Designer-RSI: Evolving Procedural Memory from User Traffic for Agentic Graphic Design

Source facts

Designer-RSI uses a frozen frontier model operating professional design software through more than 230 tools and maintains an external procedural-memory bank of natural-language skills. The memory widens by adding procedures for uncovered subtasks and deepens by revising procedures against successes and failures, with a replay gate intended to prevent regressions. The preprint reports five rounds over 1,406 user briefs and 1,869 graded trajectories, with the skill bank growing from 76 to 139 procedures.

Analysis

This is a concrete pattern for continual agent adaptation without weight updates: accumulate reusable procedures, revise them against execution history, and gate changes against prior successes. The approach is notable because the feedback is noisy and there is no reliable programmatic oracle for graphic-design quality.

Uncertainty/Doubt

• This is an arXiv preprint; the reported quality gains, grading process, and regression controls were not independently reproduced here.