Applying the bitter lesson to my LLM workflow: skip agents, just talk to the model
esrtweet · x · 2026-11-09
Developer esrtweet explains why he deliberately keeps his LLM workflow primitive: no subagents, no accumulated skills, no unattended "software factory"—just conversing with the model to analyze and write code.
His reasoning: all these elaborate scaffolds are coping mechanisms for frontier models not yet being smart enough, and models keep improving. Rather than invest in procedures with a short half-life of usefulness, he applies Sutton's bitter lesson to his own workflow and saves his think time for actual problems.
More from AGI Musings
- Dev reframes LLMs as "universal translators" translating between representations — brandon_xyzw · 2026-11-09
- The bitter lesson isn't 'naive wins' — it's whatever scales best with compute — pzakin · 2026-11-09
- Owain Evans: now is the right time to switch into AI safety work — OwainEvans_UK · 2026-11-09
- Henry Farrell: the "machine gods" metaphor shrinks AI policy debate and should be resisted — sebkrier · 2026-11-09
- Thomas Dietterich: map what LLMs can assess reliably, design UX for the rest — JessicaHullman · 2026-11-09
- Bitter lesson redux: methods that generalize beat compute brute force, Sutton weighs in — ryanallred · 2026-11-09