Loading…
It feels like January 2020: you know something big is coming, but the world keeps going as normal. 42% of Danish companies say they use AI, yet only 5.5% globally see real bottom-line impact.
So much is happening in AI right now that I sometimes walk around in a fog, unsure whether I'm the one going crazy.
The feeling reminds me of January 2020. Imagine knowing what's coming when the coronavirus arrives. But the world around you just kept going as normal. People booked summer holidays. Managers set budgets. Nobody wore a mask.
That's how I feel about AI now.
I won't play expert and claim I know exactly how this ends. But there is a gap between what's happening at the front line of AI development and what's actually happening inside Danish companies. And that gap is far bigger than most people think.
Earlier this week, Greg Brockman (OpenAI) wrote a long post about the world entering a "compute-powered economy". His point: the friction between idea and reality is disappearing. Small teams can build what used to take large ones. You no longer have to translate your intentions into instructions. The computer adapts to you. Close to a billion people already use ChatGPT weekly, according to OpenAI itself.
At the same time, there are rumors about next-generation models. GPT-5.5 from OpenAI, and a project at Anthropic that some call "Mythos". To be clear: these are rumors, and neither is a confirmed product as I write this. But things are moving. Fast.
This is where the fog comes in. Because when you hold that up against what's actually going on in Danish businesses, the picture doesn't fit.
In the fall of 2025, Statistics Denmark published numbers that look positive on the surface. The share of Danish companies using AI went from 15% in 2023 to 28% in 2024 and 42% in 2025. Nearly a tripling in two years.
Split by company size:
If you only look at those numbers, it sounds like I'm wrong. Companies are on board. Especially the big ones.
Then McKinsey's State of AI report from November 2025 ruins the party. Globally, 88% of companies say they use AI in at least one function. Fine. Only 5.5% report real financial bottom-line impact, and about two thirds of all AI efforts are still in "pilot or experiment mode".
That's the gap I feel when I'm out hearing cases in my network. On paper, everyone uses AI. In practice it's ChatGPT for writing an email, a flow in Make or n8n, and maybe an agent that makes one small decision about where data should go.
That's not crazy. It's not the sharp end of the fighter jet. It's automation with AI frosting.
Many companies have leaned hard on Microsoft 365 Copilot. My gut says it isn't worth the hype. To be fair, I haven't used it in depth myself, so I'm far from an expert here.
But let's look at the numbers.
A Forrester Total Economic Impact analysis of Microsoft 365 Copilot from 2025 points to 9 hours saved per employee per month, an ROI of 116% over three years, and 70% of users reporting higher productivity. About 70% of the Fortune 500 have integrated it in some form.
Those are real numbers. And they contradict my gut feeling, at least partly.
But: Microsoft commissioned the Forrester study, and you should keep that in mind. And what Copilot solves is still the "classic" kind of task: PowerPoint drafts, email drafts, meeting summaries, Excel formulas. That saves time. That's value. But it's not the value I'm talking about.
It's productivity inside existing workflows. Not redesigning the workflows.
It's the difference between a faster horse and a car.
AI creates serious value when it's autonomous. Not in the science fiction sense, but in the down-to-earth one:
It knows who the company is. It knows who I am. It knows the product, the tone, the customer, the context. I can give it a task, not an instruction but a task, and if it doesn't know how to solve it, it figures it out.
A practical example: a company sells reception furniture to dentists, law firms, and accountants. A well-built AI agent can do this without help:
Or the painter who scrapes Boligsiden's sold listings and automatically sends a postcard two months after the sale: "New start. Does the house need a coat of paint? Call Maler Mads." Or the carpenter who analyzes Google Street View for worn roofs in one postal code and sends exactly those houses an offer. Not many people get angry about receiving an offer for something they need.
That's what AI can do today, if someone bothers to build it. I almost never see it out there.
I think two explanations do most of the work:
1. Companies don't set aside resources to investigate AI. Not in the sense I mean. Few small and medium-sized companies have one full-time person whose job is to experiment broadly and find out what actually works. Statistics Denmark backs part of this up: about half of Danish companies name lack of legal clarity as a barrier, and 40% point to concerns about GDPR and data protection. That's legitimate. It's also an excuse for doing nothing.
2. Building real agents still takes heavy IT knowledge. Architecture, tools, databases, orchestration. You can prompt your way to a simple agent in an afternoon, but it will only be that good. What creates real value is agents that understand context, have memory, make decisions, and run unsupervised. That still takes people who know what they're doing.
My thesis is that we're approaching what you could call the Apple Store moment for AI agents. The point where the barrier to launching an agent is so low that a salesperson, an HR employee, or a logistics manager can do it themselves without calling IT.
"I need an agent that can do this." And then there is one.
When that point hits (and it will, the only question is when), adoption will outrun everything we've seen before. Faster than cloud. Faster than smartphones. Because the entry barrier effectively drops to zero.
Until then, people like me build them for you.
This is the question I chew on most these days: should you fit AI into a company's existing IT architecture and existing work processes? Or should you build something entirely new next to them?
Most companies pick the first. It's the easiest. You bring in Copilot, you automate a small piece here and there, and otherwise you carry on as before. And that's fine. It's just not transformation. It's optimization.
My thesis: a fair number of companies would be better off running parallel AI-first departments. A parallel sales department where you ask: what would this look like if we built it from scratch in 2026 with AI as the foundation, not the add-on? A parallel marketing department. A parallel logistics function.
Not to replace the existing ones. Not to fire people. But to find out how far you can get without spending 80% of your energy on change management and internal compromises.
Then, if it works, pull people over. With good UX, people adopt on their own once they can feel the value. That's what happened with the iPhone. That's what happens with every tool that is noticeably better. It just has to be noticeably better.
If people can see the boring work disappearing, and that they get to focus on strategy and human contact and the things they actually like, they come on their own.
McKinsey's numbers support this, by the way. The companies that get real value from AI are the ones that rewire their workflows, not the ones that layer AI on top. That's the main difference between "high performers" and everyone else in their 2025 report.
I don't know if I'm crazy. I don't know exactly how fast this will change. I do know that 42% of Danish companies "use AI", and that only 5.5% globally see real bottom-line impact. I know Greg Brockman is signaling much stronger models very soon, and that a billion people already use these tools weekly.
And I know I see very few companies out there preparing for what happens when the agents get twice as good and the price drops 90%.
If you're sitting in a company reading this: the dumbest thing you can do is stall because you're waiting for the fog to lift. Assign one person. Let them investigate broadly. Build something small on the side, without change management. See what happens.
That's how you find out whether you're crazy, or whether you were the only one who saw the storm coming.
Tell me what you are building. I will tell you straight what is worth doing.
Start a conversation →