Writing
Notes on building, governing, and running AI agents.
Practical pieces for the people putting agents into production: how multi-agent workflows actually fit together, what it takes to keep them EU-resident, and why memory is the thing that makes agents useful past the first session.
Why your AI agents should keep a record of their mistakes
A durable, written record of past errors turns a capable AI agent into a reliable one. Here is how a lessons log compounds and beats raw chat history.
Why AI agents need memory: context that compounds vs session resets
The difference between stateless chat and agents with persistent memory, why it changes what agents can do, and the payoff of context that compounds over time.
Not every feature needs its own agent
A restraint principle for agentic architecture: fold most capabilities into existing flows, and reserve standalone agents for work that truly needs one.
Running AI agents in the EU: data residency, GDPR, and the EU AI Act
A practical guide to keeping AI agent workloads EU-resident: what GDPR means for model traffic, what the EU AI Act asks of deployers, and how EU hosting helps.
Let your AI agents open pull requests, not commits
The safest way to put AI to work on a codebase: agents work on a branch, open a pull request, review their own diff, and a human merges. Here is why.
Multi-agent AI workflows, explained
What a multi-agent AI workflow is, how it differs from a chatbot, and when you actually need coordination and delegation, with a concrete coder-reviewer example.