By Frank Gartland
The conversation around AI has finally started to mature. We’ve moved past the “I’ve seen the movies; AI will eliminate all the jobs” panic and landed on something more practical and honest: How do we generate measurable business impact and genuine gains from AI? That’s the question most organizations are wrestling with right now and, for a lot of them, the answer isn’t encouraging.
According to Bain & Company, nearly 40% of companies measuring AI cost savings are landing in the 0–10% efficiency range—roughly half of what they set out to achieve. The Verge reports that Uber’s COO recently said it was “getting harder to justify” spending money on AI because it was “very hard to draw a line” from that spend to business value. And Yahoo! Finance reports that Amazon SVP Dave Treadwell recently warned employees, “Don’t use AI just for the sake of using AI. Use AI to help you solve customer problems, to help you solve business problems, to innovate.”
Ask the Right Question
We’re training people on AI when we should be training them to work differently.
The fervor behind the “everyone has to learn AI” movement drove organizations to create and deliver traditional training that attempts to teach people how to use the tools. Of course, software and data engineers have been learning about machine learning, natural language processing, and the technologies that make “AI” possible since the 1960s. But it’s the approaches we’ve taken to train business users, students, and others that have been missing the mark.
The right question is not how we train people on AI. It’s how can AI change the way people create, iterate, and accelerate the process of getting work done. That question leads you to consider the tasks each group of people perform—or could perform—every day.
The real skill isn’t using AI. It’s knowing when to trust it — and when not to.
Starting Points
Here’s what most people are finding out the hard way: AI doesn’t give you answers. It gives you starting points. A fast, capable, confident-sounding collaboration partner that is sometimes (ok, yes, very often) just plain wrong.
When you push back, by simply asking it, “Are you sure about that?” or “That doesn’t sound right” you’ll find that most AI tools will readily admit they just guessed or made a bad assumption. Often, it almost seems like AI is just filling a gap in time with something it hopes you find plausible and don’t second guess.
Most people don’t think about doing that, let alone know they must. Instead, they take the first output without question, send their results along, and are stunned when a problem develops downstream. In one case, an unnamed enterprise company ran up a $500 million AI bill in a single month, according to Yahoo! Finance; not because the technology failed, but because nobody had developed the judgment to use it well. Deloitte Australia famously had to refund the Australian Government part of a $290,000 fee when it leaked that the firm used AI to generate part of the report, Fortune reports.
That judgment and ability to know when to trust the AI’s output, when to challenge it, and when to set it aside and think for yourself is what ultimately drives value. These types of higher cognitive skills cannot be learned by watching a course or completing a module. Judgement must be built through experience.
Less Content; More Practice
When new technologies are launched, product leaders expect users to explore them firsthand. They understand that people learn fastest when they can interact directly with the system, test boundaries, and see outcomes for themselves. The same principle applies to AI.
Allowing individuals to engage in role-based, highly contextualized AI training gives them the opportunity to explore how AI changes their work processes and make decisions on when to use it, what to trust, and what to change. Instead of passive learning through courses or classroom teaching, individuals can experience firsthand what happens when an AI agent works correctly. Better still, they can learn what to do when things go wrong and how to correct them.
It wouldn’t surprise me if many leaders push back here, concerned about token cost and operational risk. Mistakes can have consequences, whether that is incorrect decision-making, flawed outputs, or misuse of data. That’s why you need to put guardrails around the AI training, such as
- Only allowing this experimentation within a controlled space, not live
- Placing limits on tokens
- Limiting—and defining—the kinds of tasks people can do with your data and software
People would still have the freedom to explore and make mistakes, but without risking a nasty data breach or accidental production deployment.
Providing a safe space also benefits your learning culture, as people are more willing to explore new capabilities when they know the consequences are contained.
When someone gets it wrong in that environment, the outcome reflects it. That feedback loop—recognize, decide, iterate, observe, pivot, understand—is what builds real capability. Completion rates tell you nothing. Whether someone can get to the right answer using AI, catch a flawed output, and know when to take over—that’s the signal that matters.
Capturing AI’s Value Starts with Experience
Grand View Research says that the global AI market is on track to reach $3.49 trillion by 2033. The organizations that capture that value won’t be the ones that trained the most people on AI. They’ll be the ones that put AI to work and build workforces with the judgment to use it well.
The question isn’t whether your people are trained on AI. It’s whether they can use it to get the right answer when it matters.
Image credit: tadamichi
