TAPE: Tool-Guided Adaptive Planning and Constrained Execution in Language Model Agents
Jongwon Jeong, Jungtaek Kim, Kangwook Lee
2026-02-25
Summary
This paper focuses on improving how AI agents, powered by large language models, perform tasks that require them to interact with an environment over and over again to reach a goal.
What's the problem?
Current AI agents struggle in situations where making even one mistake can ruin the entire process, especially when there are strict rules about what actions are allowed. The main issues are that these agents don't plan very well, and even when they do, their actions aren't always reliable due to randomness in how they choose what to do next. Basically, they can get off track easily and have trouble recovering.
What's the solution?
The researchers developed a new system called TAPE, which stands for Tool-guided Adaptive Planning with constrained Execution. TAPE improves planning by creating a network of possible plans and using a separate tool to find a plan that actually works. When carrying out the plan, TAPE tries to be more precise in its actions to avoid errors, and if things go wrong, it quickly adjusts and creates a new plan instead of sticking with a failing one.
Why it matters?
This work is important because it makes AI agents much more reliable in challenging situations. The experiments showed that TAPE significantly improved success rates across different tasks, especially in difficult scenarios and when using less powerful language models. This means we can build AI systems that are better at solving real-world problems where mistakes aren't easily tolerated.
Abstract
Language Model (LM) agents have demonstrated remarkable capabilities in solving tasks that require multiple interactions with the environment. However, they remain vulnerable in environments where a single error often leads to irrecoverable failure, particularly under strict feasibility constraints. We systematically analyze existing agent frameworks, identifying imperfect planning and stochastic execution as the primary causes. To address these challenges, we propose Tool-guided Adaptive Planning with constrained Execution (TAPE). TAPE enhances planning capability by aggregating multiple plans into a graph and employing an external solver to identify a feasible path. During execution, TAPE employs constrained decoding to reduce sampling noise, while adaptively re-planning whenever environmental feedback deviates from the intended state. Experiments across Sokoban, ALFWorld, MuSiQue, and GSM8K-Hard demonstrate that TAPE consistently outperforms existing frameworks, with particularly large gains on hard settings, improving success rates by 21.0 percentage points on hard settings on average, and by 20.0 percentage points for weaker base models on average. Code and data available at here.