I think 'agent' may finally have a widely enough agreed upon definition to be useful jargon now

Willison's September 2025 concession that 'agent' now means something — specifically, an LLM that runs tools in a loop to achieve a goal. The piece then contrasts Claude's opt-in memory tool (explicit tool calls, visible in the conversation, users can see what the agent reads and writes) with ChatGPT's automatic injection model, and argues the visibility difference matters for trust, debugging, and safety. He also flags the lethal trifecta — private data plus untrusted content plus external communication — as the attack surface any memory-enabled agent has to defend. Short, skeptical, and useful because Willison is usually the last person to adopt new terminology. When he does, the field has actually settled something.