POrchestra lets agents revise task setups mid-task
In parole semplici
POrchestra is a way to coordinate AI helpers working on long tasks. If a helper discovers that its goal or tools need to change, it can pause, tell the coordinator what is happening and why, and then continue with its earlier work still in its history.
On GAIA2, a test of assistants adapting to updates in simulated apps, POrchestra had higher task success than ReAct with the same execution budget. Success means the fraction of scenarios judged valid; this comparison does not show that one feature alone caused the difference.
La ricerca
When a delegated helper finds that its goal or tools no longer fit, the orchestrator may not learn about the mismatch until the helper finishes or a preset trigger occurs. Can a helper request a change while it works, and can feedback from that work also guide better orchestration decisions?
POrchestra lets helpers request setup changes mid-task and continue with their history, while OPOD turns attributable feedback into training guidance alongside final rewards.
GAIA2 tests digital assistants handling updates and requests in simulated applications. Can orchestration adapt while a helper is still working, and does that matter for task success?
With DeepSeek-V4-Flash as executor and the same total execution budget, POrchestra reached 55.5% success across 128 scenarios, versus 44.5% for ReAct. Success is the fraction judged valid.
This result supports POrchestra in this comparison, though the systems differ in more than their feedback mechanism.
The comparison changes the overall systems and model roles, so it does not show that proactive feedback alone caused the difference. The results cover the tested systems and tasks; they do not establish gains for every agent or setting.
Analisi
Dettagli
- Autori
- Xueyang Feng, Chaozhuo Li, Xiaohe Bo, Jiakai Tang, Chenxi Li
- Licenza
- arXiv non-exclusive