AI workflows from real work, not demos.
Practitioner notes on building with Claude and other frontier models. Agent design, tool design, prompt and context engineering, evals, model choice, real-world integrations. Patterns that transfer across greenfield and inherited systems.
A new pillar article every fortnight; short videos most days. No listicles, no hype, no thread-farming. The harness around the model is where the work lives.
Featured
-
What is an agent harness, and why it decides whether your agents work
The agent harness is the deterministic infrastructure around the model. Code, caching, prompts, skills, tools. It decides whether agents work.
Recent Posts
-
Claude Code skills, tools, and hooks: when to use each
A practical rule for Claude Code extensions: tools give capability, skills give judgment, and hooks enforce invariants.
-
How to design subagents that help rather than thrash
Subagent design is four choices per role: context, model, prompt, tools. Here's how to pick them so subagents help instead of thrash your main agent.