Most business owners I meet frame AI wrong from the first sentence. They ask "which tool should I buy?" — the same way they'd ask about accounting software or a faster laptop. And then they're surprised when the "tool" improvises, overpromises, contradicts company policy, and occasionally invents a discount. No laptop ever did that.
Here's the frame that actually works: you didn't buy a tool. You hired a junior employee. A strange one — tireless, weirdly well-read, shameless about guessing, with no institutional memory and no fear of embarrassment. But the category is unmistakably employee, not equipment. Equipment does the same thing every time. Employees interpret, improvise, and make judgment calls in your name. The moment your AI talks to a customer, quotes a price, or drafts a reply that goes out under your logo, it's making judgment calls in your name.
Once you accept the frame, a century of ordinary management wisdom becomes applicable overnight — and most AI "best practices" turn out to be things any decent manager already knows.
You wouldn't let a new hire talk to customers on day one without a job description. Yet companies routinely connect a bot to their busiest channel with instructions that amount to "be helpful." A junior employee with a vague job description doesn't become creative; they become erratic. Write the AI the same document you'd write a human: what it does, what it never does, what it may decide on its own, and when it must fetch a manager. I call this a service passport, but the name matters less than its existence. If you can't write that page, the problem isn't the technology — you haven't decided what the job is. No hire, silicon or human, can succeed at an undefined job.
You wouldn't skip the probation period either. Nobody sane gives a new junior the keys to everything in week one; you give them a narrow slice of work, you check their output daily, you widen the mandate as trust accumulates. With AI, this is the difference between a pilot and a rollout. Give the agent one process, real but low-stakes, and read its transcripts the way you'd review a trainee's emails — not the summary dashboard, the actual conversations. The transcripts are where you meet your new employee's judgment. Dashboards are where that judgment goes to hide.
And you wouldn't accept "trust me, it went great" from a junior, so don't accept it from a machine. A human employee leaves a paper trail: sent emails, signed forms, call logs. Insist that your AI leaves one too — a record of what was asked, what was promised, what was done. Call it an AI receipt. The first time something goes sideways with a customer, that record is the difference between a five-minute correction and an unwinnable argument. Managers have known forever that supervision without records is just mood. It doesn't stop being true when the subordinate is a model.
There's also a lesson in the other direction — about what this employee is for. Juniors don't replace your senior people; they absorb the work your senior people should never have been doing. The mistake of firing the humans "because the AI handles it now" is the mistake of a manager who never noticed which parts of the job carried the judgment. The junior does the volume; the human does the exceptions, the empathy, the calls that decide whether an angry customer stays. Companies that get this split right report the boring, wonderful outcome: same headcount, more capacity, calmer people. Companies that get it wrong discover that they've put a very confident intern in charge of their hardest conversations.
One more habit transfers: performance reviews. Not "do we like the AI?" — measure it like a manager measures anyone. Did the work get faster for the customer? Did errors and redos go up or down? What do the people who work alongside it say? Put a date in the calendar, every quarter, and be willing to demote it — narrow the mandate, add a human checkpoint — when the review is bad. An employee you never review doesn't stay good; they stay unobserved.
I find this frame does one more quiet, useful thing: it kills both hype and panic at once. Nobody expects a junior hire to transform the company in a month, and nobody fears one either. They expect the junior to need a clear job, close supervision, honest records, and a gradually earned mandate — and they expect that, managed well, the junior makes the whole team more productive. That's precisely the right level of respect for what you've brought into your business.
So the question was never "which tool should I buy?" It's the question every owner already knows how to ask: am I managing this employee, or just hoping about them? Hope, as always in management, is not a strategy — but management, it turns out, still is.