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
- Perplexity launches Search SDK for agents, enabling deep research in code — AravSrinivas · 2026-08-15
- Self-bench: Open-source tool to auto-build private coding-agent benchmarks from your repo — ricklamers · 2026-08-15
- 27B agent Faraday beats Claude Opus 4.8 and GPT-5.5 on research replication via new Replica method — omarsar0 · 2026-08-15
- Seeking LiteLLM Alternatives with Enterprise-Grade Security and RBAC — Screwtapeworm69 · 2026-08-15
- What Hidden States Should an AI Agent Track When Diagnosing CI Failures? A Developer Seeks Feedback — Elegant_Quantity_583 · 2026-08-15
- AI analysis tool runs 67 minutes, gathers 1,063 evidence items, verifies 532 claims — ChrisGPT · 2026-08-15