The AI Order Got Cancelled. Your Strategy Was Never Signed Either

Washington delayed Trump's AI executive order hours before signing, then signed it anyway amid what multiple outlets describe as open White House infighting. That's the headline. It's not the story. The story is buried in a Fortune report that should be setting off alarms in every boardroom: thousands of CEOs are now admitting, on record, that AI has produced zero measurable impact on employment or productivity in their companies. Zero. Not "less than expected." Zero.

I've been building software for twenty years. I run an AI development company in Europe out of Jönköping, a city most Americans couldn't find on a map, and I'm telling you this productivity paradox isn't new. It's the same ghost that haunted the PC revolution and the internet rollout in the 1990s. Economists call it the Solow Paradox: you see computers everywhere except in the productivity statistics. We're watching it happen again in real time, except this time it's compressed into eighteen months instead of a decade, and everyone with a LinkedIn account is pretending they saw it coming.

The Real Failure Isn't Technical, It's Organizational

Harvard Business Review published something more useful than the Fortune piece this week. They found that managers and executives inside the same companies fundamentally disagree on AI strategy. Not disagree on execution. Disagree on the goal. One VP thinks AI is about cutting headcount. Another thinks it's about speeding up existing workflows. The CFO thinks it's a cost center to justify. The CTO thinks it's an arms race they can't afford to lose. Nobody in the room agrees on what "success" looks like, so nobody can measure failure either, and the money keeps burning.

This is the actual crisis. Not the technology. I've deployed AI agents into production systems that save clients real hours and real money. The tech works when the problem is defined. What doesn't work is bolting AI onto an organization that hasn't decided what it's trying to fix. You can't automate confusion. You can only make it move faster.

Meanwhile Microsoft's CEO is touring London this week pushing Copilot into every enterprise deal he can close, and EY has gone all-in publicly on agentic AI as a strategic bet. That's the US and UK enterprise sales machine doing what it does best: selling conviction ahead of evidence. It's not dishonest, necessarily. It's just business as usual in markets that reward speed over correctness. The problem is when European companies watch that and think they need to match the pace instead of matching the clarity.

Sweden's Slowness Is Suddenly a Feature

Here's what I see from Jönköping that people in San Francisco or London don't see from where they sit. Swedish companies are slow to adopt things. We argue in committee. We want consensus before we commit budget. Every founder who's tried to sell into a Swedish enterprise knows the pain of a six-month procurement cycle that would take six weeks in Texas.

For the last decade that's been a liability. In 2026, with the American AI enterprise machine discovering that half their AI rollouts produced nothing measurable, that Swedish habit of forcing alignment before spending looks less like bureaucracy and more like risk management. You can't blow the productivity paradox budget on a system nobody agreed the purpose of, if your culture never let you buy it in the first place without agreeing on the purpose.

Breakit and Di Digital have both covered the caution in Swedish boardrooms toward generative AI spend this year, and the tone from Stockholm has consistently been "show us the unit economics" rather than "show us the demo." That instinct, annoying as it is when you're trying to close a deal, is exactly the discipline that HBR is now telling American executives they skipped.

I'm not saying Sweden is smarter. I'm saying Sweden is forced into the discipline that everyone else is now discovering they needed. The EU AI Act, whatever its flaws, has made every company operating here ask "what are we actually building and why" before deployment, because the compliance burden makes sloppy AI expensive. That's regulation working as friction, and for once the friction is doing something useful. It's slowing companies down to the point where they have to answer HBR's question before they spend the money, not after.

Where EU Policy Still Isn't Ready

I don't want to oversell the Swedish miracle. Our regulators aren't ahead of this either. The EU AI Act was built around a model of AI as a static system you certify once. It was not built for agentic AI that makes autonomous decisions in loops, which is exactly what companies like Dify, Langflow, and AutoGPT are enabling at massive open source scale right now. Nobody in Brussels has a clean answer for who's liable when an agent chains six decisions together and the fifth one is wrong. That gap is real and it's coming for us within eighteen months, not five years.

Where This Goes: 2 to 5 Years Out

Here's my actual prediction, not the safe consultant version. The productivity paradox resolves the same way it did with PCs and the internet: not because the technology gets better, but because organizations restructure around it instead of layering it on top of the old structure. Companies that treat AI as a bolt-on will keep reporting zero productivity gains for another two to three years. Companies that redesign the actual workflow, the actual org chart, the actual decision rights, will start pulling ahead sharply around 2027-2028. The AGI conversation makes this more urgent, not less. Every lab still racing toward general intelligence is racing past the exact organizational question these CEOs are failing right now. If you can't get consensus on what a narrow AI agent should optimize for today, you have no chance of governing something more capable tomorrow. The companies winning in five years won't be the ones with the best model access. Every company will have access to roughly the same frontier models by then, this is already becoming true with open weights closing the gap. The winners will be the ones who solved the boring organizational alignment problem years before their competitors admitted it existed.

Regulators everywhere, EU included, are going to keep chasing the last generation of AI risk while agentic systems quietly restructure how decisions get made inside companies. Expect a real regulatory scramble around 2027 when the first major agentic AI failure makes headlines the way a bank collapse does. Someone will lose real money because an autonomous agent chain made a call nobody reviewed, and every government will suddenly discover urgency they didn't have this week.

What To Actually Do About This

Stop reading AI news for a signal on what to build. Start with an internal alignment exercise before you spend another euro. If your leadership team can't write, on one page, the specific business metric your AI initiative is supposed to move, and everyone in the room agrees on that metric, you are the company Fortune is writing about. You will report zero impact next year too. If you're past that stage and actually need to build, here's what I'd point a serious CTO toward this week. Dify is genuinely useful for prototyping agentic workflows without locking yourself into one vendor's roadmap, which matters given how fast the underlying models are shifting. Langflow is worth a look if your team needs a visual layer to get non-engineers involved in designing the agent logic, which, going back to the HBR finding, is exactly the alignment problem you need to solve anyway. And if you want to see where the open source agent ecosystem is heading at scale, AutoGPT is still the reference point everyone else is building against, even if the hype around it has cooled.

None of these tools fix your alignment problem. No tool will. That's a leadership job, not an engineering job. But once you know what you're solving for, these are legitimate ways to build fast without locking yourself into a Copilot-shaped hole that costs you flexibility in three years.

Build Only After You Can Answer This

We run every client engagement at HEIMLANDR through the same filter before we write a line of code: what specific decision or task are we changing, and how do you measure it. If a client can't answer that, we don't start with rapid MVP work, we start with a strategy conversation, because building the wrong thing fast is worse than building nothing. This is also why custom AI solutions built for your actual workflow consistently outperform enterprise platforms sold on a demo stage in London. The platform isn't the problem. The unexamined assumption underneath it is.

Signal vs Noise

The noise this week is Washington's executive order drama and the parade of enterprise AI sales tours. The signal is quieter and less flattering: most companies deploying AI right now don't know what they're trying to fix, and no regulation, no model upgrade, no keynote from Redmond is going to solve that for them. Sweden's cautious, consensus-obsessed culture, the one that makes sales cycles painful and procurement slow, is accidentally solving the exact problem American boardrooms are just now discovering they have. That's not luck. That's what happens when you're forced to agree before you spend.

Get your alignment right first. The technology has been ready for a while now. Your organization might not be, and that's the actual executive order nobody signed yet.

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.