I wrote code for 25 years. The last 138,000 lines, I didn't.
I help companies integrate AI reliably into their own software development.
AI tools increasingly take over the implementation. The value moves to where it stays demanding: describing precisely what should be built and why — and structuring knowledge so machines can work with it reliably. That is my field.
Software development is changing fundamentally. AI tools take over the pure implementation — faster, cheaper, and increasingly in teams without classical programmers. What remains is the genuinely hard part: clarifying what should be built, and why. And capturing domain knowledge so that AI assistants work with it reliably instead of guessing.
I built software for a quarter of a century — and over the past few years I rebuilt my own way of working from the ground up: I write specifications, schemas and knowledge structures; agents write the code. That experience is what I pass on — to teams with their own development as much as to small companies that want to become independent of their agency.
AI integration into development
Helping teams adopt AI tools productively and reliably in their software development — including teams without a classical programming background.
Specification & domain knowledge
Preparing domain and system knowledge so AI assistants can genuinely work with it. Structure instead of prose — verifiable, versioned, kept current.
AI-assisted implementation
Architecture, toolchains, reviews: modern, heavily AI-assisted development — grounded in more than two decades of full-stack practice.
Not every company has developers — many depend on their agency for every small change. For exactly this case I am currently developing a method: rebuilding or migrating small web projects AI-assisted, with everything in your own hands from day one. By the end, you handle most everyday changes yourself — and know exactly when to call me.
So that nothing breaks along the way, the guardrails are built in from the start: protected version states, automated checks before anything goes live, every state recoverable. And because everything runs on your own accounts, the project belongs to you from day one — including a structured knowledge base your AI assistant can work with reliably.
Memstead — living proof
Memstead is my open-source project: a durable, typed memory for AI agents. Knowledge lives as plain markdown in a git repository — readable by humans, diffable in review, with no database and no vendor lock-in. Agents read and write it through a schema you control, enforced on every single write.
Above all, Memstead is my demonstration piece: this is what software development looks like when you run it consistently AI-assisted — with clean specification, enforced schemas, and a human at the helm. And the proof is not a claim: every line, every commit, every decision is publicly verifiable in the repository.
Incidentally, this website was built the same way — specified and reviewed by me, written by AI agents.
Let's talk.
My consulting practice is still in formation — until then I'm happy to connect and trade notes: on AI-assisted development, on Memstead, or on hands-on experience. Write me — short and informal is perfectly fine.