11 tiny coding agent fixes with a stupid amount of payoff
Cole Medin · youtube · 2026-09-02
Cole Medin shares 11 small changes that make coding agents far more reliable, agent-agnostic and backed by arXiv studies:
- Write for the agent, not humans; instruction files rot — one in four repos have stale rules
- /compact isn't worth it: only 10% of context survives and safety constraints erode — put load-bearing rules in hooks
- Less context beats more; projects with no AI config grow twice as complex
- Don't escalate models mid-task (recovers less than half the gap); naming an agent coordinator doesn't improve results
- Never let the writer approve the work; 85% had a better answer before the last iteration — over-revising is real
- Validation is a system, not a step
His skills repo is open-sourced alongside the video.
More from coding & agent
- LLM one-shots a Minecraft 3x3 piston door by building its own Linux test infra — Vjeux · 2026-10-18
- Dev builds Mumbai-inspired 3D manga background generator from OpenStreetMap data — CurieuxExplorer · 2026-10-18
- Solo founder builds a money-printing data service on Claude's Pi Durable harness — MicahBerkley · 2026-10-18
- Dev builds Arch-based OS with custom desktop and browser using Claude Code — Snoo-10493 · 2026-10-18
- Pi Durable went through 7 rewrites yet agent session files still lack a standard — ricklamers · 2026-10-18
- Five SaaS agent workflows fail the same way — and it's never model intelligence — usehive · 2026-10-18