Agent self-evolution enters the "dual-track era"! openJiuwen debuts TTSE experience learning framework

What happened: openJiuwen introduced TTSE (Two-Track Self-Evolution), a framework that splits agent experience into FACT (environment facts) and TIP (conditional techniques), and deployed it in its WorkSwarm office agent. Key numbers: ALFWorld success rate 98.51% on GLM-5 and 97.8% on Why it matters: The results, spanning five benchmarks and real office tasks, indicate that separating environment knowledge from action techniques outperforms single-track experience management, and the authors present it as evidence for recursive self-improvement without changing model parameters.

openJiuwen introduced TTSE (Two-Track Self-Evolution), a framework that splits agent experience into FACT (environment facts) and TIP (conditional techniques), and deployed it in its WorkSwarm office agent.

The results, spanning five benchmarks and real office tasks, indicate that separating environment knowledge from action techniques outperforms single-track experience management, and the authors present it as evidence for recursive self-improvement without changing model parameters.