AI Agents: Research & Applications

LLM-based autonomous agents are presented as a structured overview of their architecture and applications in gaming, governance, science, and robotics. The development of agents is traced from symbolic and reactive systems through reinforcement learning, where AlphaGo defeated human Go champions, to LLMs acting as the agent's "brains." Profile modules assign roles or personas, and multi-agent systems such as ChatDev and MetaGPT use these roles for cooperative problem-solving. Memory is covered as either unified short-term memory, limited by context windows, or hybrid memory that stores long-term insights in vector databases for similarity-search retrieval.