Google WikiSkill paper: Agents reuse experience to outperform larger models
mark_k · x · 2026-08-28
Google Research published "WikiSkill," a method where agents turn execution traces into persistent knowledge stored in a wiki, allowing skills to compound over time.
Results: Qwen 3.5 9B with evolved skills achieved 47.4% accuracy, beating Qwen 3.6 27B without skills (39.4%). On SpreadsheetBench, Qwen 3.6 27B jumped from 40.8% to 81.7% using these skills.
Cross-model transfer: Skills learned by one model often transfer effectively to others, sometimes outperforming self-learned skills. This suggests a key path for agents: accumulating experience to improve capabilities over time, rather than just scaling model size.
More from coding & agent
- YC F26's Praxis launches AI agents to run law firm ops end to end — ycombinator · 2026-10-16
- Codex's year-long rise: 40M+ weekly users and its leader now runs all of OpenAI's core products — danshipper · 2026-10-16
- Developer builds inter-agent messaging for his AI agents — complete with message undo — baditaflorin · 2026-10-16
- How Codex Happened: inside OpenAI's comeback in coding — danshipper · 2026-10-16
- iOS 'dots' workflow: one Siri command turns screenshots into a background shopping and saving agent — nickbaumann_ · 2026-10-16
- Your widget tree is already a list of options: where Jev fits in Flutter — rseroter · 2026-10-16