cs.LG · 2610.13841 Copy arXiv ID · Oct 9, 2026 Save Returning to Development States: When Full Sandbox Snapshots Help Coding Agents Authors: Daniel Thi Graviet , Tanmay Shah
Organizations: Daytona, San Francisco, California, USA · University of Waterloo, Ontario, Canada
Abstract Coding agents often make useful progress before they fail, but retry-from-scratch discards the machine state created along the way. This makes failure recovery an interesting search problem: if intermediate environment states preserve meaningful progress, restoring and branching from them could outperform repeatedly restarting from the initial state. We implement Go-Explore-style search over full Daytona snapshots and find that higher and more efficient solve rates depend mainly on progress that sits outside the repository, such as databases, services, and other runtime state. Our results suggest that full-snapshot search is most effective when long-horizon coding work spans multiple services whose state cannot be reconstructed from a single diff.
Explore similar work Jul 21, 2026 · Qijia He, Jiayi Cheng, Chenqian Le +8 AI Coding Agents Cost-Aware Inference
University of Washington · New York University · ByteDance +1
Jul 2, 2026 · Yuvraj Virk, Zack Edds, Chunqiu Steven Xia +1 Automated Algorithm Discovery Large Language Model-Guided Optimization
University of Illinois Urbana-Champaign
Jun 5, 2026 · Shaoqiu Zhang, Yuhang Wang, Jialiang Liang +8 AI Coding Agents Agentic Search
1Shanghai Jiao Tong University · 2Xinjiang University · 4Independent Researcher +2
Jul 21, 2026, cs.AI J/K move · Enter open · S save
Qijia He, Jiayi Cheng, Chenqian Le, Rui Wang +7
University of Washington · New York University · ByteDance +1
Coding agents increasingly operate in executable environments where a failed attempt produces actionable feedback rather than merely an incorrect answer. Existing cost-aware systems typically treat such failures as cascade decisions: try a cheap model first, then escalate hard cases to a stronger and more expensive model. In coding, however, execution feedback can also make further cheap-model recovery worthwhile, raising a budgeted deployment question: when should an agent spend more cheap compute, and when should it escalate? We formulate this post-failure decision as recovery routing over heterogeneous actions and train a supervised router from execution rollouts. To make the same router usable under changing budgets, we add a Conformal Risk Control (CRC) layer that selects a deployment-time cost penalty without retraining and provides marginal expected-cost control under exchangeability. Across held-out failures from five coding benchmarks, cheap recovery and escalation exhibit complementary success patterns. The calibrated frontier improves over fixed actions, prompt-only routers, and a binary cascade baseline; in the main GPT-5.4-nano/GPT-5.4 setting, one CRC-calibrated frontier point exceeds always-escalate solve rate while using 35% of its mean recovery cost. Code is available at https://github.com/Qijia-He/agent-budget-control.