When to Hire a Human Instead of Building Another AI Agent
Build another AI agent for work you can judge. Hire a person when you can't tell a great result from a dangerous one. Three questions to ask.
Hire a person when you can't tell a great result from a dangerous one, or when a mistake would be expensive, legal, or reputational. Build another AI agent for repeating work whose output you can judge. Three questions sort a task before you build.
Hire a person when you can't tell a great result from a dangerous one. An AI agent multiplies work you can judge. When you can't judge the output, your review turns into a guess.
Your review works because you're reading the result and you know good from bad. My Studio page says it plainly: an AI agent can ignore its instructions, and it can make things up. You still read the result before it goes anywhere. Keeping a person responsible for the result is a management rule, and delegating execution to an AI agent without delegating responsibility covers why.
Which three questions sort a task before you build?
Can you tell a great result from a dangerous one in this kind of work? If you can, an AI agent can carry it while you review. If you can't, start with someone who can.
Would a mistake be expensive, legal, or reputational? Then a qualified person comes first, even if an AI agent helps them. The same page says to talk to a qualified professional first for anything with contracts, payments, or regulated work, and for a legal, tax, or compliance review of your AI agents.
The third is whether the work is a one-time foundation or a task that repeats. A specialist can set up the foundation once, and your AI agents can run the routine on top of it.
What does it sound like when you reach the edge of what you know?
In my October 7 session, someone asked how the platform I teach on compares to Co-Pilot. I said "I am not an expert in Co-Pilot to be able to compare and contrast each feature, but I can connect you with someone who is an expert in both."
What do you build once the person has done their part?
Write what the specialist sets up into the AI agent's instructions, and let the AI agent run it. Then test it before it touches live work. The 35 AI Agent Studio Guide has four tests for that step.
Run the three questions on the AI agent you're about to build. If the answer comes out as a person, call them first.