# Agents Learn Their Runtime: Interpreter Persistence as Training-Time Semantics

arXiv: https://arxiv.org/abs/2603.01209 · HTML: https://mlsanity.com/paper/2603.01209/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Placed in Software engineering coding agents, one of 16 directions in LLM agents and tool use.
- Builds on 17 of its 23 references that are in the map.
- First posted on arXiv 1 Mar 2026.

## Placed in

- LLM agents and tool use / [Software engineering coding agents](https://mlsanity.com/agents/software-engineering-coding-agents/)

## Builds on (most cited first, 17 of 17)

- [LoRA: Low-Rank Adaptation of Large Language Models](https://mlsanity.com/paper/2106.09685/) (2021)
- [Efficient Memory Management for Large Language Model Serving with PagedAttention](https://mlsanity.com/paper/2309.06180/) (2023)
- [SWE-agent: Agent-Computer Interfaces Enable Automated Software Engineering](https://mlsanity.com/paper/2405.15793/) (2024)
- [τ-bench: A Benchmark for Tool-Agent-User Interaction in Real-World Domains](https://mlsanity.com/paper/2406.12045/) (2024)
- [Program of Thoughts Prompting: Disentangling Computation from Reasoning for Numerical Reasoning Tasks](https://mlsanity.com/paper/2211.12588/) (2022)
- [Executable Code Actions Elicit Better LLM Agents](https://mlsanity.com/paper/2402.01030/) (2024)
- [MINT: Evaluating LLMs in Multi-turn Interaction with Tools and Language Feedback](https://mlsanity.com/paper/2309.10691/) (2023)
- [ToolSandbox: A Stateful, Conversational, Interactive Evaluation Benchmark for LLM Tool Use Capabilities](https://mlsanity.com/paper/2408.04682/) (2024)
- [ToRA: A Tool-Integrated Reasoning Agent for Mathematical Problem Solving](https://mlsanity.com/paper/2309.17452/) (2023)
- [FireAct: Toward Language Agent Fine-tuning](https://mlsanity.com/paper/2310.05915/) (2023)
- [InterCode: Standardizing and Benchmarking Interactive Coding with Execution Feedback](https://mlsanity.com/paper/2306.14898/) (2023)
- [MEM1: Learning to Synergize Memory and Reasoning for Efficient Long-Horizon Agents](https://mlsanity.com/paper/2506.15841/) (2025)
- [ReSum: Unlocking Long-Horizon Search Intelligence via Context Summarization](https://mlsanity.com/paper/2509.13313/) (2025)
- [Budget-Aware Tool Use Enables Effective Agent Scaling](https://mlsanity.com/paper/2511.17006/) (2025)
- [CostBench: Evaluating Multi-Turn Cost-Optimal Planning and Adaptation in Dynamic Environments for LLM Tool-Use Agents](https://mlsanity.com/paper/2511.02734/) (2025)
- [Agent Data Protocol: Unifying Datasets for Diverse, Effective Fine-tuning of LLM Agents](https://mlsanity.com/paper/2510.24702/) (2025)
- [Calibrate-Then-Act: Cost-Aware Exploration in LLM Agents](https://mlsanity.com/paper/2602.16699/) (2026)

## Abstract (arXiv)

In CodeAct, language-model agents write Python that calls tools and use execution feedback to choose actions. Persistent runtimes preserve Python variables between actions; stateless runtimes clear them without resetting task progress. Training traces demonstrate task-solving strategies and runtime-specific ways to store and recover intermediate results. We study runtime transfer: whether agents trained under one contract remain effective under the other, a dependence that fixed-runtime evaluations can conceal. Across three tasks requiring a working record built from tool feedback, we fine-tune separate Qwen3-8B agents per task and runtime on instance-paired persistent and stateless traces and evaluate all four training-deployment combinations. We vary the per-turn tool-call cap: how many calls one Python action may execute. In our primary task, agents inspect hidden item attributes and select a high-value subset under a weight limit. At 80 calls per turn, a persistent-trained agent deployed statelessly scores 0.61 of optimal value against 0.77 under matched persistent deployment, but uses 4.4 times as many tokens rebuilding its record. Tightening the cap to 25 reduces mismatched quality to 0.07 while matched quality remains 0.66. Trace replay shows that the cap interrupts inspection, the reset erases the partial record, and rebuilding consumes calls needed for progress. The widening gap replicates across training seeds, a second rollout, and two additional base models. Across the settings studied, cap tightening produces a mismatch-specific collapse only where interruptions are frequent and matched execution resumes while mismatched execution restarts. Evaluations of trace-fine-tuned agents should specify episode limits, test runtime transfer across caps, and report cost alongside quality; otherwise, the same mismatch may appear as redundant computation or near-total quality loss.