AgentFactory: Successful Solutions Stored as Executable Subagent Code, Not Text Memories.

ArXiv 2603.18000 proposes preserving successful task solutions as directly executable Python subagent code rather than natural-language summaries. Unlike textual experience logs (lossy, non-executable, non-portable), these subagents carry standardized documentation and improve through execution feedback. The framework demonstrates progressive capability accumulation: the subagent library grows as more tasks are encountered. This solves the episodic memory problem — agents that remember what worked but can't reproduce it. arXiv