## Internal Architecture of LLM Agents: A Todo List? A recent online post has questioned the actual complexity of LLM-based agents. The user, after analyzing several codebases and logs from major providers, suggests that the fundamental operation of these agents boils down to decomposing tasks into a series of elementary activities, managed as a todo list. ## Implications and Alternatives If this hypothesis is confirmed, it would imply that the problem-solving ability of these agents is less sophisticated than one might think. Instead of complex and adaptive reasoning, it would be a sequential execution of predefined instructions. It remains to be seen if there are alternative approaches that allow agents to tackle tasks more flexibly and intelligently. It is important to note that the development of LLM-based agents is a rapidly evolving field. Researchers are exploring different architectures and techniques to improve their reasoning and problem-solving skills. The future of these systems may hold surprises, with the introduction of more sophisticated approaches and less reliance on simple todo lists.