Simon Willison says loops are becoming obsolete as models handle long tasks natively
teropa · x · 2026-07-23
Loops may be becoming unnecessary for long-horizon model work
Simon Willison argues that loops were a short-lived workaround for models that could not reliably keep working on a long task until they reached a defined goal.
He says newer systems like Fable and GPT-5.6 — and probably Kimi K3 too — can already do this out of the box, without needing an external loop or orchestration layer.
More from Models
- Codex rumor points to GPT-5.6 Sol “spark” on Cerebras at 750 tok/s — haider1 · 2026-07-23
- Gemma 4 tops 300 million downloads three months after launch — osanseviero · 2026-07-23
- OpenAI opens GPT-5.6 to the public across ChatGPT, Codex and API — emmanuelvivier · 2026-07-23
- Laguna at low quant seems to overthink and burn through context fast — IUseClifford · 2026-07-23
- Qwen-Image-3.0 gets put through layout-heavy tests against GPT-Image-2 — Scobleizer · 2026-07-23
- Dean Ball says Kimi is strong at coding, but open-weight economics still hurt — koltregaskes · 2026-07-23