About me
I have spent 25 years in IT doing every job between the strategy and the server. The question I work on now is what has to be true inside an organisation for AI to run safely in production.
I started as a .NET and SharePoint architect, building for Kadaster, CIZ, Wageningen UR, Malmberg and a dozen other organisations. Along the way I co-created a content management system, built a complete club platform from private cloud to kiosk app, and published an open AngularJS library. I learned early that the interesting problems are never the code; they are the organisation around it.
From 2016 I moved into enterprise architecture and cloud transformation: Azure environments, a data platform built as the path to machine learning, business continuity, a laboratory system migration without a minute of downtime. At BuyBay, as Manager IT & Architecture and then Strategic Technology Leader, I rebased IT for a scale-up, led Azure and Microsoft 365 adoption, and replaced the warehouse and transport systems. At MultiBel, as Head of IT, I introduced AI-assisted working across departments and took the security management system towards ISO 27001.
Today I am Strategic Product Manager Information & AI at PinkRoccade Local Government, also known as Head of Product Data & AI, after roles as Development Manager and Manager Customer Care on the management team. I lead the move from a portfolio of loose products to one platform, and the development of the open-source ZGW specifications for Common Ground. Before that, at MultiBel, I introduced agents, skills, MCP servers and a company memory into daily work, and the quality-assurance intake went from three to four hours to thirty minutes per inquiry. My thesis is that AI for municipalities only works on top of a complete, governed platform, and on my own time I built a prototype of one to test it: policy enforced outside the prompt, human approval where the stakes require it, evidence that cannot be edited afterwards.
I build with AI agents every week, under written rules about what they may never do. That is where my point of view comes from: not from reading about AI governance, but from being the person who has to answer for what the agents did.
How I work
Evidence over trust
If a system cannot show afterwards which rules applied and who approved what, it is not ready for a decision that affects a person. I design for the audit before the feature.
Grade every claim
Concept, prototype, pilot, deployed, measured. Each piece of work on this site carries its grade, and I expect the same of any AI plan put in front of me.
Humans keep the keys
Agents can draft, retrieve, prepare and propose. A named person takes the decision that changes someone’s rights, and the override rate tells you whether that oversight is real.
Build it to understand it
I do not trust an architecture I have not tried to break. Most of what I know about AI governance I learned by building a platform and watching where it failed.
The purpose selects the tool
I have a Microsoft background and a preference, and I am agnostic in practice: sovereignty, cloud, stack and language are angles with trade-offs, not positions. Some languages and technologies fit the AI era better than others, and the outcome decides, not the habit.
What I am into right now
- Governed agent runtimes: what a trust root looks like when no model provider or cloud is allowed to be it
- The EU AI Act in practice: turning Article 14 human oversight into something an engineer can implement
- Which programming languages fit the AI era, with predictions I will grade in public in May 2027
- Common Ground and ZGW as the platform base for municipal AI, and what the layer above them needs
Beyond work
I live in Rosmalen with my family, and we are building a house, so most evenings involve a floor plan, a 3D printer running in the corner and a home-automation problem I created myself. Technology is the job and the hobby. The rest is people: friends at the table, and three years on the board of a tennis club, which taught me more about stakeholder management than any project did.
