Dom Michalec
Notes on product, engineering, AI agents, and teaching people to work well with AI.
Building Pong with Claude forces five decisions you will make on every real project: model and effort, plan before code, where durable context lives, how the model verifies its work, and what it is allowed to do.
A step-by-step tutorial for building an Agent Skill that turns raw meeting notes into linted decision log entries, uploaded with the Skills API, run with code execution, and reused in Claude Code and the Agent SDK.
What the Claude API effort parameter changes, measured across all five levels on Sonnet 5.5 and Opus 5.5, plus the three production traps: per-model defaults, thinking-mode 400s, and prompt cache invalidation.