Stop Forecasting the AI Weather. Dress for the Climate.

Almost everyone is using AI now. Stanford’s AI Index 2026  found that 88% of organizations have adopted it in some form. Yet hardly any have AI running on its own, without a person steering it. That number is still in the low single digits.

That gap is what makes AI so hard to think about today. The headlines say you’re falling dangerously behind. The reality on the ground says almost nothing runs by itself yet. Both are true. And every expert reads them differently.

The temptation is to pick a side: go all in, or pull back. The better move is to notice that nearly everything you’re being asked to react to is a prediction. So the real question isn’t whose forecast is right. It’s whether your strategy should depend on a forecast at all.

Weather Is Not Climate  

Think of it like the weather and the climate.

Weather is the daily noise: the soaring valuations, the new model launches, the “AGI is almost here” claims, the “it’s all a bubble” essays. It’s loud, dramatic, and changes by the week. Nobody can reliably predict it: not the optimists, not the pessimists, not you.

Climate is the underlying trend: AI keeps getting cheaper and more capable, year after year, through every boom and every scare. That doesn’t make headlines, because it doesn’t change between Tuesday and Thursday. But it’s what actually determines what’s possible in your business over the next few years.

You don’t sell your house because of a storm. You also don’t build one without knowing the climate it has to stand in. The common mistake is confusing the two: making a multi-year decision based on this week’s weather. Panicking at a gloomy essay is the same error as scrambling after a flashy launch. Both let a forecast you can’t trust drive a commitment you can’t easily undo.

The Bubble Is Real, but It’s Not Your Risk  

Is there an AI bubble? Maybe. The money is staggering: global corporate AI investment hit roughly $580 billion in 2025, more than double the year before. A handful of giant firms control nearly the whole supply chain, from chips to apps. If a few of their assumptions turn out wrong, a lot of money is exposed at once.

But notice whose risk that is. It belongs to investors and vendors, the people betting on which AI labs survive and at what price. It tells you almost nothing about whether the specific, practical use of AI on your own roadmap will pay off. That’s the weather. It’s someone else’s forecast to worry about.

The Climate Is Steadily Improving  

Underneath the noise, the trend is remarkably consistent. The cost of using a leading AI model fell from about $20 to $0.07 per million words of text (tokens) in roughly two years, close to 300 times cheaper. The tools keep getting better, the value people get from them is rising fast, and most of them are still free.

The lesson: don’t freeze just because something better might arrive next year. The AI in front of you today is already cheaper and more capable than it was a year ago, and it’ll be cheaper and more capable again next year. That’s the climate you’re planning for.

Why Good Teams Still Get Caught in the Rain  

If the trend is so favorable, why do so many AI projects flop? Usually not because of the technology. They fail for ordinary reasons: no clear goal, so nobody can say what success looks like; no real place in the workflow, so the output goes nowhere; no rules around it, so problems surface the hard way; no clear owner, so the project quietly stalls.

It helps to know AI’s strange shape. It can be brilliant at one task and hopeless at the one right next to it. A top model can win a maths olympiad yet read an ordinary clock face correctly only about half the time. The takeaway isn’t “AI is unreliable.” It’s: point it at the things it does well, and it works for you; point it at its weak spots, and you get rained on.

Two quick examples. IKEA’s operator bought an AI tool to plan delivery routes and cut wasted driving, expecting to save around €100 million a year. Not glamorous, but focused, practical, and paying off. By contrast, MEDVi, a US telehealth platform, boasted of $1.8 billion in projected sales while collecting an FDA warning letter  and lawsuits over its AI claims. That’s chasing the hype and leaving the basics behind.

How to Dress for the Climate  

Five simple moves:

• Start with a real business problem, not a model. Choose uses you can measure. If you can’t measure it, it’s a science project.
• Run lots of small, cheap experiments. Expect most to fail, by design. The few winners pay for the rest.
• Set the rules before you scale. Training, data protection, and basic guardrails aren’t red tape; they’re what let you move fast without getting hurt.
• Stay flexible on vendors. The “best” model changes every few months. Keep things loosely connected so switching is an upgrade, not a rebuild.
• Plan for next year’s AI, not today’s. Today’s limits won’t last. Assume it keeps getting better, because it reliably does.

A Bursting Bubble Would Be Good News  

Here’s the surprising part. If the bubble does burst, a disciplined team isn’t a victim, but a winner. Computing gets cheaper. The hype-driven waste clears out. Talent frees up. And vendors can no longer sell on buzz; they have to prove real value. A burst clears the weather. It doesn’t change the climate.

That’s why the people who win the next decade won’t be the best forecasters. Nobody can forecast the weather. The winners will be the ones who stop trying, and build something that works in any weather.

Ready to turn the AI climate into measurable business impact? Explore our Data Analytics & AI service. Or, if you would rather start with your specific challenge, Let’s Talk .