The same mistake everyone makes with AI
It happens in university classrooms and Fortune 500 conference rooms. The error is the same.
Shubham Chandra, Head of Curriculum7 min read
Key takeaways
- The biggest barrier to AI adoption is the mental model people bring to it. They treat AI like a search engine and judge it on search engine terms.
- Students and executives make the same error, and the only thing that has shifted it, in either room, is structured practice on real work.
- AI is a capable but context-starved collaborator. Brief it like a colleague, not a search bar.
The universal error
- of leaders would not hire someone without AI skills
- 66%
- of leaders would not hire someone without AI skills
- Microsoft and LinkedIn, Work Trend Index, 2024
- wage premium for workers with AI skills
- 56%
- wage premium for workers with AI skills
- PwC, Global AI Jobs Barometer, 2025
The sequence runs like this: a vague question goes in, a surface-level answer comes back, and the person decides AI is not that useful and stops. The whole cycle usually takes a couple of days.
I've seen it in a University of Chicago classroom and in a Fortune 500 conference room. Students ask AI to "analyze this" and get back observations they already knew. They accept it as the ceiling. So do VPs.
The twenty-two-year-old and the C-suite executive walk away thinking the same thing: "it's just giving me what I already know." Neither realizes that the question produced that answer.
Two thirds of leaders told Microsoft and LinkedIn they would not hire someone without AI skills. What they mean by skill is judgment: knowing where AI fits in real work and when to trust what comes back, and judgment is something no tutorial has managed to teach.
Why the search engine metaphor breaks everything
Google rewards keywords and hands back an answer. AI rewards context and hands back a starting point. Most people bring the Google habit to the second tool, then blame the tool.
Treated like search, AI gets queries with no context about who is asking or what they need, and the first result gets accepted the way a top search result does.
The result is a population of professionals who tried AI once, got a mediocre answer, and now quietly think it is overhyped. They are wrong for an understandable reason, which is that nobody ever corrected the metaphor.
A bad AI interaction and a useful one usually differ by thirty seconds of context the person did not provide.
The reframe that helps
Think of AI as a capable colleague who just walked into the room with no context. Quick and willing, and unaware of the situation or of what "good" looks like here.
Brief it the way a smart colleague gets briefed on day one: the role, the audience, what success looks like, what has already been tried. When people make this shift, output jumps from generic to useful, and the only thing that changed was the input.
Most training programs skip this and open with the features of ChatGPT. Features change every quarter, so that lesson expires fast, while a mental model keeps paying off for years.
What this means for training
The tools work fine. ChatGPT, Copilot, Claude and Gemini are all capable enough, so the thing that needs fixing is the mental model, and a lunch-and-learn has never fixed one.
What fixes it is practice on real problems: the report due Friday, the analysis the board wants next week, the proposal that has sat half-written for a month. Working one of those through with AI is what moves the mental model.
PwC's 2025 Global AI Jobs Barometer found a 56% wage premium for workers with AI skills. The premium rewards judgment about when AI helps and when it does not, and the people earning it got there by doing the work differently, over and over, until the new way became the default.