How to Speculate about Uncertainty in Agentic Coding? A Draft-Model Gate Method

2026-09-04

AI Summary

本文提出了一种名为推测不确定性(SU)的方法，通过分析代理输出的令牌来预测其失败的可能性，无需访问内部参数。该方法在软件工程中减少了执行错误率和成本。

Abstract

LLM agents deployed for software engineering fail expensively: they act confidently wrong, and bad actions are recognized only after costly execution and retry. We present Speculative Uncertainty (SU), a method that recovers a predictive failure signal for a black-box agent from its output tokens alone, with no access to logits, weights, activations, or repeated sampling. Inverting speculative decoding, a small open-weight draft model scores the agent's already-generated trajectory in a single forward pass. From these speculative cross-likelihoods we extract phase-aware features by separating the reasoning and action spans, and calibrate them against a verifiable objective. SU produces a failure-likelihood score that any downstream policy, such as routing, human intervention, or extra test-time compute, can consume directly. To show the signal is actionable, we instantiate one such policy, a pre-execution veto gate, on software engineering agents Qwen3-Coder-480B and closed-source Claude 3.5 Sonnet, cutting execution error rate by 6-8 percentage points and token cost by 14-19% in deployment, transferring to out-of-distribution benchmarks without retraining, and generalizing across agent models.

Problem

• Speculative Uncertainty
• Software Engineering
• Failure Prediction
• Agent Output Tokens

Innovation

• Speculative Uncertainty
• black-box agent
• speculative decoding
• phase-aware features
• failure-likelihood score