Build vs Buy Is Dead. Swedish Companies Clinging to It Bleed Cash.

The framework is dead. The invoices are not.

You are sitting in a boardroom right now debating build vs buy software. Stop. The framework is dead. It died sometime in the last twelve months and nobody sent the memo to your procurement committee.

Here is what actually happened: AI coding tools have collapsed the cost of building custom software by 70-90%. A single senior developer with Claude Code or OpenHands ships what a five-person team shipped in 2024. The math that made buying SaaS rational, the math that said "it's cheaper to pay Salesforce than to build," no longer holds. Every SaaS subscription you carry is now technical debt you are paying someone else to own on your behalf. And you are paying them monthly, forever, for the privilege.

I run HEIMLANDR.IO from Jönköping, Sweden. We build software. We build AI agents. So yes, I have a bias here. But the numbers do not care about my bias, and neither should you.

The old math is gone. The new math is brutal.

Let me be concrete. In 2023, custom SaaS development for an internal tool, say a bespoke CRM layer on top of your ERP, cost you 800,000-1,500,000 SEK and took four to six months with a team of three to five developers. You could buy a SaaS seat license for 500 SEK per user per month and be live in two weeks. Buying was the rational choice. Obviously.

In 2026, that same internal tool costs 100,000-300,000 SEK to build and takes two to four weeks. One senior dev. AI-assisted from architecture to deployment. The SaaS seat license still costs 500 SEK per user per month, which means you break even in under a year, and after that you own the asset outright. No API rate limits. No forced migrations. No "we're sunsetting this feature." No vendor deciding your data model for you.

VentureBeat reports this week that AI has already cratered software build costs but enterprise governance has not caught up. That is the polite way of putting it. The blunt way: your buying decisions are running on 2023 assumptions in a 2026 cost environment, and every month you delay, the gap between what you pay and what you should pay gets wider.

Sweden's position: perfect setup, terrible execution

Here is what frustrates me from Jönköping. Sweden is perfectly positioned to exploit this shift. We have some of the highest developer quality in the world. Our senior developers charge 750-900 SEK per hour, which used to be expensive compared to offshore shops billing 300-400 SEK. But when one Swedish senior dev with AI tooling outputs what five offshore developers produce, the per-feature cost flips. Swedish software development is suddenly the best value proposition on the planet, measured by what actually gets shipped.

We have the infrastructure. We have the education base. We have a culture of building, from Spotify to Klarna to Minecraft. Stockholm is a startup factory. Even here in Småland, where I sit, there is serious engineering talent that the big cities ignore.

And yet. Swedish companies, especially the mid-market and enterprise layer, are paralyzed. They are hiding behind procurement committees. Behind "enterprise readiness" checklists that were designed for a world where buying was correct. Behind EU AI Act compliance discussions that, while important, have become a convenient excuse to do nothing.

I talk to CTOs at Swedish industrial companies. Companies with 500-5,000 employees. Companies that pay millions in SaaS licensing annually. They know the math has changed. They tell me privately over coffee. Then they go back to the office and renew the contracts because changing the procurement process is harder than explaining the overspend.

Compare this to what I see from US founders. GeekWire reports that engineers in Seattle are calling Claude Code "a new era of software development." They are not debating. They are shipping. YC-backed startups are launching products built by one or two people that would have required seed rounds and ten-person teams two years ago. The velocity gap between American builders and Swedish committees is growing every week.

And in Asia, the picture is different again. Chinese and Indian development shops are seeing their labor cost advantage evaporate. When AI is the multiplier, the quality of the human in the loop matters more, not less. Sweden should be winning this race. Instead, we are watching from the sidelines, filling out compliance forms.

The SaaS trap nobody wants to name

Let me say something that SaaS vendors will not like. The entire SaaS business model was built on an arbitrage: it was cheaper for a vendor to build once and sell many times than for each customer to build their own. That arbitrage depended on building being expensive. Building is no longer expensive.

What SaaS vendors still offer is maintenance, updates, and "not having to think about it." Fair enough. But what they also deliver is lock-in, data dependencies, rigid workflows that do not match how you actually operate, and an ever-growing line item on your P&L that your CFO frowns at but nobody kills.

CIO.com is running a piece right now arguing that the decision is no longer build vs buy but "how to combine the two." They are half right. The combination model works for commodity infrastructure. You should probably keep buying your email service and your cloud compute. But for anything that touches your core business logic, your customer experience, your operational workflows? Building is now cheaper, faster, and gives you ownership of the asset.

Every month you pay a SaaS vendor for something you could build, you are renting when you could own. And the rent goes up.

What the AI tooling stack actually looks like today

I am not speaking theoretically. Here is what we see and use at HEIMLANDR right now.

Claude Code has changed how we do MVP development. A senior developer with the right prompting framework, good architecture instincts, and Claude Code as the execution layer can produce production-quality code at a pace that would have seemed absurd in 2024. The everything-claude-code repo (180K+ stars on GitHub right now) shows how seriously the community is taking optimization of these workflows. Skills, memory, security patterns, research-first development. This is not toy stuff. This is production engineering.

OpenHands (73K+ stars) is pushing AI-driven development further toward full autonomy. We are not there yet. The "AI writes the whole app" dream is still mostly a dream for anything non-trivial. But AI writes 60-80% of the code, and a skilled human shapes, reviews, and architects the rest. That ratio is moving fast.

RTK is a Rust-based CLI proxy that reduces LLM token consumption by 60-90% on common dev commands. Single binary, zero dependencies. This is the kind of infrastructure tooling that makes AI-assisted development economically viable at scale, not just for experiments.

For API development and testing, the open-source ecosystem around Hoppscotch and Bruno means you do not need Postman licenses anymore. You build your API, you test your API, you own your toolchain. Streamlit still dominates for rapid data app prototyping. The whole stack is shifting toward open, composable, AI-augmented.

This is what "build" looks like in 2026. It does not look like 2019.

Where this goes: 2027-2030 and the AGI overhang

If the current trajectory holds, and I believe it will accelerate, here is what the next few years look like.

By late 2027, a non-technical founder with strong domain knowledge will be able to ship a production SaaS product with zero traditional coding skills. The AI tooling will handle code generation, testing, deployment, and basic maintenance. The human provides the business logic, the taste, the domain expertise. This is not speculation. The components exist today. They just need to get better at talking to each other.

By 2028-2029, the SaaS vendor market consolidates violently. Mid-tier SaaS companies that do not have deep data moats or network effects will see churn rates they cannot survive. When your customer can rebuild your product in a weekend, your only defense is the data you hold hostage and the switching cost you create. Expect vendors to get more aggressive about lock-in, not less, in the next two years. Plan accordingly.

The AGI question hangs over all of this. If we get something approaching general intelligence in the 2028-2030 window, which serious people at every major lab are working toward, then the concept of "buying software" may become as quaint as buying a pre-built spreadsheet template. You will describe what you need and an agent will build it, maintain it, and adapt it in real time. The companies that survive that transition are the ones that own their data, their processes, and their institutional knowledge in structured, portable formats. Not locked inside someone else's SaaS database.

Regulators are not ready. The EU AI Act is focused on risk classification and transparency requirements for AI systems. It says almost nothing about the economic disruption of AI-assisted development collapsing the software market. Swedish policymakers at Regeringskansliet are still talking about "digital transformation" as if it is 2021. The gap between policy and reality is large and growing. I do not expect this to change soon. Which means the opportunity is there for companies willing to move without waiting for permission.

What to look at

If you are a CTO or founder reading this, here are specific things to evaluate this week:

everything-claude-code. The most comprehensive optimization framework for Claude Code workflows. If your developers are using AI coding tools without structured skills and memory management, they are getting maybe 40% of the potential output. This repo closes that gap.

OpenHands. AI-driven development platform. Worth evaluating as a complement to your existing dev workflow, especially for internal tooling and automation tasks that you currently buy SaaS for.

RTK. If token costs are a concern (and they should be at scale), this Rust CLI proxy cuts consumption dramatically. The economics of AI-assisted development depend on tools like this.

Your own SaaS spend audit. Seriously. Pull your vendor list. For each line item over 10,000 SEK per month, ask: could we build this in two to four weeks with one senior dev and AI tooling? If the answer is yes, you have found your next project.

What to actually do

I will make this concrete because vision without action is just entertainment.

Step one: Pick one SaaS tool you are paying for that covers a non-commodity function. Not email. Not cloud hosting. Something that touches your business logic. Customer onboarding flows. Internal reporting. Workflow automation. Something you have complained about for at least a year.

Step two: Scope a rapid MVP to replace it. Two weeks. One senior dev. AI-assisted. The goal is not perfection. The goal is to see, with your own eyes, what the new cost curve looks like for your specific context.

Step three: Compare. Total cost of the MVP build plus estimated annual maintenance versus annual SaaS cost. Include the cost of not having control. The features you cannot get. The integrations that do not work. The data you cannot export.

Step four: Make the decision based on 2026 math, not 2023 math.

This is not about building everything. It is about recognizing that the line between "build" and "buy" has moved dramatically, and most organizations have not updated where they draw it.

From Jönköping, with conviction

I write this from a small city in southern Sweden where we build real things. At HEIMLANDR, we work with companies that are tired of paying rent on software they should own. We do custom SaaS development, we build AI agents, we ship MVPs fast. Not because building is always the answer. But because in 2026, it is the answer far more often than your procurement process admits.

The build vs buy framework is dead. What replaces it is a more honest question: do you want to own your future, or rent it? Swedish companies have every advantage needed to own it. The engineering talent, the infrastructure, the culture of building. What they lack is the willingness to act before the committee approves.

The committee will never approve fast enough. The math will not wait. Build.

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.