22 Agentic Engineering Practices I Actually Use Daily
No IDE heroics: plan.md, voice, cmux tabs, remote control, skip-permissions discipline, research-before-plan, and resting before AI psychosis.
I mostly do not open an IDE or type production code by hand. I talk, plan, and build in loops. Twenty-two practices that stuck:
Plan first (plan.md), treat plans as agent checklists not essays, use plans for non-code work, default to voice, run 4–6 cmux tabs, open terminals into Codex, remote-control sessions, skip permissions only with completion alarms, GPT Pro plans while Codex builds, research with last30days before planning, paste full transcripts with consent, humans steer taste, HyperFrames for video, notes as agent memory, VPS+mosh+tmux, share plans with non-devs, skill-ify repeats, PR open source the same way, carry batteries, automate local life with K-skills, rest before AI psychosis, and yes — this post used the same loop.
The job is direction and standards. Volume is the fleet’s problem.
訂閱電子報
獲取最新 AI 洞見。