TOM-SWE: User Mental Modeling For Software Engineering Agents

2025-10-24

AI Summary

Existing code-generation agents achieve notable progress on complex tasks but struggle to accurately infer and continuously track ambiguous or context-dependent user intentions. This paper introduces ToM-Agent, the first Theory-of-Mind–enabled agent for software engineering, featuring a dual-agent collaborative architecture: a lightweight Theory-of-Mind (ToM) agent dynamically models user goals, constraints, and preferences while maintaining a persistent, stateful memory of user context; the primary SWE agent then generates intent-driven code decisions grounded in this model. Our approach integrates interaction history analysis, probabilistic intention inference, and context-aware prompt generation. On the state-aware SWE benchmark, ToM-Agent achieves a 59.7% task success rate—surpassing OpenHands (18.1%). In a three-week deployment with real developers, it was rated useful 86% of the time. The core contribution is the systematic integration of theoretical Theory-of-Mind capabilities into coding agents, enabling long-term, dynamic, and interpretable modeling of user intent.

Technology Category

• Cognitive Modeling & Cognitive Systems: Agent Architectures
• Natural Language Processing: Code Generation / Program Synthesis from Natural Language
• Multiagent Systems: Modeling other Agents

Application Category

• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Search and Retrieval-Augmented AI: Agentic search
• Responsible Web: Machine-in-the-loop, human agency and autonomy

Abstract

Recent advances in coding agents have made them capable of planning, editing, running, and testing complex code bases. Despite their growing ability in coding tasks, these systems still struggle to infer and track user intent, especially when instructions are underspecified or context-dependent. To bridge this gap, we introduce ToM-SWE, a dual-agent architecture that pairs a primary software-engineering (SWE) agent with a lightweight theory-of-mind (ToM) partner agent dedicated to modeling the user's mental state. The ToM agent infers user goals, constraints, and preferences from instructions and interaction history, maintains a extbf{persistent memory} of the user, and provides user-related suggestions to the SWE agent. In two software engineering benchmarks (ambiguous SWE-bench and stateful SWE-bench), ToM-SWE improves task success rates and user satisfaction. Notably, on the stateful SWE benchmark, a newly introduced evaluation that provides agents with a user simulator along with previous interaction histories, ToM-SWE achieves a substantially higher task success rate of 59.7% compared to 18.1% for OpenHands, a state-of-the-art SWE agent. Furthermore, in a three-week study with professional developers using ToM-SWE in their daily work, participants found it useful 86% of the time, underscoring the value of stateful user modeling for practical coding agents.

Problem

• Software engineering agents struggle to infer user intent from ambiguous instructions
• Existing systems lack persistent memory of user goals and preferences
• Current coding agents fail to track context-dependent user requirements

Innovation

• Dual-agent architecture with user mental modeling
• Persistent memory tracks user goals and preferences
• Stateful user modeling improves task success rates