The AI Productivity Lie: 80% See Zero Gains. Sweden Can't Afford It.

Eighty percent. Let that land.

Over 80% of companies report no meaningful productivity gains from AI. Not "disappointing gains." Not "slower than expected." Zero. This is the finding making rounds across Fortune, Tom's Hardware, and Mashable right now, as of today, May 31st 2026. Billions invested. Entire C-suites reorganized around "AI transformation." And the number is 80%.

Here is what makes it worse. The executives at these companies keep saying AI is saving time. Their own employees disagree. We have a credibility gap so wide you could park a Volvo truck in it. And I am sitting here in Jönköping watching Sweden sleepwalk into a demographic crisis where the one technology that could actually help is being deployed as PowerPoint decoration instead of operational infrastructure.

This is the most important story in tech right now. Not a new model release. Not another funding round. The story is that we are lying to ourselves. And for Sweden specifically, the cost of that lie is about to become very real.

The executive theatre problem

Let me be specific about what is happening. A CEO reads about AI. The board asks questions. A "Head of AI" gets hired, usually someone from consulting. A pilot project launches. It touches maybe 40 people in a company of 4,000. Six months later, the CEO tells investors AI is "transforming operations." Meanwhile, 3,960 employees are doing exactly what they did before. Maybe with a ChatGPT login they use to rewrite emails.

That is not AI automation for business. That is theatre.

The surveys confirm this split. Executives say AI saves time. Employees say it doesn't. Both are telling the truth from where they sit. The CEO sees the pilot team's metrics. The employees see their actual workday, which has not changed. The disconnect is not about the technology. It is about deployment. Companies are buying AI. They are not building AI into how work actually gets done.

I have seen this pattern with our clients at HEIMLANDR. The companies that come to us after their "AI initiative" failed almost always made the same mistake. They treated AI as a product to purchase, not as an engineering problem to solve. They wanted a tool. What they needed was a rethinking of their process, followed by custom AI solutions built to fit that new process.

Sweden's collision course

Now layer Sweden's reality on top of this.

Sveriges Radio is reporting right now on a massive labour shortage in elderly care. The numbers are brutal. Sweden's population is aging fast. The ratio of working-age adults to retirees is shrinking every year. There are not enough people to staff elderly care facilities today, let alone in five years. This is not a forecast. It is happening now.

At the exact same moment, Sweden is weighing stricter family visa rules. The political logic, I assume, is about migration control. But the practical effect is this: you make it harder for people to bring families, you make it harder to attract and retain international talent. You accelerate the talent drain at exactly the moment you need more workers, not fewer.

So Sweden has two problems colliding. A shrinking workforce and an AI deployment failure rate of 80%. The one technology that could offset the demographic crunch is being wasted. And the policy environment is making the talent side worse, not better.

From San Francisco, this looks abstract. From Jönköping, it is concrete. I talk to Swedish companies that cannot hire. Not "struggle to hire." Cannot. The people do not exist. And when I ask what they are doing about it, too many of them point to an AI chatbot on their website and call it digital transformation.

What the Nordics get right. And wrong.

Sweden and the Nordics have real advantages here. High digital literacy. Strong infrastructure. A culture of trust that makes organizational change easier. A workforce that is, broadly, comfortable with technology. These are not small things. Most countries would kill for this starting position.

But we are squandering it. The EU's AI Act creates compliance overhead that disproportionately hits smaller European companies. A startup in Gothenburg faces regulatory requirements that a startup in Austin does not. That is a choice we made. Maybe it is the right choice long-term. But right now, in 2026, it is slowing deployment at the moment we need speed.

Sweden's public sector, which is enormous and which runs most of the elderly care system, moves slowly on technology procurement. By the time a municipality goes through its procurement process for an AI system, the technology has moved two generations forward. The system they buy is outdated before it is installed.

Meanwhile, China is deploying AI in elder care at scale. Not perfectly. Not without problems. But at scale. The US private sector is moving fast on AI agent development in healthcare, logistics, and customer service. Europe is writing frameworks. Sweden is writing reports about writing frameworks.

I say this as someone who loves this country and builds here deliberately. We have everything we need except urgency.

What actually works

At HEIMLANDR, we have been building AI agents and custom AI solutions for companies across Europe. The ones that see real productivity gains share three things.

First, they start with a specific bottleneck, not a vision. Not "we want to use AI." Instead: "We have 14 people spending 60% of their time on document classification and it is the reason our processing time is 11 days instead of 2." That is a solvable problem. AI is very good at solvable problems. It is terrible at vague ambitions.

Second, they build, they do not buy. Off-the-shelf AI tools are fine for individual productivity. For organizational productivity, you need systems built into your actual workflow. That means custom development. That means someone who understands both the technology and your operations. The 80% failure number is largely a story about companies buying generic tools and expecting specific results.

Third, they measure before and after. Sounds obvious. Almost nobody does it. If you cannot say "this process took X hours before and Y hours after," you do not know if AI is working. You are guessing. And executives who are guessing tend to guess optimistically, which is exactly the gap the surveys are showing.

Where this goes: 2027-2030

Here is my honest read on the next few years.

The hype cycle is going to correct. The 80% number will become common knowledge. Some companies will use it as an excuse to pull back on AI entirely. That will be a mistake. The 20% that are seeing gains are seeing serious gains. The gap between companies that deploy AI properly and those that don't is going to become a competitive moat within 18 months.

AI agent development is where the real shift happens. We are moving from "AI as a tool a person uses" to "AI as an agent that executes tasks independently." This is a fundamentally different thing. A tool augments a person. An agent replaces a task. When you have a workforce shortage, you don't need augmentation. You need task replacement. You need agents that can handle scheduling, document processing, initial patient intake, logistics routing, compliance checking. Things that currently require a human not because they require human judgment, but because nobody has built the system to do it otherwise.

On the path toward AGI, I think the interesting question is not "when" but "what happens to organizational structure when AI agents can handle 40% of current white-collar tasks." My answer: the companies that figure this out first win. Not because they save money, though they will. Because they can actually operate in a labour market where the workers do not exist.

For Sweden, the trajectory is clear. Either we get serious about deploying AI in public services, elder care, healthcare, municipal operations, or we face a service quality collapse driven by demographics. There is no third option. Immigration policy cannot fill the gap fast enough even if it were moving in the right direction, which it is not. Domestic birth rates are not going to suddenly reverse. The math is the math.

Regulators need to understand that the AI Act should enable deployment in critical public services, not just regulate risk. Right now it is mostly the latter. We need regulatory frameworks that make it easier for a Swedish municipality to deploy an AI agent for elder care scheduling. Not harder.

What to look at

If you are a CTO or technical founder thinking about this seriously, here are things worth your attention right now.

Langflow (148k+ stars on GitHub) is an open-source tool for building AI-powered agents and workflows. If you want to prototype an agent internally before committing to a full build, this is a solid starting point. Visual workflow builder. Good for getting buy-in from non-technical stakeholders who need to see what an agent actually does before they approve budget.

Claude Code (128k+ stars) is Anthropic's agentic coding tool. It lives in your terminal, understands your codebase, and handles routine development tasks through natural language. If you are running a small engineering team and need to move faster, this is real productivity gain, not theoretical. This is the 20% working.

AutoGPT (184k+ stars) remains relevant as a framework for building autonomous AI agents. The project has matured significantly. Worth revisiting if you looked at it a year ago and dismissed it as a toy.

And if you are thinking about what an AI development company in Europe can do for you specifically, we build this stuff. Not as a vendor selling you a dashboard. As engineers who sit with your team, find the bottleneck, and build the agent that solves it. That is what we do at HEIMLANDR.

The actual ask

If you are a CEO reading this: stop asking your team "are we using AI?" Start asking "which specific process has AI made measurably faster this quarter?" If the answer is vague, you are in the 80%. That is not a technology problem. It is a leadership problem. And it is fixable.

If you are in Sweden, add a second question: "What happens to this operation in three years when we cannot hire for 30% of these roles?" Because that is where we are heading. And the time to build the systems that absorb that impact is now, not when the crisis is acute.

I write this from Jönköping on a Saturday morning. It is quiet here. The lake is calm. Sweden is beautiful and comfortable and that comfort is, right now, the biggest risk we face. The world is not going to wait for us to finish our fika before it moves on.

Build the thing. Measure the thing. Stop pretending.

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.