What’s an AI Agent, Anyway?
Think of it like hiring someone new. When you bring on a real employee, you write a job description — their role, responsibilities, skills, maybe even the personality you’re looking for. An AI agent works the same way.
Instead of posting on LinkedIn, you describe the role in a plain text file. Give it a name, a job, a set of capabilities. That file becomes your AI “employee.” I know it sounds either too simple or too abstract — but stay with me.
Start With a Manager
Here’s the part that surprised me most: you don’t build a team by creating every agent yourself.
You start with one agent — an orchestrator, which is just a fancy word for a manager. Then you tell that manager what kind of team you need.
I typed something like:
I need a fitness coach, a secretary, and an editor. Here’s what they each need to do...
My orchestrator, Sage, did the rest. It wrote the job descriptions and set up the agents.
At one point, I asked my secretary Ivy to handle business card scanning. She couldn’t. Sage went ahead and “hired” a new specialist “Kai” without me asking. Same logic as good management: a great hire can find better people than you would on your own.
Skills Are Just Job Capabilities
Every employee has things they’re good at. So do AI agents — these are called “skills,” and they’re defined in separate files attached to each agent.
A skill might be:
check my schedule every morning and flag conflicts.
Or:
when I send you a draft, proofread it and fact-check it.
Or:
scan business card photos and log contacts in a spreadsheet.
The better you define the skill, the better the agent performs. Garbage in, garbage out — same as managing real people.
What This Means for You
I currently have five agents running: Sage (chief of staff), Rex (fitness coach), Ivy (secretary), Vera (editor), and Kai (card-scanning intern). I didn’t manually build all of them — I just told Sage what I needed.
You don’t need a technical background to do this. If you’ve ever written a job description or briefed a new hire, you already have the mental model. Start small. Give one agent a job. See what it does. The tools are more accessible than they look.
