Models improving doesn't obsolete your agentic coding scaffolding, argues pushback on viral take
max_paperclips · x · 2026-09-30
maxpaperclips pushes back on housecor's viral take that as models and harnesses improve, developers' complex layers atop agentic coding tools (skills, custom loops, MCPs, memory systems) become unnecessary or even harmful. He argues models don't suddenly memorize repetitive tasks, gain awareness of available APIs, or accumulate reusable utilities as they improve. The real change is only that instructions and prompts can be less detailed; while-loops, task tracking, and other scaffolding don't disappear because a model got 2% better at TerminalBench.
More from coding & agent
- More detail: LLM-planted decision points make replay-based Minecraft verification seamless — gandamu_ml · 2026-09-30
- OpenAI DevDay's overlooked signals: plugin distribution window, Codex security agent, Decisions API — dhruv2038 · 2026-09-30
- Dev uses LLM-recorded execution order to make Minecraft worldgen verification deterministic — gandamu_ml · 2026-09-30
- CaptchaArena releases 50K verified CAPTCHA trajectories; 9B agent hits 71.7% Pass@1 — ColumbiaUniversity · 2026-09-30
- Google's TabFM-Auto: LLM agent evolves data pipelines, lifting TabFM by 228 Elo and topping MLE-Bench — google · 2026-09-30
- LEGO-Anything: coding agents write Blender code to rebuild editable 3D scenes, 62.7% gains — AWS · 2026-09-30