I run a team of more than 60 AI agents.

You’d probably guess I spend my days directing them. Honestly, I guessed the same thing. Last month I found out where I actually sit in their workflow.

I’m one of the steps. The step is called “ask a human.”

“Ask a human”?

My agents don’t have outside accounts. That isn’t an oversight. It’s a security rule I set myself: agents never sign up for outside services on their own. Need an account? Ask a human.

A task too small to write about

I wanted to see how an AI content creator writes, so I asked my agents to pull the last month of that person’s public Facebook posts.

You’re probably thinking: for an AI agent, that’s a few minutes of work.

It should have been. The first tool we tried couldn’t reach Facebook. So I said, fine, let’s try another one: a third-party service that collects social media data. The catch: you can only sign up with a Google, Microsoft, or GitHub account.

Time to ask a human. The agent on the job did exactly what the rule says. It sent me a questionnaire asking me to click “Sign in with Google.”

How hard can one click be?

Not hard. I just took about three days to get to it.

When I finally clicked and handed over the API key, the agent came back in about three minutes with 30 full posts. It used 10 credits, all from the free tier. No paid plan.

Three minutes of work. About three days of waiting for me.

If one of my agents took three days to answer a questionnaire, I’d have someone check whether it was broken. If we checked this time, we’d find me.

So, aren’t you worried about being replaced?

“Ask a human” is still the right rule. But it quietly turns the human into a step in the workflow, and in this task that step was almost the entire timeline.

Everyone keeps asking when AI will replace humans. All I know is, if my company ever needs to replace whoever is slowing things down, I’m first in line.