Soundings
Our Perspectives and Insights
AI Is Not a New Business Problem. It’s an Old One.
Last year you hired a guy named AJ.
Brand new, came in off the street. You did not check many references, but he was sharp, said the right things, and you needed help in finance. So that is where you put him.
AJ moved fast. Within a few weeks he had quietly brought on four or five more people to help him, and you let him, because the work was getting done. Soon AJ and his team had the books. All of them. They were invoicing your customers, moving money between accounts, closing the month, recommending decisions that sounded good in the meeting. You nodded along, because it did sound good. Then one morning you realized you could not actually say what AJ was doing in there. You just knew he was doing a lot of it, and you had stopped checking a while ago.
If that made your stomach drop, good. Now read it again, and everywhere you saw AJ, put AI.
That is not a hypothetical. That is what “let AI run our finances” looks like in practice, and companies are doing it right now, today, with a straight face. They are wiring a capable, fast, completely unvetted newcomer straight into the most sensitive parts of the business and acting surprised when it does something a human in that seat would have been walked out for in week one.
Early in my career I was the project manager on a build, and because I owned the client relationship, I owned whatever went out the door. My director gave me one rule for developers new to the project: trust, but verify. Let them do the job, then check it where it matters, before the client ever sees it. AI is that new developer now, and your name is still on what ships.
Would You Hire a Finance Department Off the Street?
You already felt the answer in your gut. No. You would never do that with people. You would not survive the quarter.
Capability without onboarding is not an asset. It is an exposure.
So why does “let AI run our finances” not trigger the same alarm? The capability is real. The speed is extraordinary. But capability without onboarding is not an asset. It is an exposure. The reframe is the whole point of this piece: AI is not a new problem. It is an old one wearing new clothes. You already know how to bring an unproven, capable, fast-learning newcomer into your business responsibly. You have done it a hundred times. It is called hiring, onboarding, and performance management, and you are good at it. You just forgot to do it.
Onboarding Is the Whole Job
When you bring on a real person, the hire is the easy part. The work is everything after. You scope the role so they know what they own. You start them with a safety net, a probation period, someone reviewing the output before it leaves the building. You watch how they do. You give feedback. You let them earn their way into more responsibility. And if they cannot do the job, you part ways.
It is not an AI problem. It is a management problem.
AI needs every one of those steps, and people skip every one of them. They install a stranger with no probation, no review, no feedback loop, and no exit plan, then call the failure an “AI problem.” It is not an AI problem. It is a management problem. Ronald Reagan borrowed a Russian proverb for exactly this situation: trust, but verify. You can trust the new hire. You still verify the work.
Credentials, Policies, and the Security Desk
There is a second thing you do with every human you bring inside that almost nobody does with AI. You check them at the door. A new employee has to meet your bar. They sign the disclosure agreements. They learn the policies: what they can touch, what they cannot, what stays in the building. You do not let someone operate at a sensitive level on day one just because they seem capable.
Drop AI into the middle of your operation and it knows none of that. It has signed nothing. It does not know your data boundaries, your privilege lines, or your policies. If you would not let a person work without clearance, do not let your AI do it either. Same standard, same reason.
Onboard Your AI the Way You Would Onboard a Person
- Scope the role. Decide exactly what it owns before it owns anything. A job description, not a blank check.
- Start with a safety net. Probation is not an insult. It is a human reviewing the output before it ships, or, better yet, doing the work in parallel so you can judge both the result and how it got there, until trust is earned.
- Monitor and give feedback. Watch the work. Correct it. Treat the first ninety days like the first ninety days.
- Credential it, do not hand it the master key. Vet its access, policies, and data boundaries the way you would any new hire into a sensitive seat.
- Be willing to let it go. If it is not performing in the role, right-size it or pull it. You would do that with a person.
The Bottom Line
The companies winning with AI are not the ones who found a smarter model. They are the ones who stopped treating it as a magic solve and started treating it as what it actually is: a new, capable, unproven member of the team. You already know how to do this. You scope. You onboard. You monitor. You credential. You give feedback, and you part ways when it is not working.
Stop hiring AI off the street. Onboard it like you mean to keep it.