Blog
Notes on research in progress: experiments, negative results, and open questions about agents, world models and generative modelling. Posts are written as short papers when the work is substantial enough, and shared early when it is not.
-
Oct 10, 2026 · exploratory pilot
Does a Failed Agent’s Workspace Help the Next One? A Controlled Pilot on Failure-Conditioned Recovery Between Coding Agents
When a coding agent fails, does handing the failed workspace to a different model help? A controlled study on SWE-bench Pro V2 with exact workspace checkpoints, pristine-start controls and three replicates — with null results, a failed replication, and an open invitation to scale it up.