How I Teach My AI Employee to Unblock Itself

A lot of travel advisors tell me the same thing. Their AI employee says it's unable to do something. Can't view this webpage. Can't do this. And then it just stops, and the task lands right back on their desk.

There's a way to fix that, and it takes one paragraph. You can teach your AI employee to unblock itself.

An AI employee does what its instructions tell it to do. If nothing in those instructions says what to do when it hits a wall, it does the polite thing and reports the wall. Nothing is broken. There's an instruction missing, the same way a new hire would come back and ask if nobody told them they were allowed to figure it out.

In my instructions file, I let my AI employee know that if it gets blocked, it needs to stop, go research the problem, and figure out a solution. Then it either comes back to me with that solution or takes the action itself, as long as the action isn't going to do something harmful. If the fix needs my permission, it asks.

Written out, the rule reads like this:

• If you get blocked or can't complete a step, don't stop and report it. Stop, research why, and look for a solution.
• If the solution is safe and inside your job, take the action and tell me what you did.
• If the solution needs my permission, or it could do something harmful, come back to me with the solution and ask before you act.

Any time it gets blocked, it goes looking for the answer first. Instead of a dead end, I get a message that says here's what happened, here's the fix, can I go ahead. Most days I just say yes. That's the difference between an assistant that hands work back and an employee that hands you options.

Don't skip the last line. You want your AI employee to be resourceful without getting reckless. Anything that sends a message to a client, spends money, deletes something, or changes a booking should always come back to you first. It works things out inside the lines, and it asks at the edges.

Put the rule wherever your AI employee reads its standing instructions, the file or settings it follows every time it starts work, not in a single chat where it'll be forgotten. Then test it on one task you know it gets stuck on and read what it comes back with. Watch the first few times it acts on its own before you stop checking.

Writing instructions that actually work is the whole game with AI employees, and this rule is one small piece of it. In my free AI Masterclass I walk travel advisors through building their AI employees live, instructions file included. Come build yours with me.