Our new AI tools and copilots have made some royal blunders. They’ve doled out bad advice with supreme confidence, they’ve gotten talked into shady deals, they’ve made things weird and turned incredibly rude. Slip-ups are rare, of course, but when they happen the internet goes to town. We love to dismiss a wayward AI.
But that’s a huge mistake.
The impulse is, in part, a result of being threatened by AI. But I think it also exposes a profound misunderstanding: We still think of AI agents as machines that aren’t capable of real growth and improvement the way that human employees are. So, we mock their mistakes — and point out their faults as though they were Roombas trapped in a corner.
In truth, however, we’ve reached a major inflection point. Today’s AI agents aren’t static. They can grow and learn if we take the time to coach them. What’s more, every company already has the power to coach AI agents themselves.
You don’t need a PhD in machine learning. In fact, I’ve met hundreds of AI agent managers who’ve never written a line of code. What they do know is how people work and how humans are managed best. And they understand that those principles now apply to AI agents, too.
Related: Entrepreneurs Are Rushing to Use AI. Here Are 8 Questions You Should Ask First.
The golden rule of managing people (and AI)
The best managers know that human error is a constant, necessary part of human learning. For an employee to truly realize their potential, they need to be given the freedom to push boundaries, experiment and even fail. Expecting a new employee to never falter isn’t simply unrealistic — it’s also unproductive. Great managers know that messing up and growth go hand in hand.
Meanwhile, exceptional managers know that it isn’t always the employee who needs to be corrected, either. It’s often the manager’s method of onboarding, training or providing feedback that needs adjustment. Large companies lose tens of millions of dollars because employees misunderstand policies or processes. However, high-performing managers don’t automatically point the finger; instead, they use those errors as a jumping-off point for introspection and improvement.
The same principles now apply when working with AI agents. They don’t arrive as finished products. Rather (just like humans), they need onboarding and a chance to learn about their new jobs. They need feedback. They need mentoring. In short, managers are discovering that AI agents need the same kind of grace that is already given to human employees.
Seizing on AI’s “teachable moments”
Say you work at a bank and you’re onboarding an AI customer service agent. You’ve uploaded to the agent every document that your human employees use to learn about company policy and procedures (those were read and digested in moments.) Company blogs and changing product details can all be tapped into the AI, too, by simply providing relevant URLs.
Then, once the AI agent is ready to start working with customers, it finally has a chance to make its first mistake. And you have a chance to make it better.
An explanation about how to open a new checking account, for instance, might be too long-winded for customers looking for a quick answer. This isn’t a fatal flaw. It’s a teachable moment. Giving direct feedback — “shorter responses, please” — translates to instant, visible improvement.
Every reaction from the agent can be shaped and crafted, with benefits that add up quickly over time. I’ve seen managers who invest time in coaching their AI employees turn an eager “intern” into a seasoned pro in a matter of months.
The real perspective shift here comes down to recognizing these agents for what they are: Fallible but eager employees, raring to learn if we give them a chance.
What’s gained by coaching AI past its mistakes
The benefits of this mind shift are manifold. In customer service, the enormous amount of time and money spent training human agents typically leads to limited returns. Industry-wide, we lose nearly half of those hires every year. It’s a sieve, with company resources flowing down the drain.
By contrast, AI agents aren’t going anywhere. Every ounce of effort poured into an AI agent’s training goes on producing returns in perpetuity. What’s more, those returns rapidly scale — a VP at Wealthsimple, a leading online investment platform, recently estimated that her AI Agent delivered the productivity of ten full-time human agents. That, by the way, allows those humans to focus on concierge experiences that are more complex and still require the human touch.
We already know that the quality management of humans is directly correlated to a higher market cap. Quality management of AI agents promises an even more positive effect. AI agents never forget and never leave, allowing management efforts to be scaled and shared.
But the upside extends beyond capable AI agents. Because AI needs human management and feedback in order to succeed, it doesn’t end up just taking jobs — but creating new, and often better, ones. I’ve seen how frontline customer service workers have taken on roles managing AI, giving them a renewed sense of ownership in the company.
Indeed, managers who learn to coach AI agents are making themselves indispensable. They’ve learned to use a tool that can boost productivity in every other department in their company.
Related: You Can Fear It and Still Use It — Why Are So Many American Workers Shy About AI?
A future where we’re all managers
Nor is this change limited to a select few roles. From here on out, just about everybody is going to be an AI manager. We’ll all have AI agents working for us, upping our productivity. And that means this mind-shift I’m describing — thinking about AI agents as teachable, ever-evolving co-workers — will be discussed far beyond the C-suite.
As a new paradigm sets in, agents will become precisely as intelligent as we, collectively, bother to make them.
It starts by extending to AI agents the same courtesy that we extend to humans — understanding that everybody (and every bot) makes mistakes. Then, we do what great managers have always done: coach, train and remove obstacles. They are learning machines, after all, just waiting for the next lesson that lets them leap forward again. And that’s where we come in.
