Large language models (LLMs) have shown remarkable abilities, but can they plan complex tasks? Recent research explores whether graph learning can boost LLMs' planning skills, particularly in agents like HuggingGPT that use tools to fulfill user requests. Think of a user asking an AI to create an image of a girl reading, matching the pose of a boy in another picture, and then describe the image aloud. This request involves multiple steps and tools, like pose detection, image generation, and text-to-speech. These steps and their dependencies can be represented as a graph, where nodes are tasks and edges are connections. Researchers found that LLMs often struggle to 'see' this graph clearly, leading to planning failures. They hallucinate tasks or dependencies that don't exist, especially as the graph grows larger. Theoretically, LLMs process graphs as sequences, which doesn't fully capture the graph's structure. The way LLMs are trained also introduces biases that hinder their graph reasoning. To address this, researchers integrated graph neural networks (GNNs) with LLMs. GNNs excel at graph tasks. The LLM first breaks down the user request into steps. Then, a GNN selects the relevant tasks based on these steps and the task graph. Surprisingly, this approach works even without training the GNN, and minimal training boosts performance further, especially with larger graphs. Experiments showed that GNNs significantly outperform existing methods, using fewer resources and less time. This research opens exciting possibilities for improving LLM-based agents. Imagine AI assistants that can seamlessly plan and execute complex tasks, from booking travel to managing projects. However, challenges remain. The current method is simple, and more advanced graph algorithms could further enhance performance. Also, building the task graph manually is time-consuming, so automating this process is crucial. This research is a significant step towards more capable and efficient LLM agents, paving the way for truly intelligent AI assistants.