Research questionHow should proactive service agents decide when and how to intervene from incomplete signals while balancing usefulness and authorization?Unlike instruction-following systems, proactive service agents must infer whether help is appropriate from incomplete environmental and user signals. An intervention may be useful but still impose interruption, privacy, misunderstanding, or overreach costs, while actions can also change the subsequent state. Latest papersRecent research connected to this question, newest first.Proactive Service Agents: A Unified Decision Framework, Methods, and EvaluationThe source surveys proactive service agents across streaming dialogue, screen, video, software-engineering, and human-agent collaboration resources. It frames decisions as partially observable and constrained by authorization and risk, covering triggering, timing, calibration, user burden, safety, and policy-value evidence; it does not establish a particular deployment capability or single algorithm.research paper · Sep 3, 2026