Profile-Then-Reason: Bounded Semantic Complexity for Tool-Augmented Language Agents
This work addresses efficiency and reliability issues for developers and users of large language model agents in tasks like retrieval and decomposition, though it is incremental as it builds on existing reactive execution methods.
The paper tackles the problem of high latency and error sensitivity in tool-augmented language agents by introducing Profile-Then-Reason (PTR), a bounded execution framework that reduces language model calls to two in nominal cases and three in worst cases, achieving pairwise exact-match advantage in 16 out of 24 configurations in experiments.
Large language model agents that use external tools are often implemented through reactive execution, in which reasoning is repeatedly recomputed after each observation, increasing latency and sensitivity to error propagation. This work introduces Profile--Then--Reason (PTR), a bounded execution framework for structured tool-augmented reasoning, in which a language model first synthesizes an explicit workflow, deterministic or guarded operators execute that workflow, a verifier evaluates the resulting trace, and repair is invoked only when the original workflow is no longer reliable. A mathematical formulation is developed in which the full pipeline is expressed as a composition of profile, routing, execution, verification, repair, and reasoning operators; under bounded repair, the number of language-model calls is restricted to two in the nominal case and three in the worst case. Experiments against a ReAct baseline on six benchmarks and four language models show that PTR achieves the pairwise exact-match advantage in 16 of 24 configurations. The results indicate that PTR is particularly effective on retrieval-centered and decomposition-heavy tasks, whereas reactive execution remains preferable when success depends on substantial online adaptation.