A friend asked me recently what I'd been spending so much time working on. I told her I'd been experimenting with AI agents. She nodded, thought about it for a second, and then asked the obvious question:
"So... ChatGPT?"
At first, that felt like an easy question to answer. I started explaining that an agent can use tools, work through tasks over time, read files, modify code, and interact with systems around it. Some frameworks even coordinate multiple agents, each with different roles and responsibilities. The explanation wasn't wrong, but the longer I talked, the less satisfied I became with it. I realized I wasn't struggling to explain the technology. I was struggling with the metaphor.
Most of the language we use to describe modern AI systems comes from organizations. We talk about workers, managers, planners, reviewers, and teams. We discuss delegation, coordination, communication, and hierarchy. If you spend enough time reading about agent architectures, it starts to sound less like software engineering and more like organizational design. At first that seemed perfectly reasonable. Agents behave enough like people that the comparison feels natural.
Lately, though, I've started wondering whether that metaphor is quietly shaping how we think about the problem itself.
