AutoDesign: Meta-optimizing agent harnesses beats Claude Design on poster generation
dair_ai · x · 2026-08-15
AutoDesign puts the agent harness itself in the optimization loop, using a meta-harness optimizer to guide a code agent in rewriting the harness iteratively. On PosterBench (100 papers across five disciplines), AutoDesign scores 78.32 vs 70.87 for closed-source Claude Design. Dropping the learned DesignHarness into seven other code-agent configurations lifts average from 54.99 to 67.39, showing it learns scaffold knowledge rather than overfitting. One full run executes 253 tool calls and 11 editing turns in 40 minutes for under $3.
Related event: Meituan's AutoDesign Optimizes Agent Frameworks(2 posts)→
More from coding & agent
- Terminal Night 3 talk slides released, missing demo video to follow — 4310sy · 2026-10-16
- Agent developers are sitting on unused API credits that can exceed actual usage — EnvironmentalDay8864 · 2026-10-16
- Notion's bet on agents: abstract away deployment so users and agents just write code — simonguozirui · 2026-10-16
- Four Gemini 4 Argon agents beat Minecraft's Ender Dragon autonomously in 27 hours — Maxence Faldor · 2026-10-16
- Tencent's Terminal-RSI builds recursively self-improving terminal agents, hitting 19.84% on Terminal-Bench 4.0 — tencent · 2026-10-16
- Tsinghua's AutoSFT uses multi-agent collaboration to automate SFT data research — TsinghuaNLP · 2026-10-16