Designer-RSI: Evolving Procedural Memory from User Traffic for Agentic Graphic Design
A frozen design agent improved by writing and revising its own reusable procedures, without model weight updates or human labels.
Designer-RSI adds an external natural-language skill memory around a frontier model operating professional design software through more than 230 tools. Across five rounds on real user briefs and automatically graded trajectories, its skill bank grew from 76 to 139 procedures. The paper reports GenEval2 execution success on Claude-Sonnet-4 rising from 72.7% to 99.3%, plus stronger benchmark win rates against a no-skill agent. The authors argue procedural memory is a practical path for adapting agents when feedback is noisy and hard to verify. ArXiv · AI/CL/LG's note
Designer-RSI adds an external natural-language skill memory around a frontier model operating professional design software through more than 230 tools. Across five rounds on real user briefs and automatically graded trajectories, its skill bank grew from 76 to 139 procedures. The paper reports GenEval2 execution success on Claude-Sonnet-4 rising from 72.7% to 99.3%, plus stronger benchmark win rates against a no-skill agent. The authors argue procedural memory is a practical path for adapting agents when feedback is noisy and hard to verify. ArXiv · AI/CL/LG's note
score 6