Building my AI operating system · in public

Prompting an LLM is easy. Changing how you work isn't.

I spent 20 years running operations. Over the last two I rebuilt how I work around
an AI operating systemA single AI system you direct, connected to your tools and data. It keeps context and calls its own skills and sub-agents to do real work.What is an AI operating system? →. It runs the real work, not a pile of chatbots.
The prompting was the easy part. The change in how I work was not.
I had to do it myself first. I help you build your own system
and make the new way stick. Then you bring it to your team,
or the teams you run. I'll also tell you when
AI isn't the answer. Often it isn't.

Updates

What's new

Wins, upcoming talks, and what I'm building right now.

Co-presenting The Fractionals Live, Amsterdam
On stage with Dorothy Seguy, to a room of senior fractional operators (CMOs, CTOs, CPOs, founders), on human-led growth and AI-assisted execution. I'll show how AI agents and workflow design help you scale outreach and follow-up, prioritize the right conversations, and cut the manual work, without automating the human side away.
Building for Yasmin at the AI for Good Hackathon
A three-day hackathon at EDGE Stadium in Amsterdam, run by Techleap as part of Amsterdam Tech Week. 48 hours, teams of five, one real person at the centre. Yasmin is 38 and was born with severe cerebral palsy. Today she speaks through an eye-controlled computer with a limited set of phrases. The challenge: build AI tools that let her communicate freely and live more on her own terms. We worked through a five-phase design sprint, from user interviews to a live jury pitch, with Yasmin and her family in the room the whole time. Every decision was tested against a real life, not a persona. The jury included Micky Adriaansens, former Dutch Minister of Economic Affairs, and Yasmin's mother Marla. Our team came in second. We built a kind of Scratch for language: the way MIT's Scratch snaps code blocks together, ours snaps pictograms into sentences and whole stories.
First place at Sprint Zero, the Techleap AI Hub hackathon
70 builders, around 20 teams, two rounds. ABN AMRO's challenge: tell whether the human on a banking video call is real. Instead of another deepfake detector, our team flipped it and built an app that proves a real person is present. It runs alongside any call with no plugins, reading LiDAR depth, rPPG and motion signals plus an ultrasonic fingerprint, scoring trust live and exchanging signed proofs over a secure channel. Demo and pitch built in two hours, judged to first place by a jury including Daniel Gebler (Picnic), Leonard Bukenya (Codam) and Lethabo Motsoaledi (Voyc). The prize: six months of full AI Hub membership for the team, worth €10,000, starting when the hub opens in September.
Building in public

I'm building an AI operating system from scratch

Not a chatbot, an operator. A personal AI operating system that plans, decides, and executes the real work across everything I do. It runs inside Conductor on Claude Code and remembers across sessions, so its reasoning compounds instead of starting from zero. It breaks work into small, self-contained packages — each one opens its own session on its own branch, does a single job, and closes against a written definition of done. And each job goes to the right model: a frontier model to orchestrate, smaller ones to synthesize and search, a local one for the cheap work.

It holds itself to the kind of engineering discipline you'd expect from a team, not a side project: a handful of firm principles, backed by checks that actually run. Nothing ships unless the code compiles, the tests pass, the internal layers stay separate, and no private data can leak into anything public. It even learns — my corrections get turned into permanent improvements, with my sign-off before any of it sticks.

I build the whole thing in the open, including the part most people hide: the first version failed, and I document exactly why. Every decision is recorded with its reasoning, so nothing gets re-litigated and the learning compounds. And the parts that prove themselves I release as free skills, prompts, and patterns you can drop straight into your own setup.

Spinning out into products

The parts that prove themselves become products you can plug straight into your own setup. First up:

Agent dashboard showing what each AI model is doing, what it costs, and how much it reuses, across Claude Code, Codex and local models

Agent dashboard

See what your AI agents are doing, what they cost, and where the money goes — live, across Claude Code, Codex, and local models. It shows how much each one reuses instead of paying for twice, how close you are to your limits, and a short write-up of every session. It runs on your own machine. Nothing leaves it, and there's no service to sign up for.

Paid · coming soon
Work with me

If you want to learn the new way of working
— and you actually want it to stick

I work with operators who are serious about AI and don't want a black box.
Founders running their own company. Fractionals running other people's.
We pick a concrete workflow. I set you up with your own AI operating
system and the thinking behind it. Then I build alongside you
until the new way of executing the workflow sticks.
Done with you, not done for you.

Hourly or project-basedHourly from €175 · projects priced to scope · startup- and solo-operator friendly
Questions & answers

Questions & answers.

A single AI system you direct, instead of a pile of disconnected apps and chatbots. It's connected to your tools and data, keeps context across your work, and calls its own skills and sub-agents to actually get things done. You manage one agent, and it runs the real work: meetings, research, writing, outreach. I'm building mine in public, and teaching others to build their own.

I build AI operating systems. A connected layer that runs a business's real work. I help operators make that way of working stick, so it spreads through their team, or the teams they run. Right now I'm building an AI operating system from scratch, in public.

With you. I set you up with your own AI operating system and the design principles behind it, then help you build your operation one step at a time. Done-with-you, not done-for-you — so you're never dependent on me.

No. I spent 20 years running operations, not writing code. The point is to give you the understanding and the tools to build, at whatever level you're starting from — from simple automation to more complex sub-agents.

No — and I'll tell you when it doesn't. Plenty of bottlenecks are solved better by a simple automation or a process change. I help you find the real bottleneck first, then decide whether AI is the right answer for it.

Tell me what you're building. A short message about where you're stuck is enough to figure out whether and how I can help.