I use AI more than most people I know. I believe less of the hype than most of them too.

Last month I helped a friend set up Claude for his consulting business. He spent twenty minutes with it. Then he looked at me and said: “This is it? This is what people are losing their minds over?”

He was right to be underwhelmed. And completely wrong about what it means.

I have 13 projects in Claude Code across fullstack applications, data analysis pipelines, and agent architectures. I built a SwiftUI app and shipped it to the App Store using AI as my primary coding partner. By most reasonable measures, I am a heavy user.

I also think 80% of the claims about AI are exaggerated.

That is not a contradiction. It is the only honest position I can hold given what I see every day.

The normal distribution argument

Most companies will use AI the way they used Excel. Badly. They will buy licenses, run a few pilots, produce some internal presentations about “AI strategy”, and then go back to doing roughly what they did before. The median outcome of AI adoption will be mediocre. This is not a prediction. It is how organizations have absorbed every technology shift in the last thirty years.

I watched it happen with digitalization. I started LUP Technologies in 2016, building software for truck drivers and fleet managers. The product worked. The technology was sound. The market was not ready. Ten years later, most of the problems we tried to solve still exist. Not because the solutions are missing, but because organizations move at the speed of their slowest stakeholder.

AI will follow the same pattern. The normal distribution applies to organizational competence just like everything else. A few companies at the tail end will do remarkable things. Most will not.

So why bother?

Because the risk is asymmetric.

Nassim Taleb wrote about this in a different context, but the logic applies directly. Consider two scenarios.

Scenario one: AI is overhyped and the current wave disappoints. You spent some time learning the tools. You got faster at certain tasks. You understood the technology well enough to evaluate vendor claims. The downside is a few hundred hours and maybe a thousand dollars in subscriptions.

Scenario two: AI is not overhyped and the current wave accelerates. You did not learn the tools. Your competitors did. The gap between you and them compounds every month. You cannot catch up because the people who started early are using AI to learn AI faster.

The first scenario costs you time. The second scenario costs you relevance.

I know which one I would rather risk.

The practical middle

My own experience is somewhere between the hype and the dismissal. AI did not replace my thinking. It replaced the slow parts around my thinking.

When I was building the SwiftUI app, AI wrote maybe 70% of the code. I still had to know what to ask for, how to structure the project, when the output was wrong. The skill was not in prompting. The skill was in having built software before and knowing what good looks like.

That is the part most commentary misses. AI is a multiplier, not a replacement. If you multiply zero domain knowledge by ten, you still get zero. If you multiply twenty years of operational experience by ten, you get something useful.

I drove trucks before I wrote software. I operated CNC machines before I managed SaaS products. I built a system that a thousand drivers actually used before I sat in meetings at Volvo Group talking about digital transformation. AI does not give you that context. AI amplifies whatever context you already have.

What this means for the person reading this

The question is not whether AI is a bubble. The question is whether you can afford the bet that it is.

If you run a business or lead a team, the responsible move is not to go all-in on AI. It is also not to wait. It is to build enough competence that you can tell the difference between a real capability and a sales pitch. That takes practice, not faith.

Start with your actual work. Not a demo. Not a webinar. Your real problems, your real data, your real constraints. Spend an hour. Form your own opinion.

The worst case is that you wasted an hour. The best case is that you found out what the tool actually does, instead of what people say it does.

That is asymmetric risk. And asymmetric risk has only one rational response.