Build Log
What I build with AI agents, and what running them every week has taught me.
What this is
Alongside the architecture work I build things, mostly with Claude Cowork and Codex, and I write up how they behave once they run every week. This site is one of them, run as an agentic publishing operation. MarTech Ledger is another, a public record of what 15 CEP, CDP and B2B marketing automation platforms ship, where the model extracts each change and deterministic scripts decide what gets published. The video factory turns articles into narrated videos, and a working Braze demo that started on this site became a sales tool.
Everything here is written from the operator’s side: the written context, the checks that block a deploy, the approval gates and the failures that passed every check. It is the same discipline around state and governance I ask of an enterprise stack, tested on a system small enough to publish its internals. Each piece says what transfers to an organisation and what only works because there is a single operator.
The operating model
Agents in Production: what an agentic operation consists of after the demo, how it fails while every check reports green, and how repeated prompts become procedures with gates. Series page
The systems
What got built and how it runs: MarTech Ledger, the video factory, the Braze demo and this site.
The year in review
What a year of building with Claude changed in how I work, and why the editor’s role came back.
