AGI Is 3 Years Away Again. Here's What Your Board Does This Quarter

I read three things this week that should have been front page news in every Swedish business outlet. They weren't covered at all. Not in DI, not in Breakit, not on SVT. Google's Sundar Pichai admitted, in public, that Search itself is "too opinionated" for an AI-first world. Sequoia published a framework splitting every company on earth into two categories: the ones rebuilt around AI and the ones bolting AI onto workflows built in 2019. And DeepMind, according to reporting doing the rounds this week, is running an internal three-year countdown to AGI.

Three genuinely enormous stories. Zero Swedish coverage. That silence is the real story.

The AGI Countdown Is Not the Point

Let me say the unpopular thing first: I don't care if AGI arrives in three years, seven years, or never in the form people imagine. I've watched this industry cry wolf on timelines since 2017. Every major lab has an internal countdown. Every countdown resets. That's not cynicism, that's pattern recognition.

What matters is not the date. What matters is that the three biggest players in AI, Google DeepMind, OpenAI's own ecosystem, and the capital allocators like Sequoia, are all repositioning right now as if the countdown is real. Sequoia isn't writing blog posts about timelines. They're changing how they evaluate portfolio companies. That's money moving. Money moving because of a belief is more important than whether the belief turns out true. Read Sequoia's public commentary and Pichai's comments on Search's limitations, and you see two of the most powerful organizations in tech quietly admitting the current paradigm of "add an AI feature" doesn't survive what's coming. That's not speculation. That's them repricing their own businesses.

Why Your Org Is Still Deploying AI Like It's 2023

Here's the uncomfortable diagnosis I give founders and CTOs who bring me in to look at their stack: most Swedish companies, even ones proud of their "AI initiative," are still running a 2023 playbook. A chatbot on the website. A copilot plugin in the CRM. A summary tool bolted onto support tickets. That was fine in 2023. It is not an AI strategy in September 2026. The Sequoia framing is blunt about this: there are companies that restructured their operating model around AI agents doing real work, and companies still treating AI as a feature you add to old software. The second group is going to get eaten, not by AGI, but by the first group, this year, with today's models. This is why AI agent development is the actual battleground right now, not some future AGI moment. Agents that execute multi-step work, hold state, call tools, and operate without a human clicking "approve" at every step. Look at what's happening in open source: openclaw just crossed a staggering star count building exactly this, agents that take real action across any OS. n8n has become the default automation layer for companies wiring AI into actual business processes instead of demo decks. This is not hypothetical tooling. This is what companies restructuring around AI are actually using, today, in production.

The Cost Question Everyone Asks Wrong

Every board conversation I sit in starts with the wrong question: "what's the AI agent development cost." Wrong framing. The right question is what does it cost you not to have this capability while your competitor does. A well-scoped agent deployment, built properly with the right architecture, costs a fraction of one senior hire's annual salary and works continuously. Companies still asking "should we do AI" in Q4 2026 are asking a question that should have been closed in Q1. The real cost isn't the build. It's the six months you spend in committee deciding whether to build it while a competitor in Berlin or Austin already shipped three iterations.

Sweden vs the Rest of the World: An Honest Comparison

I love this country. I built HEIMLANDR here in Jönköping because I believe in Swedish engineering discipline, our trust culture, our flat hierarchies that let good ideas move fast internally. But on this specific question, we are asleep. In the US, this AGI/agent conversation is dinner table talk in every tech-adjacent boardroom. Sequoia, a16z, and every major fund are publicly restructuring their thesis around it. In China, the state has explicit AGI-adjacent industrial policy, funding compute and model development as a strategic asset, not a side project. Even the UK, post-Brexit and scrambling for relevance, has a formal AI Safety Institute publishing regularly on frontier model risk and capability jumps. Sweden? I searched. Genuinely searched. Nothing from DI, nothing from Breakit, nothing from SVT on the AGI countdown story, the Sequoia framework, or the Millennium Project's business-implications research that's circulating in policy circles right now. Not a single Swedish outlet is asking whether our regulatory framework, our labor model, or our SME funding structure is ready for a world where AI agents do the work of junior analysts, developers, and case handlers within 24 to 36 months. This isn't a criticism of Swedish engineers. Our engineers are excellent, arguably some of the best per capita in Europe. This is a criticism of Swedish boardrooms and Swedish media. We are extremely good at building the thing once someone tells us what the thing is. We are not good, right now, at asking the strategic question before it becomes obvious to everyone. The EU AI Act, meanwhile, is still primarily oriented around risk classification for existing model categories, not the agentic, semi-autonomous systems that are already deploying in production today via tools like AutoGPT descendants and enterprise agent frameworks. Brussels is regulating yesterday's AI while today's AI is already three generations past the framework.

What Jönköping Gets Right That San Francisco Doesn't

To be fair to us: Swedish companies that do move on this move with better unit economics and less hype-driven waste than their Silicon Valley counterparts. We don't burn venture money on agent projects with no revenue model attached. When we build, we build for margin. That discipline will matter a lot when the AGI hype cycle corrects, and it will correct, because it always does. The winners in three years won't be whoever announced AGI first. It'll be whoever built profitable, durable AI-native operations while everyone else argued about timelines.

Where This Actually Goes

Here's my honest forecast, and I'll own it publicly so you can hold me to it. Within 18 months, "AI agent" stops being a differentiator and becomes table stakes, the same way "has a website" stopped being impressive around 2005. Companies without functioning agent infrastructure handling real operational load will look the way companies without a website looked twenty years ago: not behind the curve, simply not credible. Within 24 to 36 months, whether or not we hit whatever the labs define as AGI, we will hit something functionally close enough to matter: models that can reliably execute complex, multi-day, multi-tool workflows with minimal supervision. That's not science fiction. Look at the trajectory from obra/superpowers and affaan-m/ECC, agent harness and skills frameworks that are rapidly closing the reliability gap that's kept agents as demos instead of production systems. That gap is closing fast, and it's closing in public, on GitHub, visible to anyone paying attention. Regulatory bodies, in Sweden and the EU broadly, are not ready for this. The AI Act wasn't built with continuously-acting autonomous agents in mind. Liability frameworks for agent-caused errors barely exist. Labor law has no answer for a world where an agent does the case-handling work of a mid-level government employee. Someone needs to be having this conversation in Rosenbad, not just Mountain View. Right now, nobody is. For builders, this means the next three years are the actual window. Not the AGI window, the market-positioning window. The companies that establish real agent infrastructure, real data moats, real operational muscle around AI now will be the ones still standing when the hype settles and the tooling commoditizes.

What To Actually Look At This Quarter

Skip the philosophy debates. Here's where I'd point a CTO or technical founder this week:

• openclaw: the most active project right now for cross-platform agents that actually execute tasks rather than just chat. If you want to understand what "agent that does things" means in practice, start here.
• n8n: if your org needs to wire AI into existing operational workflows without a six-month engineering project, this is the fastest legitimate path. We use variants of this pattern in custom AI solutions work constantly.
• ollama: if data sovereignty matters to you, and in Sweden it should, running models like DeepSeek, Qwen, or GLM locally instead of shipping everything to a US API is not paranoia, it's basic GDPR hygiene.
• superpowers: a genuinely useful methodology for structuring how your team actually builds with agentic tools, rather than improvising every project from scratch.

What Your Board Does This Quarter

Not a five-year AI strategy deck. Not another vendor evaluation cycle that drags into Q1. Three concrete moves: First, get one real agent into production somewhere in your operations this quarter. Not a pilot that lives in a slide deck. Something touching real revenue or real cost, even something small. If you don't have the internal bandwidth, this is exactly what Rapid MVP work is for, get a working version live in weeks, not quarters. Second, ask your legal and compliance function directly: are we ready for an agent-caused error, and who is liable? If nobody in the room has an answer, that's your actual risk exposure, not the AGI timeline. Third, stop asking "when is AGI coming" in board meetings. Start asking "which of our competitors already restructured around AI agents, and what do they know that we don't." That is the Sequoia question. It is the only question that matters at your scale, today.

If you're building this properly, from agent architecture to the underlying platform, that's the work we do daily at HEIMLANDR, whether it's fullstack development for the systems these agents plug into or full AI agent builds from scratch. As an AI development company in Europe, we don't sell AGI hype. We sell working systems that make money this quarter.

I don't know when AGI arrives. Nobody does, including the people running the countdowns. What I know is that the companies winning in 2029 are being built right now, quietly, by people who stopped debating the timeline and started shipping the infrastructure. Sweden has the engineering talent to lead this instead of watching it happen elsewhere. We just need boardrooms willing to ask the question out loud.

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.