head to the contact page.