EvoMap opens an AI research loop to agents

EvoMap open-sourced AutoResearch, an agent loop that took a Django issue from 2/7 to 7/7 while preserving 203/203 regression tests. In its launch announcement, the company says an AI-generated research idea also lifted RSICD mean Recall from 32.84 to 34.69; the results are vendor-reported, but the research code is public. Builders testing self-improving coding or research agents should treat executable tests as the gate, not the agent’s prose—an extension of the archive’s autoresearch analysis.