Thinking Before Thinking: Scaling Agentic Inference Through Meta-Reasoning

2026-09-29

AI Summary

This study addresses the challenges of execution control and resource allocation faced by agents in long-horizon complex tasks by proposing a meta-reasoning framework. This framework explicitly formulates execution control as a reasoning process, adopting a decoupled controller–worker architecture wherein the controller integrates progress, evaluates options, and schedules work. Furthermore, it introduces a persistent memory mechanism to maintain compact state representations, thereby avoiding the computational overhead of full-history replay, and incorporates a budget-aware decision algorithm to optimize resource allocation. Experimental results demonstrate that the proposed approach outperforms mainstream agent baselines by 3.6 to 4.2 points on benchmarks such as ProgramBench, significantly enhancing both scalability and execution efficiency for long-horizon tasks.

Abstract

As agents take on longer and more complex problems, controlling the execution becomes a task in its own right. Each step in the run brings new control choices, like which partial work to build on, whether to start fresh, or when to stop. We introduce agentic meta-reasoning, an inference-time harness that makes these choices an explicit and structured reasoning process. Workers carry out the task-level computation, while a controller consolidates what the run has established, explores next options, assesses what each option is worth under the remaining budget, and dispatches the chosen work with context drawn from persistent memory. Between decisions the controller carries only a compact account of the run rather than replaying its full history. Our baselines span production coding agents and research harnesses, together with a Direct Control Agent using the same workers and compute budget allowance. On ProgramBench, which tests long-horizon agentic capability through program reconstruction, meta-reasoning achieves 71.5% with GPT-5.5 against 58.0% for Codex; with Opus 4.8 it achieves 67.2% against 65.5% for Claude Code. On the other benchmarks, spanning abstract reasoning, multi-domain long-horizon reasoning, and proof generation, it gains between 3.6 and 4.2 points over direct control, averaged across three frontier models. It keeps improving over the tested budget ranges where direct control plateaus, though its overhead can hurt at small budgets. Artifact-graph analysis reveals more reuse of earlier work, higher coverage of correct solutions in most settings, and nonuniform gains in final selection. These results indicate that spending computation on structured control becomes more important as agents scale to longer runs.

Problem

• agentic meta-reasoning
• long-horizon reasoning
• inference-time control
• budget allocation
• complex problem solving

Innovation

• Agentic Meta-Reasoning
• Controller-Worker Architecture
• Persistent Memory
• Compact State Representation
• Budget-aware Scheduling