How I work

1. Understand the process and the result you want. I talk to your team and see how the work is really done: your tools, your data, the exceptions nobody wrote down.

2. Pick the right task. I map your systems and flows and pick the task worth handing to AI. The same map shows where the risk is under the AI Act, GDPR and security.

3. Agree the scope and the acceptance criteria. We put in writing what gets built and what counts as done.

4. Build and integrate. I build what the task needs on your real data: an AI agent, an automation or software. It connects to the tools your team already uses.

5. Validate it with you. Before the system runs on its own, we check together that it meets what we agreed.

6. Maintain and measure. The build has an end date. Then I maintain it, with the scope each project needs, and every month you get a report on results and cost.

Until now, buying software meant buying a tool. Someone on your team had to learn it, feed it and check what it said. If nobody did, the tool sat unused.

Now AI can do that work end to end: qualify incoming leads, write the report or check the case file. That is how I deliver it: you get the work done.

This model is called Service-as-Software. Foundation Capital named it in 2024. In March 2026 Julien Bek, a partner at Sequoia, took it further in “Services: The New Software”.

Sequoia put it in one line: “A copilot sells the tool. An autopilot sells the work.”

An AI system that works in a demo is useless if it does not understand how your company works.

So I do not work from the outside. I work the way a forward deployed engineer does: with your team, I get to know how the work is really done and build the solution inside your systems, until it works.

The term comes from Palantir: engineers who work inside a single client, build whatever that client needs and are judged by its result.

That is how I work inside Sesame HR too, where every team is an internal client.

Before building, we put in writing what counts as done and how it is checked. For example, what a contact must have to count as a qualified lead, or which questions a report must answer.

With that standard we validate the system with you before it runs on its own. And the same standard is measured every month.

Each proposal separates three things.

The fees: building the system and integrating it into yours.

Third-party costs: the AI models and tools the system uses. AI is not free to run, and you see these costs in the monthly report.

Maintenance, with the scope the project needs.

Prices and conditions are set in writing in each proposal, before starting.

Security comes first: keys outside the code, only the permissions each task needs, and nothing from outside is treated as an order. What arrives in an email or a web page is information, not instructions.

GDPR and the AI Act are part of the design from the start: which data is used and where it travels.

Automatic tests and usage limits. Every change passes checks before it goes live.

Everything is logged: what each agent does, why, and what it costs. Every piece is documented, tested and connected to the others. What no longer serves is retired.

Repetitive tasks whose result can be checked: qualify incoming leads, write the report or check the case file.

Work the company already pays someone else to do, or that its team does. There is a clear idea of what done looks like.

Low liability. Where the law asks for a professional's signature, a person signs.

Not everything needs an agent. Sometimes an automation or software is enough. Other times, a rule and a person in charge.