About

I'm Luc Bocahut. I design and build software systems, and I have been doing it long enough to care most about one property: whether the thing still works, unattended, years later. The trading and risk systems I built for a proprietary desk in the 2000s are still in production more than twenty years on, and I still occasionally maintain them. That experience shaped everything I have built since.

How I think

A few principles keep resurfacing in my work, and I hold them as mottos because they earn their keep.

Ingest cheap, process lazy, enrich on demand. Most data is never looked at. Systems that do expensive work up front pay for everything; systems that defer work pay only for what gets used.

Explore agentic, build deterministic. Language models are extraordinary for finding out what a system should do, and unreliable as the system itself. I use agents to explore, then freeze what works into deterministic code. The corollary for products: the application owns the semantics, the model only translates intent into the application's language.

Live in production. Long feedback loops hide problems. I would rather ship small, watch a real system behave, and correct quickly than rehearse in staging. It takes judgment about what can safely move fast, and that judgment is most of the job.

I remove more than I add. The most useful thing I do in most projects is ask what can be deleted, consolidated, or made someone else's problem. Variation is cost: when everything shares one framework, one deployment pipeline, one set of conventions, both humans and machines stop improvising and start completing patterns.

What I enjoy building

Systems that run lives and businesses, not demos. The common thread across very different domains, trading infrastructure, an ERP for a bespoke tailor, excavation tracking for a metro project, a personal agent OS, is that each one is the operational backbone for someone's real work, built full-stack from the infrastructure to the interface, and expected to keep working after I leave the room.

The project that currently holds most of that ambition is Anton, a personal agent OS that runs my family's daily logistics in production: a custom TypeScript runtime, sandboxed skills, layered memory, a policy engine evaluating every action, and local inference on hardware in my home. It is also an old idea finally buildable. In 2015 I sketched a personal digital twin concept called Bots for Humanity as my MBA capstone. The technology took a decade to catch up; the design goals barely changed.

I like working close to the people who use what I build, pair programming with the engineers I work with, and writing things down properly. If any of this resonates, the ideas are the best place to see how I think, and the work is the evidence.