From AutoResearch to Continual AutoResearch: Do Agents Learn Across Tasks?
Published in NeurIPS 2026 MetaAgents Workshop, 2026
We introduce Continual AutoResearch, an open-ended environment that compares persistent and reset instances of the same agent under matched task streams, data and time budget, across nine task families. Across five frontier agents, persistent state improves research efficiency, but these gains do not consistently translate into better final solutions.
Recommended citation: Qin, J., Deng, R., Ahmed, M. O., Tung, F. & Feng, L. (2026). "From AutoResearch to Continual AutoResearch: Do Agents Learn Across Tasks?" NeurIPS 2026 MetaAgents Workshop.
Download Paper