Your CRM Is Building Itself. Your Board Doesn't Know Yet.

Somewhere in your organization right now, an employee has given an AI agent access to your CRM, your ticketing system, or your finance stack, and it has already changed something. Not tomorrow. Not next quarter. Right now, while you're reading this. Boston Consulting Group just told CIOs their software is "building itself" and their control window is shrinking. They're not wrong. They're just late to tell you what's already happened inside your own building.

I run HEIMLANDR out of Jönköping. Small city, small team, big opinions. And the thing I keep seeing, the thing nobody in Stockholm boardrooms wants to say out loud, is that Swedish mid-market companies are sitting on the best agentic AI opportunity in Europe and treating it like a compliance problem instead of a competitive weapon.

The Gap Between What BCG Sees and What Sweden Is Doing

BCG's framing is right for their audience: enterprise CIOs at companies with 40,000 employees, twelve layers of approval, and SaaS vendors who are quietly turning their products into autonomous agents that rewrite workflows without asking permission. That's a real phenomenon. Salesforce, HubSpot, SAP, they're all racing to make their platforms agentic before someone else's agent eats their category.

Temporal's 2026 developer report backs this up with numbers that should worry anyone still debating pilot programs: a 70.8% jump in engineers running AI agents in production this year alone. Not experimenting. Production. That's not a trend anymore, that's the floor.

Here's what nobody's connecting: Swedish companies are structurally built for this moment, and almost nobody here has noticed. We don't have BCG's problem. Most Swedish mid-market orgs have three, maybe four layers between the CEO and the person doing the actual work. Flat structure. Low bureaucracy. Direct lines of communication. That's exactly the environment where agentic AI deploys fast and clean, because there's nobody standing in the hallway asking for a steering committee.

Instead, what I hear in meetings across Jönköping, Gothenburg, even Stockholm, is "is this GDPR compliant." Which is a fair question. It's also the wrong first question, and it's being asked six months after the agent already touched production data.

Spiris Just Showed Us What Sweden Can Actually Do

I want to give credit where it's due, because this matters. Spiris just opened its financial platform to AI agents through MCP, the Model Context Protocol. That's a real, live, shipped example of Swedish infrastructure moving at the speed this moment demands. Not a whitepaper. Not a "strategic AI roadmap" slide deck. An actual open door for agents to interact with a financial system.

That's the exception, not the rule. Look at what's dominating Swedish tech media coverage right now: basic tutorials on how to call the DeepSeek API. That's fine content for someone starting out, but it tells you where the median Swedish org actually sits. We're teaching "how do I build an agent" while BCG's global audience has moved on to "how do I govern the four hundred agents already running inside my company."

That gap is the story. Not AI hype. Not doom. A gap between infrastructure moving fast (Spiris) and organizational thinking moving slow (everyone reading a beginner's guide to an API in 2026).

Why Small and Flat Beats Big and Careful

Silicon Valley's agentic AI rollout is happening inside organizations wrapped in governance theater. Every agent deployment at a Fortune 500 company goes through security review, legal review, a pilot with a control group, a steering committee, and eighteen months later they have a case study. Meanwhile a 40-person manufacturing company in Småland could deploy the same agent stack in a week, because there are three people who need to say yes, and two of them are in the same office. This is Sweden's actual advantage and almost nobody is using it. We complain about lacking Silicon Valley's capital and scale. Fine, we don't have that. But we have something they've engineered themselves out of: speed of decision. A Swedish mid-market CEO can look at an AI agent deployment, understand the risk, and greenlight it before lunch. That's not recklessness, that's the absence of bureaucratic drag. Use it.

What I keep telling founders and CTOs who call us: the AI agent development cost question you're asking is the wrong first question too. You're comparing quotes for building a custom agent against doing nothing, when the real comparison is building a custom agent against a competitor down the road who already has one running your exact workflow, quietly, without a press release. Custom AI solutions built for your actual process beat generic SaaS agents bolted onto someone else's assumptions almost every time we've built one.

The Regulatory Reality: EU AI Act Meets Agents That Don't Wait

The EU AI Act is a serious piece of regulation and I respect the intent. But it was drafted for a world of discrete AI systems doing discrete tasks. It was not drafted for a world where an agent chains six tool calls together, touches your CRM, your invoicing, your customer data, and your Slack, all in one autonomous run, in under thirty seconds, without a human reading a single log line. Swedish and EU regulators are not stupid, but they are structurally behind the deployment curve, and that's not a knock on individuals, it's what happens when law-making speed meets exponential technology speed. The IMY (Sweden's data protection authority) is going to have an increasingly hard time defining "who made this decision" when the honest answer is "an agent chain, orchestrated by another agent, that nobody specifically approved this exact configuration of." If you're a Swedish CTO, don't wait for regulatory clarity that isn't coming fast enough. Build governance into your agent architecture now: logging, human checkpoints on anything touching financial or personal data, and a clear owner for every autonomous workflow. Not because a regulator is about to knock. Because when they do, in eighteen months, you want to already have the answer instead of building it retroactively under a fine.

Where This Actually Goes

Two to five years out, here's what I believe happens, and I'm willing to be wrong publicly about this because that's how you have a real opinion. Agentic SaaS becomes the default interface for enterprise software. You won't log into a CRM to update a deal stage, an agent will have already updated it based on an email thread, a call transcript, and a Slack message, and you'll just get notified. BCG's framing of "software building itself" is the correct one, and it accelerates, not slows. The companies that win in this window aren't the ones with the biggest AI budget. They're the ones with the fewest approval layers between "we should try this" and "it's running in production." That is Sweden's structural gift, if we choose to use it instead of legislating ourselves into caution before we've even tried. On the AGI question: I don't think we need AGI for this to matter. Agentic AI, the narrow, tool-using, multi-step kind that's already shipping, is enough to restructure entire job categories inside mid-market companies over the next three years. Customer service, first-line finance ops, basic sales qualification, internal IT support. Not eliminated. Restructured, with a human supervising three or four agents instead of doing the work directly. If AGI arrives on top of this infrastructure, it plugs into agent orchestration that's already normalized. The org chart shock happens before the AGI shock, and it's happening now.

What to Look at Right Now

If you're a CTO or technical founder trying to move from theory to deployment this quarter, here's where I'd actually point you, based on what we're using and testing at HEIMLANDR: n8n for workflow orchestration with native AI steps, self-hostable if data residency matters to you, which in Sweden it should (n8n-io/n8n on GitHub). Ollama if you want to run models like DeepSeek, Qwen, or GLM locally instead of shipping every prompt to a US cloud, which matters more than most Swedish companies currently act like it does (ollama/ollama on GitHub). Firecrawl if your agents need to actually read the web reliably instead of hallucinating what a page says, which is a bigger operational risk than most teams have budgeted for (firecrawl/firecrawl on GitHub). And keep an eye on agent harness frameworks like obra/superpowers, because the next competitive edge isn't which model you use, it's how disciplined your agent's operating methodology is. Model choice is commodity now. Harness quality isn't. If you want this built for your actual business instead of duct-taped together from tutorials, that's literally what we do, from rapid MVP builds to full custom AI solutions that plug into what you already run.

What You Should Do Monday Morning

Stop asking "should we use AI agents." Someone in your org already has, informally, without asking. Your real question is whether you find out from a security audit or from a competitive advantage you built on purpose. Audit what's already touching your systems. Pick one workflow, ideally something boring like invoice matching or lead qualification, and build a proper agent for it with logging and a human checkpoint, not a shadow one built in someone's browser extension. Talk to your legal and compliance people, but don't let "is this GDPR compliant" be the sentence that kills momentum for six months. Ask it in parallel with building, not before building. Sweden doesn't need to out-fund Silicon Valley on this one. We need to out-execute them, and our flat, fast, low-bureaucracy structure is the exact shape that agentic AI deployment rewards. Nobody's using that advantage yet. I'd rather be early and wrong about the timeline than right and six months behind.

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.