The AI Productivity Paradox Is Back. This Time It's Eating the C-Suite.

Billions In. Nothing Out. And Nobody's Allowed to Ask Why.

Here's the number that should be on every board deck this week: over 80% of companies report zero productivity gains from AI. Not "marginal." Not "hard to measure." Zero. Thousands of CEOs are admitting, right now, that their AI investments have had no measurable impact on employment or productivity. Meanwhile, the companies selling them this stuff just successfully lobbied to kill AI oversight in the US. Think about that for a second. The people who sold you the thing are also the people making sure nobody gets to audit whether it works.

If you're a CEO, CTO, or founder sitting in Stockholm, Gothenburg, or here in Jönköping where I run HEIMLANDR.IO, you need to decide what this means for you. Not philosophically. Practically. Because the clock is ticking on a decision that will define the next five years of your company.

Solow's Ghost Is Laughing at Us

Robert Solow said it in 1987: "You can see the computer age everywhere but in the productivity statistics." Economists are dusting off that quote right now because we're living through the exact same paradox. Except this time, the spending is bigger, the claims are louder, and the gap between promise and proof is wider.

Let me be direct about what I see from the ground. I build custom AI solutions for companies. Real ones, not slideware. And the pattern is always the same. A company buys a platform license. They bolt a chatbot onto their customer service. They give everyone Copilot seats. They call it "AI transformation." Then six months later, nothing has changed. The chatbot hallucinates. The Copilot generates code nobody trusts enough to ship. The customer service team is still doing the same thing they did before, just with an extra tab open.

The problem is not the technology. I want to be clear about that. The underlying models are genuinely powerful. The problem is that most companies are buying AI like they bought ERP in the 2000s. Top-down. Platform-first. Before anyone has figured out what specific problem they're solving.

And according to Harvard Business Review, managers and executives don't even agree on what AI's role should be inside their own organizations. Executives see cost reduction. Managers see workflow assistance. Nobody has aligned on what "working" looks like. So of course nobody can measure it.

The Oversight Kill: What Just Happened in Washington

This week, US tech CEOs successfully blocked an AI oversight executive order. Trump delayed signing it hours before the ceremony. The details matter less than the signal: the industry does not want measurement. It does not want independent auditing of AI performance claims. It does not want standardized productivity benchmarks.

Ask yourself why.

If AI was delivering the 10x productivity gains that every pitch deck promises, the industry would be begging for measurement. They'd be funding their own third-party audits. They'd be publishing the numbers. Instead, they're spending lobbying dollars to make sure nobody looks too closely.

I'm not a conspiracy thinker. I'm a builder. But when the people selling me something actively fight against anyone checking if it works, I pay attention. You should too.

The Swedish Situation: Shield or Coffin?

Here's where it gets interesting if you're building in Sweden or anywhere in the EU.

We sit inside the EU AI Act. We have GDPR. We have a regulatory framework that, whatever you think of it, at least attempts to put guardrails on AI deployment. The US just threw theirs in the trash. China never had any worth mentioning.

KPMG's Global Tech Report 2026 and a recent Jönköping University thesis on digital transformation leadership both point to the same gap: executive teams across Nordic companies are structurally misaligned on where AI value actually sits. The C-suite says "strategic transformation." Middle management says "I need help with my spreadsheets." Engineering says "just let me ship." Nobody is wrong, but everybody is pulling in different directions, and the result is a lot of money spent on platforms that nobody fully adopts.

Swedish companies have a cultural advantage here that most of them are ignoring. We have flat organizations. We have consensus culture. We have a tradition of involving workers in technology adoption. This is exactly the structure you need to actually make AI work, because AI doesn't create value from the top down. It creates value at the edge, in specific workflows, where individual contributors know what's broken.

But Swedish executives are also sitting on their hands. I talk to founders and CTOs in this market every week. The typical posture is: "We're watching. We're waiting. We'll move when it's clear." That's understandable. It's also dangerous. Because the 18-month window I keep talking about is real. By late 2027, two things will be true:

1. Companies that have built real AI capability, not bought platform licenses but built actual custom AI automation into their business workflows, will be pulling away from competitors in measurable ways.
2. The EU regulatory framework will either be a competitive moat (because your AI is compliant, auditable, and trustworthy) or a coffin (because you spent so long complying that you forgot to build anything).

Which one it becomes depends entirely on what you do in the next 18 months.

Where the Actual Value Is

Let me tell you where I see AI actually working. Not in theory. In practice, in projects we build at HEIMLANDR.

It works when you start with a specific, measurable workflow problem. Not "transform our business." Something like: "Our sales team spends 11 hours a week copying data between three systems and writing follow-up emails." That's a problem you can solve with AI agent development. You build an agent that does that specific thing. You measure the hours saved. You expand from there.

It works when the people doing the work are involved in designing the solution. Not consulted after the fact. Involved. The flat structure Swedish companies already have is perfect for this, if anyone would actually use it.

It works when you own the implementation. Not when you're renting it from a platform vendor who can change their pricing, their API, or their model quality at any time. Self-hosted, custom-built AI automation is more work upfront. It's also the only way to build durable competitive advantage.

That 80% failure number? I'd bet most of those companies bought a platform. Very few of them built anything.

Where This Goes: 2027-2030

Three things are converging that will make the next four years very different from the last two.

First, the measurement reckoning is coming. You can delay oversight. You can lobby against audits. But eventually, CFOs will do what CFOs always do: demand ROI numbers. And when those numbers don't show up, the budget cuts will be brutal. We're going to see a massive AI hangover in enterprise spending by late 2027. The companies that survive it will be the ones who can actually point to productivity data.

Second, the path toward AGI changes the calculus. Even if current models plateau, the trajectory toward more capable AI systems means that the companies building real AI infrastructure now, internal tooling, data pipelines, custom agents, will be positioned to plug in more powerful models as they arrive. The companies that only bought SaaS seats will have nothing. No institutional knowledge. No custom infrastructure. No competitive moat. Just a subscription they can cancel.

Third, the regulatory gap between the US and EU is going to widen. The US is going full deregulation. The EU is tightening. This creates a genuine strategic question for every European company: do you build to EU standards and sell compliance as a feature, or do you try to compete in a deregulated market on speed? I think the answer is obvious. Build compliant AI. Make it a selling point. The world is going to want trustworthy AI, and the EU is the only jurisdiction that's even attempting to define what that means.

What to Look At

If you're a technical leader trying to figure out where to start, here are some concrete things worth your time this week:

Langflow (148K+ stars on GitHub) is a solid open-source tool for building AI agent workflows visually. If you want to prototype AI automation for a specific business process without writing everything from scratch, it's a good starting point. Not a production solution by itself, but a fast way to test whether an AI agent approach makes sense for your use case.

Claude Code (126K+ stars) is worth evaluating if your engineering team wants to integrate AI into their actual development workflow instead of using a separate tool. It lives in the terminal, understands codebases, and handles git workflows. The real test is whether your team actually ships faster with it. Measure that. Don't assume it.

awesome-selfhosted (294K+ stars) is the single best resource if you're evaluating self-hosted alternatives to commercial AI and SaaS platforms. Given what I said about owning your implementation, this is required reading for any CTO thinking about AI infrastructure.

And if you're past the evaluation phase and ready to build, we do this every day at HEIMLANDR. AI agents, custom AI solutions, and rapid MVPs that actually ship. From Jönköping to wherever you are.

The Real Question

I'm going to end with something uncomfortable.

The AI productivity paradox isn't just a macro trend. It's personal. If you're a CEO or CTO who approved AI spending in 2024 or 2025, you need to ask yourself an honest question: can you prove it worked? Not with anecdotes. Not with "the team likes it." With numbers. Hours saved. Revenue attributed. Costs reduced.

If you can't, you're part of the 80%.

That's not a death sentence. It's a starting point. But you have to be honest about it before you can fix it. Stop buying platforms. Start building solutions to specific problems. Involve the people who do the work. Measure everything. And for the love of God, stop listening to the CEOs who are simultaneously telling you AI will transform everything and lobbying to make sure nobody ever checks.

The tools are real. The potential is real. But the way most companies are deploying AI right now is theater. And the people running the theater just made sure nobody can turn on the house lights.

You have 18 months. Use them.

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.