Piste / advanced

AI agents in production

AI agents in production on WhatsApp, on the phone, in email and inside assistants over MCP. I've shipped them for startups, SMEs and enterprises, often on my own.

When it fits

A process jams every day, or a prototype only works in the demo, with someone next to it steering.

How I work

Scope first: what the agent has to finish, what it needs to know, when it stops and hands over to a person. Then the harness: typed tools, permissions in code, evals on real cases, observability. The model is the last thing I pick.

Contractor for an MVP, partner for a product, fractional for a team. Same method.

What I need

  • Real examples, ugly ones included.
  • Someone who knows the domain and answers questions.
  • Access to the real systems.

What you get

An agent in production: you know when it gets something wrong, and it can only act where it's allowed to.

We can sort out the rest in a 30-minute call.