Field Notes: behind the builds

Honest notes from shipping AI products, and the occasional essay on where this is all going.

I stopped prompt engineering. The work moved to the harness. (Essay): Models got smart enough to infer. The emerging craft is context and harness engineering: rich context, then memory, markdown, skills and MCPs around the agent.
Inference: the AI cost worth watching (Essay): Every model upgrade quietly changes your unit economics. The useful habit is knowing what a workflow costs to run, and how that shifts when the model does.
Knobs beat sliders: making the Claude API something you can feel (Field Notes): Fluffy Parrot puts Claude API parameters on knobs and per-run tabs with cost and time, so tuning feels like using an instrument.
Strategy grounded in the build (Essay): Commercial judgment and hands-on delivery belong in the same seat. The work you can show is how you know the recommendation is real.
What I learned letting two agents maintain a website (Field Notes): We Love Claude keeps a Claude feature guide current: two agents crawl the apps and open PRs. The hard part wasn't the agents.