Search papers, labs, and topics across Lattice.
This paper explores a novel approach for LLM agents that allows them to dynamically delegate tasks to stronger models during reasoning, rather than relying solely on pre- or post-response delegation. By framing this intra-generation delegation as a Bayesian optimal-stopping problem, the authors derive a time-varying escalation threshold and demonstrate its effectiveness through theoretical proofs and controlled simulations. The results show that this method significantly outperforms traditional routing strategies, with a regret bound that improves with the number of labeled calibration trajectories.
Dynamic delegation during reasoning allows LLM agents to outperform static routing strategies, achieving better task success rates with a Bayesian approach.
Current LLM agent systems decide delegation before reasoning begins (a router picks a model) or after a response is complete (a verifier scores it and may retry). We study a third regime: an agent that recognises, during its own reasoning, that it is unlikely to succeed and transfers control to a stronger model. We formulate intra-generation delegation as a Bayesian optimal-stopping problem over a learned competence posterior -- an online estimate of the agent's eventual task success whose sufficient statistics are learned from labelled trajectories, not read off raw entropy. We derive the myopic escalation threshold in closed form, characterise the optimal policy via dynamic programming, and prove that the optimal policy is a time-varying threshold with no shape assumption on the raw signal. We further prove exponential separation of the oracle belief at the Chernoff-information rate of the signal, a regret bound governed by the calibration of the posterior, and a finite-sample guarantee: with n labelled calibration trajectories the deployed plug-in policy's regret decays as 1/sqrt(n). A controlled simulation study confirms each prediction of the theory, including the predicted 1/sqrt(n) rate. We additionally report a real-model validation on a Qwen2.5-Coder 1.5B->7B code cascade (MBPP, 257 tasks), confirming two of three pre-registered predictions: the escalation frontier dominates post-hoc routing at equal cost, and the cumulative competence belief's discrimination rises over generation.
