Learn · Article
Stop training your people on tools
For heads of learning and development. Every course named after a product is an asset with a two-year half-life that nobody books as a write-off.
By Nandan Gijare · 20 September 2026 · about 3 minutes
Open your company's learning portal and count how many of the AI courses are named after a product.
In the book, Arjun spends two weekends in February clicking through his employer's version. Cursor for forty-five minutes. ChatGPT prompting basics for thirty. Eight courses, each one on an individual tool. Not a single course on how to think about any of them.
I have looked at a number of these portals. That ratio is common enough to be the norm.
The sunk cost nobody books
A company that trains its people on Cursor specifically will discover, in about two years, that Cursor is no longer the tool anyone is using, and that the training has quietly become a write-off. Nobody records it as one. The budget was spent in a prior year, the course still sits on the portal, and the people who took it have a certificate that now refers to something they no longer open.
Do that four times, across four tools, over four years, and you have spent a meaningful training budget on an asset with a two-year half-life while believing you were investing in capability.
The alternative is not exotic. Train people on the thinking that lets them evaluate any new tool, choose between alternatives, prepare a decent input, judge an output against criteria they set in advance, and stitch work together across several tools without the join showing. That trains on something that compounds. The tools become details, which is what they are.
Why training departments get this wrong
Not incompetence. Three structural reasons, and they are all reasonable from inside.
A tool course is procurable. It has a vendor, a duration, a completion rate and a certificate, and it can be bought in a quarter and reported on in the next one. A course on judgement has none of that. It is harder to specify, harder to source, and much harder to show as a number on a slide.
A tool course is also measurable in the way organisations like: seats filled, modules completed, a pass rate. Whether anyone's judgement improved is a question that takes six months to answer and cannot be answered by the portal.
And the demand comes in that shape. When a directive says the company must adopt AI, what travels down the line is usually a tool name, because tool names are the only part of the directive that is concrete. Management forwards it without translating it, largely because they do not know how to translate it. So the request that arrives in the training department is "we need Copilot training", and the training department, being a service function, delivers Copilot training.
What happens downstream
The work of figuring out what AI integration actually means in a specific role does not disappear when it is not trained for. It lands on individual contributors, who get no support, no worked examples and no modelling from above.
What follows is the split described in the book. Some people jump in headfirst and outsource their thinking. Some dig in and refuse. Most are stuck in the middle doing neither thing well. And the people who are cut first, when the company decides it has been too slow, are the contractors and the juniors and anyone without political cover, none of whom were given the thing they needed to adapt.
That is the part I would put in front of a head of learning. The failure is not that money went to the wrong course. It is that the absence of judgement training is not neutral. It selects, quite reliably, against the people with the least protection.
A test for your own portal
Pick any AI course in your catalogue and ask what a person who completes it can now do that does not depend on the tool it was named after.
If the honest answer is nothing, you have bought a manual.
There is a second question, harder and more useful. Ask a senior leader to walk you through a real decision they made recently in which they used one of these tools: what they decided before they opened it, what they asked it, where they disagreed with what came back, and how they checked the result. If they can do that, they can model it, and modelling it in public is worth more than any course you can buy. If they cannot, that is where the work starts, and it starts above the training department rather than inside it.