Only Context Agents Can't Find: Goals, Constraints, and Done Criteria
FinanceYF5 · x · 2026-11-12
Part 2 of the coding-agent management thread: only add context the agent can't find itself. Code structure, conventions, and testing patterns are usually readable from the repo. What you actually need to spell out: task goals and rationale, constraints, what NOT to do, and completion criteria — tell it the outcome, don't dictate every step.
(Quoted part 1: give agents self-verification via tests/builds/lint/screenshots/diffs and require them to show results.)
Related event: Working with Coding Agents Is Management: 12 Practical Lessons(4 posts)→
More from coding & agent
- Dev oneshots RK3588 DDR timing with a frontier model, building first 10GbE 4x NVMe NAS — teortaxesTex · 2026-11-12
- Anthropic's Lydia Hallie on Claude Code best practices: 'The modern programming language is English' — ghumare64 · 2026-11-12
- Holiday Extras rebuilds its main app in two weeks with OpenAI Codex — OpenAI · 2026-11-12
- Addy Osmani's guide to picking a Claude model: start with Opus at medium effort — bibryam · 2026-11-12
- United Airlines bans AI agents from booking flights on behalf of travelers — unconventionalbook · 2026-11-12
- Fable 5.2 ports a bespoke CUDA voxel renderer to Vulkan, running 4K on an RTX 5090 — jongranskog · 2026-11-12