The AI Productivity Paradox Is Back. CEOs Are the Bottleneck.

Six thousand executives just told us the truth. Most of them won't act on it.

Here is the number that should keep every CEO and CTO awake tonight: over 80% of companies report zero measurable productivity gains from AI, despite billions in investment. Not "disappointing gains." Not "below expectations." Zero. That's according to a survey of 6,000 executives covered by Tom's Hardware this week. And the kicker? Those same leaders admit they personally use AI tools roughly 90 minutes per week.

Ninety minutes. That's less time than most of them spend in a single status meeting.

Meanwhile, Fortune is running headlines about the Solow Paradox making a comeback. For anyone who missed it the first time: in the 1980s, economist Robert Solow observed that "you can see the computer age everywhere but in the productivity statistics." Forty years later, swap "computer" for "AI" and we're right back where we started. History doesn't repeat, but it sure does rhyme.

I'm sitting in Jönköping reading these numbers and I'm not surprised. I'm frustrated, but not surprised. Because the pattern I see from the companies we work with at HEIMLANDR is always the same. The technology works. The leadership doesn't.

AI isn't failing. Leadership is failing AI.

Let me be blunt about what is actually happening. Most companies bought AI the way they buy enterprise software. They signed a deal with a vendor. They rolled out licenses. They sent an email saying "we now have AI, please use it." Then they went back to running the company exactly the same way they always have.

That's not adoption. That's a purchase order.

The HBR piece from this week makes it painfully clear: managers and executives are in fundamental disagreement about AI strategy. Executives want transformation. Managers see a tool that threatens their authority, disrupts their workflows, and creates ambiguity about who owns what. Both sides are talking past each other, and the result is organizational paralysis dressed up as "responsible adoption."

I've seen this in Swedish companies too. I talked to a mid-size manufacturing CEO in Småland last month who proudly told me they'd invested 2 million SEK in AI initiatives. When I asked him what he personally used AI for, he said "I had my assistant show me ChatGPT once." That's not a technology problem. That's a leadership problem.

If you're a CEO and you don't use AI every single day, you have no business setting AI strategy for your company. Full stop. You wouldn't let a CTO who's never written code make architecture decisions. Why is AI different?

The 90-minute problem

Ninety minutes per week tells you everything. It means these executives are dabbling. They are treating AI like a novelty, not a core operating tool. And the signal that sends to the rest of the organization is devastating.

When I build something at HEIMLANDR, I use AI agents and AI-assisted workflows for hours every day. Not because I'm an enthusiast. Because it makes me faster and better at my job. Code review, market analysis, drafting, debugging, data extraction. These are tasks that used to take my team hours. Now they take minutes, if you set it up right.

The "if you set it up right" part is where most companies are failing. Off-the-shelf AI tools give you maybe a 10-15% improvement on individual tasks. That's nice. It's not transformational. The transformation comes from custom AI solutions built around your actual workflows, your actual data, your actual bottlenecks. It comes from AI automation tailored to your business, not from a generic chatbot sitting in a sidebar that nobody opens.

The companies seeing real returns are the ones that went deep. They mapped their processes. They identified the 20% of tasks consuming 80% of human time. They built or commissioned specific AI agent development for those tasks. They measured. They iterated. They didn't just buy a platform and hope.

The Sweden problem (and the Europe problem)

Here's what makes this worse from where I'm sitting in Jönköping. Sweden is supposed to be a tech leader. We have one of the highest rates of digitalization in Europe. We produced Spotify, Klarna, King. We have some of the best engineers in the world per capita.

And yet.

Look at Northvolt. Sweden's flagship industrial tech bet is struggling to hit basic production targets. This is a company with billions in funding, massive government support, and a mission that basically everyone in Sweden is rooting for. The innovation was never the problem. The execution is the problem. And that same pattern runs through Swedish AI adoption. We're great at talking about being innovative. We're less great at the hard, unglamorous work of actually changing how we operate.

The Nordics have a cultural blind spot here. We are consensus-driven. We like to get everyone aligned before we move. That's a strength in many contexts, but it's a death sentence for AI adoption, because AI rewards speed and iteration, not committee approval. By the time a Swedish enterprise has run its pilot program, formed its ethics review board, completed its risk assessment, and gotten buy-in from all stakeholders, a competitor in Shenzhen has already deployed and iterated three times.

Meanwhile Microsoft's Copilot push continues relentlessly in the Nordics, and EY is showcasing agentic AI adoption as a competitive differentiator. The big consulting firms smell blood. They know that most companies can't do this themselves, and they're positioning to be the middlemen. I have a problem with that, because what most companies need isn't a 200-slide PowerPoint from a Big Four firm. They need someone to actually build the thing. That's what we do at HEIMLANDR's AI agent development practice. Build first. Strategy decks second.

Trump's AI executive order: theater or signal?

On the regulatory side, things are getting weird. Trump just signed an executive order demanding early government access to new AI models before public release. The original version was apparently pushed back on by tech CEOs, who then got a softer version more to their liking. The New York Times coverage makes it clear this is less about safety and more about control and access.

From a European perspective, this matters because it signals where the US is heading: AI as a national security asset, not a commercial product. The EU AI Act is taking a different approach, focused on risk classification and transparency. Sweden is somewhere in the middle, nominally following EU rules while privately hoping Swedish companies can still move fast enough to compete globally.

Here's my honest take: neither approach is right. The US approach is too captured by incumbent interests. The EU approach is too focused on what could go wrong and not enough on what happens if Europe falls behind. And the Swedish government? I don't think they have a real AI strategy at all. They have talking points. There's a difference.

If you're a builder in Europe right now, the regulatory environment is not your biggest risk. Organizational inertia is your biggest risk. The companies that move now, that invest in real AI automation for their business, are the ones who will still exist in five years. The ones waiting for regulatory clarity or perfect conditions will be case studies in MBA programs about how not to handle a technology transition.

Where this goes: the 18-month window

I think we are in the most important 18-month window for AI adoption since GPT-4 launched. Here's why.

The models are getting dramatically better at agentic behavior. We're not just talking about chatbots anymore. We're talking about AI systems that can execute multi-step workflows, make decisions within guardrails, and operate semi-autonomously. The tools repo AutoGPT was early to this vision, and while the early versions were rough, the ecosystem around autonomous agents has matured enormously. Langflow now makes it possible to build and deploy AI-powered agent workflows with a level of sophistication that would have required a full engineering team two years ago.

The path toward AGI, whether it arrives in 2028 or 2035, makes this even more urgent. Every generation of AI models that arrives between now and then will be more capable, more autonomous, and more disruptive to existing business processes. Companies that haven't built the organizational muscle to adopt and integrate AI by 2028 won't be able to absorb AGI-level capabilities at all. They'll be like companies in 2010 that still hadn't figured out the internet.

The gap between AI-native companies and AI-tourist companies will become extinction-level. Not metaphorically. Literally. If your competitor can operate at 3x your efficiency because they have AI agents handling 40% of their operational load, you don't get to catch up gradually. You lose customers, you lose margins, you lose talent, and then you lose the company.

What to actually do about it

Enough analysis. Here's what I would do if I were a CEO or CTO reading this right now.

First: Use AI yourself, daily, for real work. Not demos. Not "exploring." Sit down tomorrow morning and do your first hour of work with Claude, GPT, or whatever model you prefer. Write your board update with it. Analyze your latest financials with it. Have it review a contract. You need to feel what this technology can and can't do in your bones, not in a slide deck someone showed you.

Second: Pick one workflow and automate it properly. Not a pilot. Not an exploration. Pick one process that eats up significant human hours and build a real AI solution for it. If you don't have the team internally, find an AI development company in Europe (or anywhere) that builds, not just advises. We do this work at HEIMLANDR through our Rapid MVP service. Get something running in weeks, not quarters.

Third: Fix the manager layer. The HBR data is clear. Your managers are the resistance point. Not because they're bad people, but because AI threatens the information asymmetry that gives middle management its power. You need to make AI fluency a promotion criterion, starting now. If a manager can't demonstrate they're using AI to make their team better, they shouldn't be managing.

What to look at

Langflow (150K+ stars on GitHub). If you're exploring AI agent development, this is the most mature open-source tool for building and deploying agentic AI workflows. Visual builder, strong integrations, production-ready. Start here if you want to understand what's possible before writing a check to a vendor.

Claude Code (133K+ stars). Anthropic's agentic coding tool that lives in your terminal. If you're a technical leader, this is the fastest way to experience what AI-assisted development actually feels like. It understands codebases, handles git workflows, and executes tasks through natural language. It changed how my team works.

awesome-llm-apps (115K+ stars). A collection of 100+ AI agent and RAG applications you can clone, customize, and ship. If you want to build conviction around what AI agents can do for your specific business, spend an afternoon exploring this repo. The practical examples are worth more than any analyst report.

gstack (112K+ stars). Garry Tan's opinionated Claude Code setup. Twenty-three tools configured to serve as CEO, designer, engineering manager, QA, and more. I don't agree with every choice he made, but the philosophy is right: treat AI as staff, not software. Worth studying regardless of your stack.

The uncomfortable truth, from Jönköping

I write this from a small city in southern Sweden, not from San Francisco or London. I don't have billions in funding. I don't have access to the people writing the executive orders or setting the EU regulations. What I have is a company that builds real things for real clients, and a front-row seat to how AI adoption actually plays out in practice.

And what I see is this: the technology is ready. It has been ready for over a year now. The bottleneck is human. It's CEOs who won't learn the tools. It's managers who protect their turf. It's boards that want "AI strategy" without understanding what AI does. It's an entire leadership class that got comfortable being the decision layer and now faces the reality that a significant portion of their decision-making can be augmented or automated.

That's scary. I get it. But the alternative is worse. The alternative is irrelevance.

You have 18 months. Maybe less. The companies that figure out how to actually use AI, not just buy it, will define the next decade. The rest will be footnotes.

Choose which one you want to be. Then do something about it today.

Fredrik Brunnberg is the CEO of HEIMLANDR.IO, building AI and software solutions from Jönköping, Sweden. This is the daily HEIMLANDR briefing. If you found this valuable, share it with someone who builds things.