Nobody Ships Repos Anymore, They Ship Judgment

Here's the thing nobody in Jönköping wants to hear at the fika table: if your engineering org still measures itself by headcount, you are optimizing for a job that is dying in real time. Not in five years. Now. This week. In the commit logs, not the press releases.

I run HEIMLANDR.IO out of Jönköping. We build with AI, blockchain, and SaaS. And the single biggest shift I've watched in the last eighteen months isn't a model release or a benchmark score. It's that writing code stopped being the scarce skill. Judgment became the scarce skill. Those are not the same thing, and most Swedish tech leadership still runs the org chart like they are.

The 95/5 Split Nobody Budgeted For

Let's be precise about what's actually happening, because "AI writes code now" is old news and it undersells the shift. What's happening in September 2026 is that agentic tools like Cursor, Claude Code, and the open-source wave around them (look at opencode, now sitting past 200,000 stars as "the open source coding agent") don't just autocomplete functions. They plan, execute, test, and iterate across entire repos with minimal supervision. Give them a well-scoped ticket and they will produce a PR that would have taken a mid-level engineer two days, in twenty minutes. That's the 95%. Boilerplate, CRUD, test scaffolding, refactors, migrations, glue code, the stuff that used to justify a headcount plan and a hiring pipeline. It's becoming free. Not cheap. Free, in the sense that the marginal cost is approaching the cost of a token. What's left is the 5%: what should we build, what tradeoff are we actually making, is this architecture going to survive contact with real users, is this the right problem to solve at all, and who is accountable when the agent is confidently wrong. That 5% was always the actual job. We just paid people to also do the 95% because there was no other way to get access to their judgment. Now there is. And that changes what "developer" means as a role, what a team should look like, and what you should be paying for.

Why This Breaks Swedish Org Design Specifically

Sweden built its tech culture on a manufacturing mindset. Predictable process, flat hierarchy, consensus, headcount as the unit of capacity planning. It worked beautifully for Ericsson-era hardware and it worked fine for the SaaS wave of the 2010s. But it assumes the constraint is labor hours. That assumption just broke. I see it constantly with founders and CTOs across Jönköping, Stockholm, Gothenburg. They still write job specs for "senior full-stack developer, 5 years experience, React/Node" as if the scarce resource is someone who can type TypeScript fast. That person's typing speed is now worth close to zero. What's worth something is: can this person look at an agent's output and know it's about to create a security hole, a data model that won't scale, or a UX decision that will tank retention. That's a completely different hiring filter, and almost nobody in Swedish tech has rewritten their filter yet.

Sweden vs. The Rest of the World: We're Behind, Not Ahead

Sweden likes to think of itself as a digitalization leader. Spotify, Klarna, decades of high broadband penetration, a government that put "digital first" in policy documents early. Fine. But this current shift isn't about digital infrastructure. It's about organizational velocity, and that's exactly where the Swedish consensus model slows down. In San Francisco, teams are restructuring around agent output right now, not next quarter. Y Combinator batches this year are routinely one or two founders shipping products that would have needed a ten-person team three years ago. In Sweden, I still see boards asking "how many developers do we need to hit the roadmap" as the primary planning question, instead of "what judgment do we need in the room to direct fifty agents doing the work of fifty developers." Swedish business press (Dagens Industri, Breakit) covers AI adoption mostly as a productivity story. Fewer meetings, faster tickets. That's not wrong, it's just shallow. The real story is a redistribution of value away from execution and toward decision-making, and almost none of our public discourse is treating it that way. Compare that to how aggressively companies in the US and increasingly in parts of Asia are collapsing team sizes and rebuilding comp structures around "how many agents can one senior person supervise responsibly." Sweden's strength, consensus and stability, becomes a liability when the market rewards speed of restructuring. The EU AI Act adds another layer. It's a real and necessary framework for high-risk AI use, but it says almost nothing about the organizational question I'm raising here: who is accountable when 95% of a codebase was generated by an agent and something breaks in production. Right now that accountability gap gets quietly absorbed by whoever's name is on the PR, usually a junior or mid-level engineer who had no real authorship over the judgment calls. That's a governance failure waiting to become a headline, and Swedish regulators have not touched it.

Where This Goes: 2 to 5 Years Out

Here's my honest read on the trajectory, not the hype version. Within 18 months, "developer" as a job title mostly disappears from job postings at companies doing this well, replaced by something closer to "product engineer" or "systems architect", roles defined by decision authority, not code volume. The people who thrive are the ones who were always good at the 5%: understanding users, understanding tradeoffs, understanding when to say no to a feature. The people who struggle are the ones whose entire value proposition was "I can implement things fast," because that's exactly what got commoditized first. Within 3 years, the org chart flattens hard. A team of three senior people directing a fleet of agents will out-execute a team of thirty mid-level developers, not because the agents are smarter than the developers, but because coordination overhead between thirty humans is a tax that doesn't exist between one human and fifty agent instances. Companies that don't restructure around this will lose on cost, speed, and eventually talent, because the best judgment-holders will go where their judgment is actually the product being sold. On the AGI question specifically: I don't think we need anything close to AGI for this shift to fully play out. Current agentic tooling, the kind powering LangChain-style orchestration and the swarm of Claude Code / Cursor competitors, is already sufficient to eat most of the execution layer of software engineering. What AGI actually changes is the 5%. If a system can genuinely reason about tradeoffs, product strategy, and long-horizon consequences the way a strong human architect does, then the remaining scarce human skill compresses again, down to something even narrower: taste, accountability, and the willingness to be the one who's wrong when it matters. That's a much smaller pool of people than "developers." Plan your career and your org accordingly. Regulation will not catch up in time. The EU AI Act was written for a world where AI is a tool humans use. We're moving into a world where AI is a coworker humans direct, at scale, without a job title for it. Sweden has a chance to lead here by being the first country to write sane rules about agent accountability in software delivery, the same way we led on data protection culture before GDPR forced everyone else to catch up. I haven't seen anyone in Swedish policy circles treating this with urgency yet.

What To Actually Do About It

Enough vision. Here's the practical version for anyone running an org or a team right now. Stop hiring for typing speed. If your interview process still centers on live-coding a data structure problem, you're screening for the 95% that's disappearing. Screen for judgment: give candidates an agent-generated PR and ask them to find the three things wrong with it, or the one architectural decision they'd overturn. That tells you far more than a whiteboard exercise. Restructure comp around outcomes and decisions, not hours or lines of code. If someone's job is now "direct five agents and be accountable for what they ship," pay them like an architect, not like a typist, even if their title still says "developer" for now. Audit your accountability chain today. Ask honestly: if an agent-written piece of code causes an incident in production, who actually made the judgment call that allowed it to ship, and does that person have the authority and the pay grade that matches that responsibility. If the answer is "the intern who approved the PR because nobody senior had time," you have a governance problem, not a technical one. If you're building fast and lean, this is exactly the environment for it. We run Rapid MVP engagements at HEIMLANDR specifically built around this new reality, small human judgment layer, heavy agent execution layer, shipped in weeks not quarters. Our AI Agents work is the same philosophy applied to internal operations, not just codebases. And if you need actual senior architecture thinking wrapped around agent-driven build, that's what Fullstack Development means to us now, not "developers who type," but people who direct systems and own the outcome.

What To Look At This Week

A few things worth your actual attention, not just bookmarking: opencode is worth running yourself if you want to feel what "95% free" actually means in practice, it's open source and moving fast. ECC, the agent harness performance system for Claude Code, Codex, Opencode and Cursor, is a good signal of where tooling is heading: not "can the agent write code" anymore, that's solved, but "how do we make agent judgment reliable, secure, and repeatable." That's the real frontier now. n8n if you haven't looked at it recently. Workflow automation with native AI is quietly becoming the layer where non-engineers direct agents to do engineering-adjacent work, which is its own threat to traditional dev headcount that most orgs haven't clocked yet. And read the actual EU AI Act text if you haven't, not a summary. Know exactly where the accountability gaps are in your own stack before a regulator or a lawsuit finds them for you.

The Uncomfortable Bottom Line

I say this as someone building a company on this exact shift, not watching it from the sidelines: if your value as an engineer, a founder, or a CTO is still "I can build it," you are standing on ground that's disappearing under you. If your value is "I know what should be built, why, and what happens when it's wrong," you just became more valuable than you've ever been. Software development in Sweden, in Jönköping, in Stockholm, doesn't need more people who can implement. It needs more people who can decide. Those are not interchangeable, and the sooner Swedish tech leadership admits that out loud, in hiring, in comp, in org design, the sooner we stop losing our best judgment-holders to companies abroad who already rewired for this.

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.