Why Company-Wide AI Training Doesn't Work: Bet on the Self-Starters

When a company decides to adopt AI, the boss’s first instinct is usually company-wide training—learn prompting, learn the tools, track study hours. But what training teaches is precisely the layer AI has already handled.

Robert Dilts, a towering figure in NLP who created the logical levels model, once said that a company can teach its employees how to use tools, follow processes, and build professional skills, but it can hardly teach them true responsibility, intrinsic drive, and judgment. He said this before AI existed. Now look back: those first three things—any large language model can walk you through them step by step. That training ticket you paid for? You bought something that’s depreciating.

Some numbers: over nine out of ten employees use AI daily for work, on their own personal accounts. Companies that actually pay for accounts? Only four out of ten. Employees didn’t need your training—they were already using it.

But there are two problems. First, employees paying out of pocket are stuck on the worst tier—free versions, low-end models, usage capped by request count. The subscription money they save gets traded for do-overs, unusable answers, and endless revisions. Work hours cost far more than subscriptions; give them the good tier and the time saved pays for itself immediately. Second, the employees who are truly good at it—how they prompt, how they break down problems, how they iterate—all that know-how lives in their personal accounts. The company can’t see it, can’t replicate it, and when they leave, it all walks out the door. What bosses should really worry about isn’t whether they’re spending money—it’s that one person’s great work never becomes the company’s asset.

The money shouldn’t go to training everyone. It should go to shifting those few who are already using it—from solving their own tasks to solving the company’s problems. Give them good tools, data and permissions, real scenarios, and a place where their craft can be preserved.

AI is an amplifier. People who were already profitable stay profitable without AI; people who weren’t don’t become profitable just by learning it. Training assumes addition—everyone gets ten more points. But AI is multiplication: multiply a self-starter and you get ten times the output; multiply someone who waits to be pushed and you still get zero. You pay for addition, but it settles in multiplication.

My advice is one sentence: skip the meetings, skip the training programs, hand out the tools, check back in three months to see who picked them up, and pour your resources into those two or three people. A premium seat runs about one to two thousand yuan a month. Even ten such people add up to less than one person’s salary. Spend one person’s salary to double the efficiency of ten people who already know how to run—do that math.

Dilts said: nurturing talent changes capability; discovering talent activates identity.

EP0098 · 2026-09-14 · 03:12. The picture uses a digital human, and the voice uses my cloned voice. This is an English translation of the approved Chinese script used to create the narration.

When a company decides to adopt AI, what’s the boss’s first instinct? Sign everyone up for training. Learn prompting, learn the tools, track study hours—you’d think they want a certificate in every hand. But have you ever considered that what training teaches is exactly the layer AI has already taken care of?

There’s a legendary figure in the NLP field named Robert Dilts—he’s the one who created the logical levels model. He once said something I’ve always remembered. He said a company can teach its employees how to use tools, how to master processes, and how to develop professional skills, but what’s very hard to teach is what true responsibility, intrinsic drive, and judgment really mean. And notice—he said this before the age of AI. Now look back at those first three: how to use tools, how to follow processes, how to build skills—any large language model can walk you through all of them, step by step. That training ticket you paid for? You bought something that’s actively depreciating.

So what’s truly scarce? It’s those other three—responsibility, drive, and judgment. These can’t be taught. No matter how much training budget you throw at them, they simply can’t be taught.

Let me share some numbers. Over nine out of ten employees are using AI every single day for work—on their own personal accounts. Companies that actually pay for company accounts? Only four out of ten. See, employees didn’t need you to teach them—they started using it on their own. And then some people ask, well, if employees are already using it themselves, why should the company spend money on this?

That’s a fair question, but you need to think through two things. First, when employees pay out of pocket, they’re stuck on the worst tier. Free versions, low-end models, usage capped by request count. That little bit they save on subscriptions gets repaid in starting over again and again, answers that don’t work, revisions back and forth. You think them using their own accounts saves you money? They’re actually spending your work hours to make up the difference—and work hours cost far more than subscriptions. Any boss can do this math: give them the good tier, and the time saved pays for itself right there.

Second, those employees who are really good at using it—how they ask, how they break things down, how they revise—all that skill lives in their personal accounts. Sure, they’re doing great work in their own accounts, but the company can’t see it and can’t replicate it. When they leave, it all walks out the door with them. This is what bosses should really feel the sting of—it’s not about whether you spent the money, it’s that one person’s great work never becomes the company’s asset.

So the money shouldn’t be spent on training everyone. It should go toward taking those few who are already using it and shifting them from solving their own tasks to solving the company’s problems. Give them good tools, give them data and permissions, give them real scenarios, and give them a place where their craft can stay behind.

I said this in an earlier video—AI is simply an amplifier. People who were already making money don’t need AI to keep making it. People who weren’t making money won’t start just because they learned AI. Training assumes addition—everyone gets ten more points. But AI is multiplication: multiply someone who tries things on their own and you get ten times the output; multiply someone who waits to be pushed and you still get zero. You pay for addition, but it settles in multiplication. Think about it—AI is inherently automatic, designed to scale. But when you sit someone who waits to be pushed in front of it, they won’t budge without a shove. They just sit there and become the pause button themselves. Dilts had another line—he said a person with no intrinsic drive, no matter how many training programs you sign them up for, will only ever be going through the motions.

I see this in my own classes too. Same lesson, same tools handed out, and after class the ones who actually go use them are always just those few. They’re not smarter than everyone else—they just don’t need you to chase them.

So my advice boils down to one sentence: skip the meetings, skip the training courses, hand out the tools, and three months later check who picked them up on their own. Then pour your resources and room into those two or three people. A premium seat costs maybe one or two thousand yuan a month. Even if you find ten people like this, the total still doesn’t add up to one person’s salary. Spend one person’s salary to double the work efficiency of ten people who already know how to run on their own—tell me that math doesn’t work out.

I’ve always remembered that line from Dilts—he said nurturing talent changes capability, but discovering talent activates identity.

That’s it for today. See you tomorrow. Bye bye.