One Size (Actually) Fits All
You’re not gonna like this take.
But hear me out, because we’ve trained some of the biggest companies in the world, at scale, so we do have SOME experience with this (he said, defensively and handsomely).
The popular - and highly rational - common belief is that the best way to train people on AI is to break it down by department. AI for sales. AI for finance. AI for operations. Everyone gets their own curriculum with their own use cases.
It sounds right! It feels right! How could it NOT be right? We've been taught our whole lives that personalization equals quality! We’re not animals, are we?
At AI Mindset, we believe this is not the right way to teach AI.
Don’t get me wrong. We DO tailor how we work with each company we train. But not in the way you think.
And I’ll take it a step further - we don’t train in use cases at all. Because it backfires.
BTW - Microsoft validated AI Mindset’s approach in a published research paper, so I’m not TOTALLY crazy. (Sheesh Conor, defensive much?? Have a milkshake, man.)
THIS IS A COACHING PROBLEM IN A CONSULTING DISGUISE
Here’s the problem, at its core:
When you break AI training into departments, you're applying a traditional consulting model.
And look - consulting is great at what consulting does. Best practices! Benchmarks! Wearing suits! Saying the word synergy! Charging a lot!
But it reflects a fundamental misunderstanding of what AI actually is.
AI is not a sales tool. It's not an operations tool. It’s not a finance tool.
It's a thinking tool.
And you don't teach thinking by department.
Here's my favorite analogy:
Imagine you were relocating your entire company to France. Nobody speaks French.
You know that they would need different vocabulary for each department, right?
And yet…you would never break those groups up, right? To teach them?
Let’s say you did - you give each department a phrase book. That way, the sales team can say the most common things like “Let’s talk about how this fits into your budget!” And the finance team says “We will require numbers to get to the EBITDA by end of quarter” - and so on.
Now they have all the phrases they need!
Or - you get the whole company together and you actually get them fluent in French. Together. And then you send everyone to France and let them figure out how to best use language to accomplish their goals. Fluently.
In that ideal world, you teach them so well that they're now thinking in French. They already know the nouns and verbs. They know their jobs. What they don't have is the language.
That's AI right now. The language is what's missing. We need people thinking in AI. Not referring back to their phrase books for sales.
YOU'VE BUILT A FENCE ON DAY ONE (THE BAD KIND OF FENCE)
But here’s the kicker (and that’s not AI saying “here’s the kicker,” it’s me, Conor. AI can’t steal that forever.)
It’s not even like it’s not the most efficient way. I would argue that teaching in use cases — breaking it down by department — actually holds people back.
Here's what happens in someone's brain when you teach them "AI for sales."
Their brain goes: Cool, new sales tool. AI gets filed in the mental box alongside Salesforce, Gong, and the CRM.
It becomes something you open when you're doing sales work.
But AI isn't a sales tool any more than thinking is a sales tool.
The moment you categorize it by function, you've told someone's brain where AI belongs - and where it doesn't.
That person isn’t thinking AI first. Which means the behavior never forms. Which means they prep for a board meeting and don’t instinctively use AI. Because AI is a sales tool. Or when they rethink their team structure or any of the thousand daily things that aren't "sales."
You've fenced it in. Before lunch. Well done, everybody.
Real adoption is when someone can't remember the last time they did deep work without AI beside them. It's the difference between thinking in French and pulling out a phrase book at a restaurant. The phrase book person knows exactly which situations call for French. The fluent person just lives in it.
YOU'RE OUTSOURCING THE GOOD PART
The most important moment in AI adoption is when a person looks at their own work and sees it differently. Not when someone tells them how AI applies. When they see it themselves.
The person doing the work knows stuff no trainer ever will. They know which report takes six hours and gets skimmed. They know which meeting could be a Slack message. They know which part of client onboarding is literally copy-paste from the last client.
They know where the bodies are buried. (Metaphorically. Hopefully.)
When you hand them department-specific use cases, you've done the thinking for them. And the thinking was the whole point.
TRAINING WHEELS VS. HANDHOLDING
Before you come at me - I am not arguing against being hands-on. Our approach is intensely guided. We walk people through every step. Nobody gets left behind.
But there's a huge difference between handholding and training wheels.
Handholding builds independence. You're right there with them, but you're building the skill. After that? They can ride anywhere.
Training wheels build dependency. "AI for sales" gives people three recipes. They follow the steps, get a result, and walk away thinking they've learned AI. They haven't. They've learned three recipes.
The difference shows up on day two, when nobody's in the room. The use-case person opens Copilot, doesn't see their situation on the list, and closes it. The behaviorally trained person opens it and starts thinking.
Because that's what they learned to do.
WE TESTED THIS, BTW
In the Microsoft Research field experiment with Gap Inc., the control group got exactly what a department-by-department model would produce: features, use cases, prompts. The AI Mindset group got behavioral training. No features, no use cases, no prompts.
The behavioral group was more than twice as likely to produce top-quality work.
Microsoft also tested mandated collaboration protocols - structured, step-by-step frameworks for working with AI. The kind of thing that feels rigorous and professional.
It backfired. Those groups scored significantly lower and were eight times more likely to produce nothing at all. Turns out, when you give people a rigid process for something that's supposed to be flexible, they just... freeze. Who could have predicted that? (Everyone. Everyone could have predicted that.)
THE BOTTOM LINE
Tailoring AI training by department feels like the right move. We're obsessed with personalization. I get it. And we DO tailor.
But AI isn't a body of knowledge to be applied. It's a way of thinking to be developed. You develop that universally, then let people apply it in their own context. The reverse doesn't work. It has never worked. And now there's a peer-reviewed study that proves it.
Use cases expire. Behavioral change doesn't. Once someone stops treating AI like a search engine and starts treating it like a thought partner, that shift transfers to the next tool, the next feature, the next generation.
That's the difference between training that depreciates and training that compounds.
AI NEWS OF THE WEEK
KPMG hands Claude to all 276,000 employees in one go
KPMG just rolled Claude out to all 276,000 of its employees worldwide and embedded it inside the platform they use for client work. Not a pilot. Not "AI for tax." A Big Four firm saying everyone gets fluent, together, which is precisely the move we've been arguing for. The receipts will show up in client work over the next twelve months. I am extremely here for it.
Google I/O 2026 happened, and the tools just lapped the workforce
Google I/O dropped roughly a hundred announcements this week. The ones that matter: Gemini 3.5 Flash, Gemini Spark (a 24/7 cloud agent that works in the background), Antigravity 2.0 for multi-agent workflows, plus Samsung's intelligent eyewear. Sundar said we are "firmly in our agentic Gemini era." The tools are now galaxies ahead of where most workforces can use them.
Andrej Karpathy joined Anthropic, and the AI is now training the AI
Andrej Karpathy, the guy who co-founded OpenAI and basically wrote the playbook for modern LLMs, joined Anthropic this week. His new job? Build a team that uses Claude to accelerate Claude's pretraining. So the AI is training the AI now. If you needed a reminder that the field is compounding on itself, this is the loudest one of 2026 so far.
