Researchers at Shanghai Jiao Tong University and Chinese technology conglomerate Tencent developed ProAct, an AI agent designed to predict user needs before users submit queries. The system uses idle time between conversations to review past interactions and prepare information in advance. According to the research paper, ProAct performed better than earlier proactive AI systems in benchmark testing, though the experiments did not involve real users. The development addresses what researchers describe as wasted computational opportunity in current AI agents that remain fundamentally reactive.
System Operates Through Multi-Stage Prediction Process
ProAct functions through multiple stages that differentiate it from conventional AI agents. The first stage, called Future-State Prediction, analyzes past conversations, user preferences, and missing information to predict likely follow-up questions. The second stage, Idle-Time Acquisition, evaluates which predictions warrant research based on relevance, timing, and potential usefulness of new information. A separate system determines whether to present prepared information immediately, save it for later use, or store it until needed.
"After each foreground interaction, the agent updates its memory, predicts possible future needs, allocates idle-time computation to valuable candidates, and decides how the resulting preparation should be handled," the researchers wrote in the paper. "This formulation ties prediction, acquisition, and delivery to a single policy, rather than treating idle-time compute as unconstrained background search."
Benchmark Testing Shows Performance Improvements
The researchers tested ProAct in 200 simulations across 40 domains, including financial planning, software release management, and cybersecurity. According to the paper, the system reduced conversation turns by 14.8% and cut follow-up requests by 11.7%. In a comparison using a benchmark called ProActEval, ProAct anticipated 703 predictable user needs versus 32 for the earlier system. The researchers also reported a 28.1% reduction in hallucinations.
"While AI agents demonstrate remarkable capabilities in reasoning and tool use, they remain fundamentally reactive: They compute responses only after explicit user prompts," the researchers wrote. "This paradigm ignores a critical opportunity: The idle time between interactions is largely wasted, leaving agents unable to prepare for future user needs."
Research Acknowledges System Limitations
Researchers acknowledged several limitations in the ProAct study. In 3% of cases, the system made responses worse by bringing up irrelevant information. The paper stated that any real-world version would need privacy protections, because the system constantly analyzes conversations and stores user data.
"Our budget analysis further shows that larger Idle-Time Acquisition budgets raise active-token cost and yield diminishing returns," the researchers wrote, "so proactive computation is an operating-point trade-off rather than something to maximize."
The research comes as autonomous AI agents spread across the tech industry, with projects such as OpenClaw and Hermes Agent delivering persistent AI assistants that handle coding, scheduling, research, and workflow automation tasks. Separate researchers earlier this month warned that AI agents may complete dangerous tasks without understanding consequences. "Like Mr. Magoo, these agents march forward toward a goal without fully understanding the consequences of their actions," lead author Erfan Shayegani, a UC Riverside doctoral student, said in a statement.
