Build vs Buy Is Dead. Swedish CEOs Are Asking the Wrong Question.

The €150/hour invoice that no longer makes sense

If you are a Swedish mid-market CEO still signing off on outsourcing contracts with Eastern European dev shops at €150/hour, I need you to stop and read this. The entire cost logic behind that decision evaporated sometime in the last twelve months, and the Swedish business press is barely talking about it.

The old build vs buy software debate assumed two things: building is expensive and slow, buying is cheaper but rigid. That framing is dead. There is now a third option that sits between them, and it changes everything. I call it "generate." A senior engineer with Cursor, Claude, and a clear spec now ships what a five-person outsourced team used to deliver. Not in theory. In practice. I see it every week at HEIMLANDR in Jönköping.

But here is the part nobody wants to hear: the companies that rush to generate everything in-house, without governance, without architecture, without someone who actually understands what the AI is producing, are building a disaster that will make traditional technical debt look like a rounding error.

This is the real conversation for 2026. Not build vs buy. Build vs generate vs buy. And most Swedish CEOs are still stuck on question one.

What actually changed

Let me be specific about what happened. Spiceworks reported this month that the build vs buy calculus has fundamentally shifted because AI rewrites every cost assumption in custom SaaS development. That tracks with what I see on the ground.

A year ago, if a mid-market Swedish company needed a custom internal tool, say a logistics dashboard or a customer portal, they had three realistic options. Buy a SaaS product and adapt their workflow. Hire an agency or outsource to Poland or India. Or build an internal team, which takes six months to hire and twelve to deliver.

Today, a competent senior developer using AI-assisted coding tools can prototype that same tool in days. Not weeks. Days. The OpenHands project on GitHub, which just crossed 74,000 stars, is building AI-driven development as an open-source platform. Tools like Daytona are creating secure infrastructure specifically designed to run AI-generated code. The ecosystem is not theoretical anymore. It is production-ready and moving fast.

This is why the €150/hour outsourcing model is dying. The arbitrage was always about labor cost. A Polish or Ukrainian developer costs less than a Swedish one. Fine. But when AI compresses a five-person-week task into a one-person-day task, the labor cost becomes almost irrelevant. What matters is the quality of the person directing the AI and the governance around what gets shipped.

The governance gap nobody in Sweden is discussing

VentureBeat warned this month that enterprise governance has not caught up with how cheaply AI lets teams spin up internal software. Companies are building faster than they can secure or maintain what they build. This is not a future problem. It is happening right now.

I look at Swedish CIO coverage and I see plenty of articles about "building effective AI teams." Good. But there is a glaring absence in the Swedish conversation: nobody is seriously discussing what happens when every department head can spin up a custom tool using AI, with no architecture review, no security audit, no plan for maintenance.

Think about it. Your marketing team uses an AI tool to generate a customer data dashboard. Your logistics team builds a route optimization app. Your HR team creates an internal portal. All built fast. All built cheap. None of them talk to each other. None of them went through security review. Three of them store personally identifiable data in ways that violate GDPR. And nobody in IT even knows they exist.

This is shadow IT on steroids. The old shadow IT problem was people buying unauthorized SaaS subscriptions. The new one is people generating entire applications without oversight. The surface area for failure is orders of magnitude larger.

Sweden and the Nordics vs the rest of the world

Here is what is interesting about the Swedish position. We have real structural advantages that most Swedish firms are not using.

Sweden produces excellent senior engineers. Consistently. The education system, the tech culture in Stockholm, Gothenburg, Malmö, and yes, Jönköping, all of it produces people who can actually think architecturally. That matters enormously in a world where the bottleneck is no longer writing code but directing AI to write code well.

In San Francisco, the conversation has moved past build vs buy entirely. The top companies there treat "generate" as default and invest heavily in governance frameworks to manage it. They have internal platform teams whose entire job is to make AI-assisted development safe and consistent.

In the EU, the AI Act adds regulatory complexity that US companies do not face. Swedish firms need to think about compliance in ways that a startup in Austin does not. But the Swedish regulatory conversation is still stuck on "what is AI" rather than "how do we govern AI-generated software at scale."

Asian companies, particularly in China and South Korea, are moving aggressively on AI-generated software for internal use. They have less regulatory friction and more willingness to experiment. The speed gap is real.

Swedish mid-market firms sit in an awkward spot. Too sophisticated to ignore AI. Too cautious to adopt it without process. And too reliant on outsourcing relationships that no longer deliver value proportional to cost. The firms that figure this out in 2026 will have a structural advantage for the next decade. The ones that do not will be paying for it in maintenance costs, security incidents, and competitive lag.

The real framework: when to build, when to generate, when to buy

Let me be practical. This is how I think about it at HEIMLANDR when we advise clients on custom SaaS development and MVP development.

Buy when:

The problem is commoditized and the SaaS vendor's core business depends on solving it. Accounting. Email. CRM for standard sales processes. Project management. Do not generate your own Slack. Buy Slack. The vendor will always outinvest you on features for generic use cases.

Generate when:

You need internal tools, prototypes, or workflow-specific applications where the logic is clear but no off-the-shelf product fits. A senior engineer with AI tools can build these in days. But, and this is critical, you need governance. Code review. Security review. Architecture that fits your stack. Maintenance plan. Without these, you are generating liabilities, not tools.

Build when:

The software IS the competitive advantage. Your proprietary algorithms. Your customer-facing platform. The thing that differentiates you in the market. This still requires deep engineering, thoughtful architecture, and serious investment. AI accelerates the build, but it does not replace the thinking. This is where working with a focused AI solutions partner or a SaaS development company that understands both the technology and the governance matters.

The fatal mistake:

Treating "generate" as "build." They are not the same. Building implies ownership, maintenance, architecture, and intent. Generating without those things creates an asset that depreciates immediately and becomes a liability within months.

Where this goes: 2027 and beyond

Let me look forward. The trajectory here is clear and it accelerates.

By late 2027, I expect AI coding agents to handle 70-80% of routine software development tasks autonomously. Not with a human prompting line by line, but with a human providing a spec and reviewing output. The ECC project on GitHub, which is building agent harness performance optimization for Claude Code, Codex, and Cursor, is a signal of where this is going. Nearly 200,000 stars. The developer community is voting with attention.

This means the "generate" category gets bigger and more powerful. It also means the governance gap gets wider and more dangerous. Every quarter that passes without enterprise-grade frameworks for managing AI-generated code is a quarter where the problem compounds.

On the path toward AGI, which I think is closer than the median Swedish CIO assumes, the question shifts again. It stops being "build vs generate vs buy" and becomes "what do humans still need to decide?" The answer will be: architecture, ethics, business logic, and accountability. The code itself becomes nearly free. The judgment around it becomes priceless.

For Swedish regulators: you are not ready. The AI Act is a start, but it does not address the specific problem of ungoverned AI-generated internal software proliferating inside enterprises. Datainspektionen will eventually have to deal with hundreds of AI-generated tools that handle personal data without proper DPIAs. The smart move is to get ahead of this. I am not optimistic that will happen.

For Swedish CEOs: the outsourcing model that served you for fifteen years is losing its economic logic. Do not panic-fire your partners. Do start investing in internal capability to direct and govern AI-assisted development. The companies that get this right will move three to five times faster than competitors at lower cost with better results. The companies that get it wrong will drown in ungovernable generated code.

What to look at

Four things I recommend spending time on this week if you are a decision-maker in this space:

Daytona (72,000+ GitHub stars). Secure, elastic infrastructure for running AI-generated code. If you are going to let teams generate software, you need sandboxed environments to run it safely. Daytona is the best open-source option I have seen for this.

RTK (54,000+ stars). A Rust-based CLI proxy that reduces LLM token consumption by 60-90% on common dev commands. Single binary, zero dependencies. If your engineers are using AI tools daily, this cuts your API costs dramatically. Practical, immediate value.

OpenHands (74,000+ stars). Open-source AI-driven development platform. Worth understanding even if you do not adopt it, because it shows the direction AI-assisted development is heading. Autonomous agents that can plan, code, and iterate.

Your own governance framework. Not a tool, but if you do not have a written policy for how AI-generated code enters your production systems, start writing one today. Who reviews it. How it is tested. Where it can run. What data it can access. This is not optional anymore.

The bottom line from Jönköping

I build software for a living. I use AI tools every day. I am not a skeptic and I am not a hype merchant. What I see is a genuine paradigm shift in how software gets created, and a massive gap between the speed of creation and the maturity of governance.

Swedish companies have an opportunity here. We have the engineering talent. We have the cultural disposition toward quality and process. We have a regulatory environment that, for all its friction, forces thoughtfulness. What we lack is urgency. The Swedish mid-market conversation about software development is still framed around outsourcing costs and vendor selection. That conversation is already obsolete.

Build vs buy is dead. The question now is build vs generate vs buy, and the answer is almost always a combination of all three, managed by people who understand what each option actually requires. If you are still asking the old question, you are solving last year's problem with last year's tools.

Start asking the right question. Then build the governance to handle the answers.

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.