Experts don’t agree on a precise definition for agents, but here’s one:
Lillian Weng of Thinking Machines defines agents as a system that combines an LLM with planning, memory, and tool use.
One example of a type of agent is Deep Research, found in ChatGPT and other models.
Right now, these tools have significant limitations. But they are already powerful enough to make programmers more productive. It’s likely that they will get better in the coming months and years. (paraphased from Timothy B. Lee)