Agent self-analysis shows late-night AI frustration comes from long sessions, not fatigue
dbreunig · x · 2026-11-06
Developer dbreunig noticed he gets more frustrated with coding agents late at night, so he asked an agent to inspect all his traces with agentsview to test two theories.
Findings:
- Theory 1 (tired, lazy prompts): not supported. Late-night prompts are slightly shorter, but prompt length doesn't hurt output — 1-5 word instructions get pushback only 4% of the time vs 13% for 100+ word ones (longer prompts mean harder tasks).
- Theory 2 (long context): half right. Late prompts sit in larger contexts (292k vs 195k tokens in the morning), but token count itself has no effect. The real variable is session length: late prompts come 21 turns into a session vs 7 in the morning, and each extra turn raises the odds of an angry prompt by 1.4% — 20 turns adds 30%.
Takeaway: clear your context and start fresh sessions. The self-audit trick — having an agent analyze your own traces — is directly reusable.
More from coding & agent
- A notebook app turns Claude into a tutor on reMarkable and iPad — tokentrillionaire · 2026-11-06
- Measured: in-process BM25 memory costs ~4.6KB RAM per entry, linear growth — Neither-Witness-6010 · 2026-11-06
- Tip: add a phone preview screen to make mobile vibe coding painless — yangyi · 2026-11-06
- Building agents: life-oriented companions beat productivity tools in real use — yangyi · 2026-11-06
- ThinkyMachines' Inkling 2 Small trains models on its own: $426 of compute doubles Qwen score in 4.5 hours — Scobleizer · 2026-11-06
- Skillry ships a Claude Skill recreating Anthropic's homepage hero design — yihui_indie · 2026-11-06