Why we picked 'coworker' over 'agent'
Every AI startup calls them agents. We don't. Here's the small word swap that changed how non-technical people felt about hiring software that does their work.
Every AI startup calls them agents. Some call them assistants. The honest engineering term is something like autonomous task-completing software with tool use. We picked coworker instead.
Not because the others are wrong. Because none of them sound like the thing you'd ask for help.
The problem with "agent"
"Agent" is the term the people building this stuff use among themselves. It comes from reinforcement learning research — an agent is a thing that observes an environment and takes actions to maximise a reward. Useful word in a paper. Cold word on a homepage.
When a coffee shop owner reads "hire your first AI agent," three things happen at once:
- They don't know what an agent is.
- They assume it's technical.
- They close the tab.
The job of a homepage is to get someone to imagine using the thing. "Agent" makes them imagine reading a manual.
Why "coworker" lands
Everyone knows what a coworker is. A coworker is someone you ask for help. They ask questions back. They get up to speed. They show their work. They have a personality you adjust to.
That's also exactly what Caminu does — chat-first, asks clarifying questions, runs on a schedule, reports back. The product mechanic was already coworker-shaped. The word just made it visible.
The hidden cost of words
Pick the term that describes the job, not the technology. "Coworker" forces us to keep building something a coworker would actually do — show up, be briefable, finish work end-to-end, report in. The day we stop deserving the word, we'll know.
Words don't just describe a product. They constrain it.