The authors study whether LLM agents actually follow the plans they declare. They introduce Planning-as-Routing, where an LLM declares one of four planning modes, Predefined, Sequential, Hierarchical, or Search, and a deterministic router sends the task to a pattern-specific executor. This separates plan selection from plan execution, which final task success alone cannot do.
Across four benchmarks and three LLMs, generic Plan+ReAct frequently fails to preserve the declared planning structure, especially for longer plans, with only 22 to 45 percent of trajectories preserving it. Pattern-specific executors enforce that structure instead. The best planning mode varies by environment and model: Search works best on ALFWorld, Hierarchical works best on SWE-bench, and the strongest mode can differ between models on the same benchmark.
The largest gains come from execution. Pattern-specific executors improve task success from 0.48 to 0.92 on ALFWorld and from 0.36 to 0.44 on SWE-bench Verified over Plan+ReAct. However, current LLMs do not reliably choose the strongest mode for each task, although few-shot examples help in some benchmark-model combinations.