Manage AI like a promising new hire, not like magic

A promising hire gets your material, your rules, and supervision before the work reaches a customer; your AI deserves the same, and it will never become a person.

Every few weeks a software advertisement promises that a tool will run the department by itself, and every few weeks an owner buys the promise and waits to be relieved of management. Nothing arrives, because nothing was ever going to arrive. The version of AI that survives contact with a real business is smaller and more familiar than the advertisement: run it the way you would run a promising new hire.

The metaphor earns its keep at onboarding first. A new employee without your material, your rules, and your standards produces work that is fluent and useless, and the system is no different. Before its output is usable, it needs your documents, your price logic, your tone, and the list of things your company simply does not say. Onboarding is the difference between a tool that guesses and a tool that answers from your ground.

Then supervision. Nobody lets a first-week employee speak directly to a customer unreviewed, and the same discipline applies: the first weeks of the system's output get read by a person before anything reaches the outside. The review is not ceremony; it is where you learn what the system misunderstood about your business, one corrected answer at a time.

Boundaries come with the desk. A good manager says clearly what the hire may decide alone and what comes back for a human decision, and the system needs the same statement: what it may answer, what it may promise, what it must hand over. An AI with no escalation path is not autonomous; it is unmanaged.

Review closes the loop. The hire is graded on outcomes, on whether customers were served and promises kept, not on how confident the sentences sounded, and the system earns its place the same way. This is why the metaphor works at all: it converts a vague technology anxiety into a management discipline companies already practice, with desks, reviews, and probation periods they already understand.

But honesty requires the place where the metaphor ends. A hire learns from correction and grows into judgment; the system does not. Its improvement is the operator's work: better material, tighter rules, and clearer boundaries, supplied by you, repeatedly. And it never becomes a person. It is a tool managed like staff, which is not the same as staff.

We manage our own systems this way, which is why our assistant tells people she is AI: tools are not hired to impersonate humans, and a company that pretends otherwise is borrowing trouble. Start smaller than the advertisement suggests, onboard before you launch, supervise the first answers, and grade on outcomes. That is the whole discipline, and it is management you already know how to do. Guidance, not guarantees: the right boundaries depend on what your business can afford to delegate.