Notes from the work, filed by the question they answer.
Essays on AI architecture, honest billing, and the craft of building software as a solo operator — written between client work, published when they’re true.
11 essays · three questions · answered in public
Start readingCan one operator carry work this size?
The question every prospect has and almost nobody asks out loud. These answer it with the way the practice is actually built — including the part where we tried it the other way and it didn’t hold.
Six months into running my consultancy, I had hired six people.
— from “Why I unhired five people.”Read the pieceCapacity versus continuity.
Two business models, two unit economics, two different markets. Why agencies sell what's easy to scale — and why operators sell what's easier to defend.
Practice5 minOn shipping: a discipline, not a verb.
What production management taught me about shipping — calendar, ops, vendors, the whole chain in one head — and why software still gets the word wrong.
How we work — one accountable operator, no hand-off layer, and a deliberate ceiling on how much runs at once.
How will I see what I’m paying for?
Every engagement runs through a portal with the hours on it, and every invoice reconciles to a published rate. These explain why we treat that as the product rather than a courtesy.
Most agency taglines are aspiration.
— from “Built with care. Billed with clarity: a thesis.”Read the pieceThe window you give the client.
Most agencies ship a status email; I ship a client portal. Same hours, same work — but one costs your trust, the other an afternoon of engineering.
Engineering7 minThe portal as first principle.
Every client engagement starts by provisioning a portal — before the contract, before kickoff, before the first line of code. Why infrastructure comes first.
Engineering12 minWhat I learned from one million accounts.
Two years as tech lead at a fintech past a million accounts. The lesson wasn't scale — it was the gap between code that runs and code that earns trust.
The client platform — hours logged to the tenth, threaded requests, and invoices that reconcile line for line against the public rate card.
How do we use AI without it going wrong?
Evaluation before agents, every model named, and the regulation closing in around all of it — written for teams who have to ship AI into something that actually matters.
For ten years I wrote unit tests.
— from “Eval harnesses are the new test suite.”Read the pieceNaming the AI agents you use.
Most agencies hide their AI use behind handwaving. Naming each agent — its model, its job, its honest limitations — turns out to be a competitive moat.
AI11 minFrom GDPR to the EU AI Act.
IT firms that built GDPR readiness in 2015–2018 billed premium fees for years. The EU AI Act opens the same window — and the bridge to AI governance is short.
AI Platforms and Consultancy — evaluation-first builds with a cite-or-abstain contract, and EU AI Act readiness you can act on.
Recent notes.
Published since the questions above were last drawn up.
Everything, by date
The same 11 essays, newest first — for when you already know what you’re looking for.
- What custom software actually costs a small business.Practice
- From GDPR to the EU AI Act.AI
- Why I unhired five people.Practice
- Built with care. Billed with clarity: a thesis.Practice
- Eval harnesses are the new test suite.AI
- The window you give the client.Engineering
- On shipping: a discipline, not a verb.Practice
- What I learned from one million accounts.Engineering
- Naming the AI agents you use.AI
- The portal as first principle.Engineering
- Capacity versus continuity.Practice