← Back

I hired someone that I never met.

And this employee has been exceptional. Comes in, does the work. When I review the work, it looks great. But I don’t know what this person looks like. I don’t know even this employee’s name.

Twist: it’s an AI agent.

I was presenting an Agent Blueprint to a client: a plan to build an AI agent that automates manual, tedious customer service ticket resolution. The presentation was going well, but the client had trouble visualizing what this agent “looks” like.

“So, is it like, an app?”

I explained it like this: imagine you hired an intern for some tasks. But you never see the intern. They come in, do all the work, and leave detailed notes of all the tasks they’ve done. All you have to do is approve or reject their work. You won’t see them, but there will be plenty of signs of their work.

It clicked.

“So, can we give it a name?”

AI agents don’t always need a UI. If an existing system has a way to document decisions, via notes or tags, then all a human needs to do is to hit “approved” on the agent’s analysis.

P.S. K Solo shared an idea about writing job descriptions for agents as part of the discovery. It’s brilliant. And I’ve incorporated it into the Agent Blueprint.