Skip to content
← Back to home

About us

Prophet is two people: Bautista Filippini and Francesco Maddonni. We both write code and we both talk to clients. There's no layer of account managers between whoever decides and whoever builds, and that's why we can walk into a company on a Monday and leave a task agentized that same week.

We used to be called The Winlab. We became Prophet in May 2026, when what we did stopped being marketing and became this: agentizing companies.

What we do

We agentize companies. It's the only thing we do and the only thing we talk about.

We take what the company already does, with the systems it already uses, and put a layer on top that turns every task into a tool for agents. A process, which is several tasks chained together, is packaged as a skill. The skills live in the company's own plugin or MCP: its connector. It's teaching an agent the company's language.

The MCP belongs to the company, not to a model: it works with Claude, with ChatGPT or with whichever agent they choose. And the company's data (ERP, spreadsheets, databases, email, documents) becomes accessible to agents through MCP, queried where it lives, with nothing migrated.

How we work

Step by step, small processes first, and each one in two phases:

  1. Guide the agent. The agent does the task and a person at the company guides and approves it.
  2. Every correction becomes a rule.
  3. Let it run alone. With that feedback, the task runs on its own, with agents that live on
  4. remote servers and work around the clock.

And three rules we don't negotiate:

What already works, stays. We don't turn anyone's systems or processes upside down. We understand what the company does today, optimize it where it pays off and put the layer on top.

Task first, code second. Before building anything you have to be able to say which task is going to be agentized and which business number has to move. If an agent isn't what fixes the problem, we say so, and we don't charge for pretending otherwise.

One finished beats five half-done. We'd rather have one task genuinely agentized in week two than a six-month plan nobody gets to see.

Our motto sums up why we do any of this: work less, achieve more, be happier. So the team works less, gets more done, and is better off.

Who we work with

Companies across very different sectors. Two examples: an agency that has its admin manager's tasks as skills in a plugin of its own, connectable to any AI, and a law firm where we replaced the legacy system with a new one with its MCP on top (court notices, payments and garnishments).

We don't publish client names. It does none of them any good for their competitive edge to show up on their supplier's website, and we think that's fair enough.

One Brain

We also make One Brain, the memory agents use: what the team decides and corrects is saved with its author and date, and the agents check it before acting.