Service

AI agents in production

I put AI agents to work on real company processes: triage, support, data analysis and internal routines. I use Claude and other models, n8n and custom code, with a record of every decision, cost control and handover to a person from day one.

Typical timeline: Pilot in production in 2 to 4 weeks, on a real process and with the metric defined before we start

Who it is for

Companies that already tried AI in a chat window and want something that runs on its own inside the process, every day, with an owner, predictable cost and a history of what was done. Also founders and CTOs who need to take a prototype all the way to production.

Problems it solves

  • An AI prototype that works in the demo and breaks with real customers.
  • Model costs that nobody tracks and only show up on the invoice.
  • An agent that decides without leaving a trace, and nobody knows why it did what it did.
  • A manual task that eats the team's week and never makes the priority list.

What I deliver

  • An agent designed for one specific process, with the tools and limits of what it may do.
  • Orchestration in n8n or in code (TypeScript and Python), depending on the case.
  • A record of every conversation and decision, and cost measured per task.
  • A handover rule and an alert when something is out of the ordinary.
  • Memory and lookup of the company's data when the process needs it (database and semantic search).

How it works

  1. Diagnosis

    One conversation to pick the process that eats the most hours. No cost, no commitment.

  2. Pilot in production

    In 2 to 4 weeks, an agent running on a real process, with the metric defined before we start.

  3. Scale

    Did it work? I replicate it across the next processes. The operation grows, the team does not need to double.

Typical timeline

Pilot in production in 2 to 4 weeks, on a real process and with the metric defined before we start.

Frequently asked questions

Which AI models do you use?

Whatever fits the problem and the cost. I have used Claude, OpenAI and Gemini, orchestrated with n8n and code in TypeScript and Python. The model is a part you can swap: the agent is designed not to be locked into one vendor.

How do you stop the agent from doing something wrong?

I limit what it can do by giving it only the tools the process needs. I record every decision, keep a person in the loop for risky cases and start with small volume before opening it to everyone.

Where do we start?

With a free diagnosis, one conversation, to choose the process that eats the most hours. Then comes a pilot of 2 to 4 weeks with the metric defined beforehand. If it works, we take it to the next processes.

Let's talk?

The diagnosis is a conversation, free and with no commitment. You leave with the list of processes that eat the most hours of your team.

Talk on WhatsApp

Other services

Want this running in your operation?

Tell me in one sentence. I reply with what can be done, how long it takes and what it costs.

Tell me the problem