Map-driven autoresearch: let Claude pick ideas, run experiments, and reallocate resources
danshipper · x · 2026-11-14
A workflow being experimented with — "map-driven autoresearch":
- Map out all ideas you'd like to pursue;
- Have Claude decide which ones to run experiments on;
- Each pursued thread runs regular autoresearch continuously;
- Update the map and reallocate resources across paths based on what was learned.
The core idea: model research ideas as an explicit resource-allocatable map, with the LLM acting as scheduler.
More from coding & agent
- YC F26 startup Anywhere launches a map app that lets AI agents read and edit your places — ycombinator · 2026-11-14
- Trail of Bits open-sources patching-bugs and regression-hunt agent plugins, cuts regression false positives 25% — JiaweiLiu_ · 2026-11-14
- Ant Group's Mara Chain: refine prompts/harness/code, matches GEPA with far fewer rollouts — jiqizhixin · 2026-11-14
- Zach Lloyd's "software factory stack": factories-as-code and five components — blaizedsouza · 2026-11-14
- AxiomCode Graph does compiler-style call resolution for coding agents in large repos — JeremyCMorgan · 2026-11-14
- GitHub ships Copilot automations, cross-repo Projects, and Fusion model routing — 0xkarasy · 2026-11-14