Build vs Buy Is Dead. Swedish Companies Haven't Noticed Yet.

Right now, somewhere in Stockholm, a CTO is presenting a build vs buy software analysis to a board. The spreadsheet has three columns. Build cost. Buy cost. Total cost of ownership over five years. Every number in that spreadsheet is wrong. Not because the math is bad. Because the cost curves those numbers sit on will not exist in 18 months. That CTO is making a decision for 2026 using economics from 2023. And nobody in the room knows it.

I run HEIMLANDR.IO out of Jönköping. We build software. We build AI systems. We see what is happening on the ground, not in analyst reports. And what I see right now is an entire industry of Swedish technology leaders sleepwalking into architecture decisions that will age like milk in the sun.

The Framework Everyone Uses Is Based on a World That No Longer Exists

Build vs buy has always been a simplification. But it used to be a useful one. You had two real options: hire engineers and build something custom, or buy an off-the-shelf SaaS product and configure it. The tradeoffs were real. Custom SaaS development gave you control but cost you time and money. Buying gave you speed but locked you into someone else's opinions about your business.

That binary is gone.

Today, the real options look more like this: Build with AI agents that write and deploy code in hours. Buy a SaaS tool that a foundation model will replicate as a feature within a year. Assemble from open-source components that are themselves being rewritten by AI contributors. Or hire a software development company in Sweden that charges €100/hour to do what Claude and Codex do in minutes.

The "10 Top Software Development Companies in Sweden for 2026" lists are circulating right now, celebrating shops charging €80-120/hr whose core value proposition is translating business logic into code. That is literally the first thing agentic AI dismantles. Not in theory. Right now. Today. Look at what OpenHands is doing. Look at everything-claude-code, which just crossed 168,000 stars on GitHub. These are not research projects. These are production systems that turn natural language into deployed software.

The question is no longer build or buy. The question is: how fast can you rebuild when the tool you bought gets eaten by a foundation model, and the team you outsourced to becomes an AI agent workflow?

Sweden Is in a Strange Position Right Now

Two things are happening simultaneously in the Swedish software market this week, and the collision is telling.

On one side, McKinsey reports a Nordic software export surge driven by companies building proprietary platforms. Swedish tech is strong. Proprietary is winning. Good news.

On the other side, the consolidation machine is running at full speed. Valsoft acquires Tiltid, a Swedish workforce management company. Bridgepoint and Polaris buy a Swedish software developer from CapMan. Roll-ups everywhere. Private equity is gobbling up mid-tier SaaS companies because the playbook says: acquire, consolidate, squeeze margins, exit.

Here is what PE firms buying Swedish SaaS companies are not accounting for: entire SaaS categories are about to become features. Not in five years. Now. Dagens Industri is reporting on emerging Swedish AI companies that could render whole software verticals obsolete. When a foundation model can do what your acquired SaaS product does, but better and for free as a side effect of a larger platform, what exactly did you buy?

You bought a customer list and some technical debt. Congratulations.

What Jönköping Sees That San Francisco Doesn't

I live in a mid-sized Swedish city. We do not have the luxury of pretending that infinite VC money solves structural problems. When we at HEIMLANDR make technology decisions, they have to work. Not in a pitch deck. In production. For real companies that need to make money.

From here, the view is different than from San Francisco. In SF, the narrative is that AI will create a new generation of unicorns. Sure, maybe. From Jönköping, the reality is more immediate: Swedish manufacturing companies, logistics firms, and mid-market enterprises need to make software decisions this quarter. They are choosing between SaaS tools that might not exist in three years and custom builds from dev shops whose pricing model is about to collapse.

Sweden gets some things right. We have strong engineering culture. Good infrastructure. A regulatory environment that, while sometimes slow, at least tries to be coherent. But Swedish companies are deeply conservative about technology adoption. The decision-making process at a Gothenburg manufacturing firm involves committees that would make a Byzantine emperor blush. By the time the build-vs-buy decision gets approved, the options have already shifted underneath them.

Meanwhile, a 22-year-old in Lagos is shipping production software using AI agents and Daytona as infrastructure, spending close to nothing. That is your competition now. Not the dev shop in Malmö.

The Three Things That Actually Changed

1. The Cost of "Build" Just Fell Off a Cliff

MVP development used to mean three months and a team of five. Now it means a weekend and someone who knows how to direct an AI coding agent. I am not exaggerating. We do rapid MVP development at HEIMLANDR and the timelines have compressed by 70-80% in the last year alone. The cost of building custom software is in freefall, and it is not going to stop falling.

This means every "buy" decision needs to be re-evaluated. If building a custom version of that SaaS tool costs 20% of what it cost two years ago, the buy calculation changes completely. The integration headaches, the feature gaps, the vendor lock-in. All of that pain was worth it when building was expensive. When building is cheap, it is just pain.

2. The Shelf Life of "Buy" Is Shrinking to Months

SaaS products used to have a useful life of 5-10 years. You would implement Salesforce and live with it for a decade. Now, entire product categories emerge and get subsumed in 18 months. The workforce management space that Tiltid occupies? Foundation models with agentic capabilities are already handling scheduling, compliance tracking, and workforce optimization as part of broader enterprise AI deployments. Not as standalone products. As features.

When you buy a SaaS tool today, you are betting that its category will still be independent in three years. That is an increasingly bad bet.

3. The Team You Outsourced To Is Becoming an AI Workflow

This is the part nobody wants to talk about openly. The SaaS development company model, where you pay a team of humans to translate requirements into code, is being disrupted from the inside. Not eliminated. Not yet. But the ratio of humans to output is changing fast. A team of 10 becomes a team of 3 with AI agents. Then a team of 1 directing multiple agents. Then, for certain categories of work, no team at all.

At HEIMLANDR, we are building AI agent systems for clients that do exactly this. Not to replace engineers. To make each engineer 10x more effective. The shops that are not making this transition are the ones getting acquired in fire sales right now.

Where This Goes: 2027-2030

Let me be direct about what I think happens next.

By late 2027, the build-vs-buy question will be replaced by the "assemble-vs-generate" question. You will either assemble solutions from AI-native components that configure themselves to your business logic, or you will generate entire applications from specifications using foundation models. The idea of buying a static SaaS product and paying per-seat licensing will look as dated as buying shrink-wrapped software from a shelf.

On the path toward AGI, every SaaS product becomes a thin wrapper around capabilities that foundation models provide natively. CRM, ERP, project management, HR systems. All of them. The value moves from the software to the data and the business logic. The companies that own their data and understand their own processes will win. The companies that outsourced their thinking to a vendor will be stuck.

For Sweden specifically: EU regulation is not keeping up. The AI Act focuses on risk classification and compliance, which matters, but says almost nothing about the economic disruption of SaaS consolidation, the labor market implications of agentic AI replacing dev shops, or the competition policy questions raised when foundation model providers absorb entire software categories. Swedish regulators at Integritetsskyddsmyndigheten (IMY) and the like are focused on data protection, which is important, but they are fighting the last war. The next war is about market structure, and nobody in Brussels or Stockholm is ready for it.

PE firms doing SaaS roll-ups in the Nordics should be terrified, but they are not, because their models run on 3-5 year hold periods and the disruption curve is just outside their spreadsheet horizon. Just barely. That is the most dangerous position to be in.

What to Look At

If you are making software architecture decisions right now, here are the things I would pay attention to:

OpenHands (72K+ stars). AI-driven development that is moving fast from experiment to production reality. If you are evaluating the cost of "build," you need to understand what this does to your estimates.

Daytona (72K+ stars). Secure infrastructure for running AI-generated code. This is the missing piece that makes AI-generated software deployable and safe. Important for any team moving toward agent-driven development.

Agno (39K+ stars). Turns agents into production software. This is where the "assemble" part of the future lives. Not demos. Not prototypes. Production agent systems that ship to real users.

Appsmith (39K+ stars). Internal tool builder that represents the current state of low-code. Worth understanding as the baseline that AI-native tools are already surpassing.

What to Actually Do About This

Stop making 5-year architecture decisions. Make 18-month bets with clear exit paths. If you are buying a SaaS tool, make sure your contract allows you to leave and that your data is portable. If you are building, build modular. Build in layers where the AI-generated parts can be swapped and regenerated as models improve.

Audit your current stack for "category risk." Which of your SaaS tools exists in a category that a foundation model could absorb as a feature? That is your highest priority for replacement planning.

Invest in your own data infrastructure. The software is becoming commodity. Your data and your business logic are not. Make sure you own both, fully, with no vendor lock-in.

If you are working with a software development Sweden partner, ask them directly: what is your AI agent strategy? How has your per-engineer output changed in the last 12 months? If they cannot answer, they are the ones getting acquired next.

And build your team's ability to direct AI systems. The skill is shifting from "can write code" to "can specify, direct, and verify AI-generated systems." Every month you wait to develop this capability internally is a month your competitors are getting ahead.

The Punk Rock Part

Here is what bothers me. The entire Swedish tech ecosystem is celebrating consolidation and top-10 dev shop lists while the ground is moving beneath everyone's feet. We keep optimizing for a world that is already ending. Nicer offices. Better Scrum processes. More efficient offshore teams. All of it is deck chairs on the Titanic if you are not fundamentally rethinking what software is and how it gets made.

At HEIMLANDR, we are not waiting for permission. We are building AI-native solutions right now, for real clients, with real revenue impact. Not because we read a trend report. Because when you sit in Jönköping and look at what is actually happening in the codebases and the model capabilities and the cost curves, the only honest conclusion is: the old framework is dead. Build something new.

The question is not build or buy. The question is: are you fast enough to keep up with a world where both options are being rewritten every quarter? If you are not asking that question in your next board meeting, you are already 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.