Type: url
Original: https://simonwillison.net/2025/Sep/18/agents/
Status: generated
Summary
The post in which the author, after years of avoiding the term, adopts “an LLM agent runs tools in a loop to achieve a goal” as a definition he is willing to use without qualification. It argues jargon is only useful when both parties share a definition, explains why memory and a user-set goal are not separate requirements, rejects the reading of agents as replacements for human staff on the grounds that accountability remains unique to humans, and names OpenAI’s three incompatible usages as the biggest source of confusion.