Rogier Muller, AI coding agent trainer in Amsterdam

Who Rogier Muller is

Rogier Muller is based in Amsterdam, works as a CTO at a fintech company, and runs Harness Institute with Vasilis Tsolis, who is in Athens and a partner at a legal AI company. Their workshops focus on scoped agent tasks, verification and review.

The relevant part is the day job. Both trainers ship production code with these tools in companies with real compliance constraints and real on-call rotations. The material comes out of that, not out of a curriculum design exercise.

What the training covers

The tools are Claude Code, Cursor, and Codex. A scoping call establishes the team’s current tools, experience and review practices before choosing exercises.

• How the agent loop works, so people stop treating output as either magic or garbage and start predicting when it will be wrong
• Repository setup: instruction files, test commands, what to exclude, what to automate in CI so the agent has something to check itself against
• Review discipline for agent-authored diffs, which is a different skill from reviewing a colleague's work
• Where to draw the line, meaning the categories of change your team should not hand to an agent yet

Sessions are hands-on in the team's own repository. The agreed preparation determines which repository and data can be used. The exercises use your build, your tests, and your actual mess, because the friction that stops adoption is always specific to the codebase.

How a session runs

A typical day starts with the team's own repository open on the screen and one real ticket on the table. We set up the instruction file together, argue about the test command, and run the first agent task in front of everyone. Then the exercises get harder: a change with an ambiguous spec, a diff containing a deliberate mistake, a task that should be refused. People leave with a written set of team rules they wrote themselves, and can test on subsequent work.

What this is not

Ask which assessment and completion evidence is included in the specific offer; those entitlements depend on the programme. It is not a promise that your team will ship twice as fast, and anyone who gives you that number in a sales call is guessing. It also will not fix a codebase with no tests. Agents amplify whatever verification you already have, so a team without a working test suite gets faster production of unverified changes, which is a worse position than before.

If that describes your repository, discuss verification readiness during scoping. A focused session may need to establish a test harness before tackling a broader agent workflow.

Getting in touch

The fastest path is a short call to work out whether the training fits where your team actually is. Come with one concrete thing that went wrong: an agent diff that slipped through review, a developer who has quietly stopped using the tool, an argument about whether to allow it on a sensitive repository. Those conversations are more useful than a feature list, and they tell us whether we are the right people for the problem.

If you want help putting this into practice, talk to us.