On the Clock: Towards Punctual and Productive Time-Budgeted AI Agents

A new arXiv paper (2610.10833v1) finds that small LLM agents fail to use explicit wall-clock time budgets productively even when they learn to respect them. Testing Qwen3.6-27B on five MLE-Bench Lite competitions and Qwen3-4B on Zork I (Jericho), the researchers found that stating the budget only in the prompt produced poor time control, while harness-injected timing information substantially improved budget adherence for Qwen3.6-27B with no measurable performance loss and enforcement hooks tightened adherence further. RL with GRPO achieved near-perfect budget adherence on Zork I and generalized to held-out budgets, but did not improve task performance on MLE-Bench, and GRPO training on multiple budgets tended to collapse toward the strategy learned for the shortest budget, revealing a gap between time adherence and productive time allocation.

arXiv:2610.10833v1 Announce Type: new Abstract: We study whether small LLM agents can operate effectively under explicit wall-clock time budgets by both respecting the allocated runtime and using available time productively. We evaluate Qwen3.6-27B on five competitions from MLE-Bench Lite and Qwen3-4B on Zork I Jericho , two agentic benchmarks where additional computational time can meaningfully improve performance. In the simplest setting, where the budget is stated only in the prompt, agents fail to translate the stated budget into controlled use of time. These failures arise from gaps in time awareness, since the harness provides no timing feedback, but also because they cannot reliably anticipate the duration of actions, and do not have a learned mapping from available time to an appropriate strategy. We investigate two complementary classes of interventions: harness-based mechanisms that expose timing information and enforce deadlines, and reinforcement learning with budget-aware rewards. Injecting timing information through the harness substantially improves budget adherence for Qwen3.6-27B without measurable loss in performance, while enforcement hooks tighten adherence further. RL with GRPO achieves near-perfect budget adherence on Zork I and generalizes to held-out budgets not seen during training, but does not improve task performance over the untrained harness on MLE-Bench. Once agents are made to respect the budget, they still fail to use additional time to improve task performance. RL-trained policies learn when to stop but often fill extra time with repeated actions, and GRPO training on multiple budgets tends to collapse toward the strategy learned for the shortest budget. Our results reveal a gap between time adherence and productive time allocation, which remains a central challenge for budget-conditioned agents.