Writing
Build logs & design notes
Things I've built, broken, and learned from.
These posts cover the systems and patterns I'm working on across agentic AI, automation, and context engineering. They lean technical—build logs, design decisions, and the occasional postmortem. Written to share what I've learned, not to sell anything.
Build logOctober 2026
A harness that delegates work to sibling agents, scopes tasks precisely, and turns its own failed runs into improvements.
A generalised multi-headed agent package that extracts five shared extensions into a single, configurable harness. Covers the head-as-data model, the TAO registry integration, and three design failures that taught more than the working parts did.
Agentic systemsInfrastructureBuild log
Design logMay 2026
Keeping agents from writing the last step first.
JSON-structured progressive-disclosure workflows that prevent agents from wandering, pre-completing steps, or running tasks out of order. Covers the problem, the tool surface, and a concrete editorial pipeline example.
Agentic systemsContext engineeringDesign
Build logApril 2026
Or: how DAD learned to delegate.
The coordination layer that connects five specialised agents—sync and async delegation, DAG execution, shared workspaces, and a live dashboard. The infrastructure behind multi-agent workflows.
Agentic systemsInfrastructure