AI engineering
Making AI-assisted development predictable: agent harnesses with fixed phases and quality gates, MCP servers that give agents real context, and a tested toolkit of skills and agents.
What I'm building →What I do
AI, platform engineering & test automation
Platform engineering
Self-service tooling that takes the friction out of environments and releases. Teams claim, reset and release more than 100 test environments themselves, and ship releases and hotfixes in one go.
Test automation
Over eleven years of test strategy, frameworks and test data, from Selenium to Playwright. I teach it too, with hands-on Playwright trainings for UI and API testing.
Prepare for the training →Right now · BMW Financial Services
Making AI predictable
At BMW Financial Services, more than 25 feature teams build on one platform. Besides the platform tooling, I build what lets them use AI in their daily work. The hard part isn't getting an agent to do something; it's getting it to do the same right thing every time.
Fixed phases, clear rules
An AI harness that takes agents through fixed phases instead of letting them roam. Nothing moves on without proof that quality is in order, and it's agreed in advance which review findings block. A run behaves the same on a laptop as on a cloud agent.
Real context, not guesswork
MCP servers that connect agents to the platform and tools teams already use, so they act on real data through well-defined, controlled actions.
A tested toolkit
Reusable GitHub Copilot skills and specialised agents, shared through a skill marketplace, covering analysis and planning through to implementation, testing and review. Every new skill or agent is tested automatically in CI.
Want the short version? Agents, Skills & MCP Servers: a practical 101 →
On stage
Talks & workshops
Mostly at the bi-monthly test automation meetups of deTesters, TestCoders and TechChamps in the Werkspoorkathedraal in Utrecht.
-
Talk
Keep your minions in line: behavioural testing for AI skills
How do you prove your AI skills and agents actually follow orders? Three trust tiers, from deterministic tool-call checks to an LLM judge and baseline comparison, plus variance analysis across runs.
-
Talk · Meetup, Utrecht
Agents, Skills & MCP Servers: a practical 101
How a team of AI agents, skills and MCP servers takes an issue from ticket to tested feature, and what that means for the way we work.
-
Talk
Introduction to Testcontainers
Real Docker containers in your tests, and how Quarkus Dev Services makes integration testing almost effortless.
-
Training
Playwright UI & API training
Two evenings of hands-on UI testing plus an API module, with slides, exercises and the Booker platform as a practice app.