Enforce agent rules in code, not prompts—and pace work to what your team can verify
gerardsans · x · 2026-11-18
- Pushing back on "everyone optimizes for speed": when agents do most of the typing, "more is more" only creates review burnout and technical debt.
- Three principles: ① treat AI as infrastructure—it builds but never decides for you; ② enforce rules in code, never in prompts (plain text gets forgotten or dismissed by the AI); ③ pace the factory to the speed your team can actually verify.
- Accompanied by a longer piece on building this agent "control plane."
More from coding & agent
- Microsoft ships MAI Code 2 and Code 2 Flash coding models into GitHub Copilot — marlene_zw · 2026-11-18
- Microsoft launches Agent 365 Business Analytics to track agent spend and ROI — donasarkar · 2026-11-18
- A permissions markdown file to tame multi-agent file edits in Antigravity — gaganghotra_ · 2026-11-18
- YC-backed Ample launches one-stop cloud for agents, hits 5,000 users two weeks after launch — ycombinator · 2026-11-18
- OmniLoRA: train any model on any GPU inside your ComfyUI graph — pente5 · 2026-11-18
- findatasets ships /feedback endpoint so AI agents can file bug reports, following Coinbase's lead — virattt · 2026-11-18