Denny Zhou: Least-to-Most Prompting Was an Early Agent Harness, Hitting 99% on SCAN
denny_zhou · x · 2026-09-08
Google researcher Denny Zhou reframes his 2022 least-to-most prompting work in today's language: it was an early agent harness — decompose a complex task into smaller ones, then repeatedly call an LLM to solve them one by one ("Decompose. Solve. Repeat.").
- On the compositional generalization benchmark SCAN, least-to-most prompting solved any split (including the length split) with at least 99% accuracy using just 14 exemplars, versus only 16% for chain-of-thought prompting.
- The retrospective connects 2022-era prompting research directly to today's agent orchestration ideas: multi-step decomposition and repeated LLM calls are the core of modern agent frameworks.
More from coding & agent
- Reticle builds a verification layer that lets AI coding agents check their own generated code — Scobleizer · 2026-10-27
- Trust your agent, but read its logs: Claude managed agents tips — brada · 2026-10-27
- Dev training a guard model to block prompt injection, tool and memory poisoning — gnukeith · 2026-10-27
- Dev trains a ~4GB-VRAM sentinel model to guard against prompt injection and poisoned memory — gnukeith · 2026-10-27
- Superset rebrands as Paradev, raises $11.5M seed led by USV — massimosgrelli · 2026-10-27
- Ex-Big Tech managers may thrive in the AI era: manage agents like you managed teams — lxfater · 2026-10-27