Start here
Clear models, architecture decisions, and real e-commerce workflows. One essay a week. Everything here is free.
If you read one thing
Eight Questions on AgentsWhat each component solves, what state it owns, and what breaks when you remove it.
It names the eight parts a working agent is made of, what state each one owns, and what breaks when you remove it. Everything else on this site assumes that vocabulary.
What an agent actually is once you take the demo apart: a model proposing, and ordinary software deciding what may run.
How an AI suggestion becomes a real-world action—and how permission checks, approval gates, and recovery logic keep a helpful agent from making an expensive mistake.
It can look like forgetting. Usually, it is something more ordinary: the information was kept in the wrong place — or it was never kept at all.
One GMV report, built three ways — and a four-layer model for knowing which layer should change.
Where the judgment belongs, what to freeze into a workflow, and who is allowed to act.
Most confusion in agent architecture begins when one system is asked to decide when work starts, how it proceeds, and when it is safe to act.
A practical framework for separating reusable capabilities from orchestration logic—and knowing when to use each.
System Prompt, Skill, Workflow, Cron, and MCP are not rivals. They are one stack.
Most of what is here belongs to one project: twenty-six weeks on agent systems, written in order, from vocabulary through a hand-written runtime to what it takes to run in production. See the series index for what is published and what is coming.
The full archive is on Writing, newest first, filterable by tag. Shorter pieces are in Notes. There is also earlier work on e-commerce, retail AI and product metrics from before this series began.
Leave your email and I’ll send new essays when they go up.