Build vs Buy Is Dead. The Real Question Is Build vs Wait.

The €2M Platform You're Speccing Right Now Will Be a Template in 18 Months

If you're a CTO sitting in a steering committee meeting in Stockholm or Gothenburg right now, arguing about whether to build or buy your next platform, I have bad news. You're asking the wrong question. The build vs buy software debate assumed a stable cost curve for custom development. That curve just broke.

GoodFirms reports this month that 91% of software companies now use AI to cut development costs. Not experiment with. Use. In production. To ship real products faster and cheaper. That number was maybe 40% eighteen months ago. The repricing is happening now, and most Swedish enterprise buyers haven't noticed because their procurement cycles still run on 2022 logic.

I run HEIMLANDR.IO out of Jönköping. We build software. We sell custom SaaS development and AI solutions. So you might think I have every incentive to tell you to keep building. I don't. I have every incentive to tell you the truth, because the companies that understand what's happening will be my best clients in two years. The ones that don't will have blown their budget on a platform that an AI agent can replicate over a weekend by 2028.

The Build vs Buy Framework Assumed Stable Inputs

Build vs buy made sense when the inputs were predictable. Developer hours cost X. Vendor licenses cost Y. You modeled the total cost of ownership over five years, picked the cheaper option adjusted for strategic fit, and moved on. Clean. Rational.

Here's what's different now. The cost of "build" is in freefall, but not uniformly. The commodity layer of software, the CRUD apps, the dashboards, the integration glue, the standard workflows, is getting cheaper by the month. AI code generation tools are genuinely good at this stuff. Not perfect. But good enough that the gap between a €150/hour Swedish developer and a €40/hour Ukrainian developer building a standard SaaS admin panel is narrowing fast. Both teams use Cursor, Copilot, or whatever the current favorite is. The AI does most of the predictable work. The human reviews, tweaks, and handles the edge cases.

So the cost advantage of building custom just evaporated for anything that looks like a standard pattern. And simultaneously, the buy side is exploding with options. No-code platforms are eating the bottom 40% of what dev shops used to sell. Vocal Media's coverage of the no-code movement documents this well. These aren't toy tools anymore. They handle real workflows, real integrations, real data.

This means the old question, "should we build or buy?", has a new default answer for most functionality: wait.

What "Wait" Actually Means

I don't mean sit on your hands. I mean stop commissioning 18-month bespoke platform builds for functionality that is rapidly commoditizing. Instead, do three things:

First, identify what is truly proprietary. Not "our process is unique" proprietary. Every company thinks their process is unique. I mean: what data, what domain logic, what competitive moat exists that cannot be replicated by someone with access to the same AI tools and public knowledge? That's a shorter list than most CTOs want to admit.

Second, build only the irreplaceable parts. The things so deeply embedded in your domain that no LLM can hallucinate them into existence. Your proprietary pricing model trained on fifteen years of transaction data. Your sensor fusion pipeline tuned to your specific hardware. Your regulatory compliance engine built on interpretations that only your legal team understands. That's custom development worth paying for.

Third, rent everything else. Use SaaS. Use no-code. Use AI-generated prototypes. Accept 80% solutions for non-core functionality. Stop gold-plating admin panels that three people will use.

The Swedish Situation: Expensive and Exposed

Let me be specific about software development in Sweden right now. Our rates run €120-180 per hour for senior work. Eastern European shops charge €35-60. That gap used to buy you quality, reliability, cultural alignment, time zone overlap. Real things. Worth paying for.

But here's the uncomfortable part. According to Belitsoft's latest SaaS outsourcing market review, those Eastern European shops are themselves being disrupted by AI-augmented delivery. They're using the same tools we are. The quality gap is narrowing because the AI doesn't care what time zone you're in or what language you speak at lunch.

Swedish dev firms that compete on execution quality alone are in trouble. The ones that compete on domain understanding, strategic thinking, and the ability to figure out what to build in the first place, those firms have a future. That's where we position HEIMLANDR. We're not selling hours. We're selling clarity about what's worth building and what isn't.

The broader Swedish tech ecosystem has a related problem. Enterprise procurement in Sweden is still incredibly slow. I see RFPs in 2026 that read like they were written in 2019. Eighteen-month delivery timelines. Waterfall milestones dressed up with agile vocabulary. Fixed-scope contracts for a world where the technology shifts every quarter. This is not a minor process issue. It's a strategic vulnerability. Swedish companies are locking in commitments to build things that will be commodity by the time they ship.

Meanwhile, Swedish regulators are focused on AI ethics frameworks and EU AI Act compliance. Fine. Important. But nobody at Regeringskansliet is asking: what happens to the Swedish software consulting industry, worth billions in annual revenue, when AI compresses 80% of commodity development work? That's a workforce question. That's an economic question. And nobody's planning for it.

What Is Actually Worth Building Custom in 2026

Let me give you a concrete framework. At HEIMLANDR, when a client comes to us for custom SaaS development, we run their requirements through what I call the Hallucination Test.

Can an LLM generate a plausible version of this feature with a well-written prompt? If yes, it's commodity. Don't build it custom. Buy it, rent it, or use AI to generate it.

Can an LLM generate something that looks right but is subtly wrong in ways that matter? That's the danger zone. This is where most enterprise software lives. It's also where the most money gets wasted, because teams build complex custom solutions for problems that are 90% solved by existing tools and 10% genuinely unique.

Is the feature impossible to specify without deep access to proprietary data, institutional knowledge, or physical systems? That's what you build. That's where human engineers paired with AI tools create actual value.

Examples of things worth building custom right now:

• Domain-specific AI agents trained on your proprietary data and processes. Not generic chatbots. Real agents that understand your business context.
• Integration layers between legacy systems that have decades of accumulated business logic nobody fully documented.
• Compliance and audit systems where the rules are ambiguous, jurisdiction-specific, and change frequently.
• Core transaction engines where correctness isn't a feature, it's the product.

Examples of things you should stop building custom:

• User management and authentication. Solved.
• Standard dashboards and reporting. Solved.
• Basic CRUD applications. Solved.
• Email/notification systems. Solved.
• Generic workflow automation. Solved.

I know this list makes some Swedish SaaS development companies uncomfortable. It should.

Where This Goes: 2027-2030

Let me project forward. AI capabilities are doubling roughly yearly. Not in some abstract benchmark sense. In practical, ship-real-code sense. Today's AI pair programmer handles maybe 60-70% of routine development tasks with acceptable quality. By 2028, that number is probably 85-90%.

This doesn't mean developers disappear. It means the ratio shifts. One senior engineer with AI tools does the work that used to require a team of five for commodity functionality. But for the hard stuff, the domain-specific, safety-critical, legally complex, deeply integrated stuff, you still need experienced humans. Maybe more of them, because the volume of "hard stuff" grows as the "easy stuff" gets automated away.

The path toward AGI changes this calculus further. If we get systems that can genuinely reason about novel domain problems, and we're closer than most enterprise buyers realize, then even the "deeply embedded domain logic" advantage has a shelf life. Not this year. Probably not next year. But within five years, the definition of "too complex for AI" will shrink dramatically.

What does that mean for the build vs wait decision? It means every custom build should have a shorter time horizon. Stop building five-year platforms. Build for two years. Assume you'll rebuild or replace major components as AI capabilities advance. Design for modularity, not permanence. This is where rapid MVP development makes more sense than monolithic platform builds.

On the regulatory side, the EU AI Act is focused on risk classification and transparency. It says almost nothing about the economic disruption of AI-driven development automation. Swedish policymakers need to catch up. The consulting and outsourcing industry employs tens of thousands of people in Sweden. What's the plan when half of that work is automated? I don't see one.

What to Look At

If you're a founder or CTO processing this, here are specific things worth your attention right now:

Plandex is an open-source AI coding agent that handles multi-file, multi-step tasks. It's not a toy. Watch how fast these tools improve. Use it on a real project to calibrate your intuition about what AI can and can't do today.

OpenDevin is an open-source AI software engineer project. Still rough around the edges. But the trajectory matters more than the current state. If you're planning an 18-month build, you need to understand where these tools will be at month 12.

Cursor and Windsurf are the AI-augmented IDEs your developers should already be using daily. If they're not, you're paying full price for half-speed work. This is table stakes in 2026.

The EU AI Act documentation, specifically the sections on general-purpose AI systems. Know what's coming for compliance requirements. If you're building AI-augmented development pipelines, these rules affect you directly.

The Uncomfortable Bottom Line

Most of the custom software being commissioned in Sweden right now is a waste of money. Not because it's bad software. Because it's software that will be available as a commodity product or generatable by AI before the custom version finishes UAT.

The build vs buy debate is dead because it assumed a world where building had a predictable cost and buying had a predictable ceiling. Neither is true anymore. Building gets cheaper every quarter. Buying gets more capable every quarter. Both curves are accelerating.

The real decision framework for 2026: Is this problem so uniquely ours, so embedded in proprietary data and domain knowledge, that no external tool or AI system can solve it well enough? If yes, build it fast, build it modular, and assume you'll rebuild parts of it within two years. If no, wait. Rent. Prototype with AI. Revisit in six months.

This is what I tell every client who walks into HEIMLANDR. Some of them don't like hearing it. The ones who do are the ones I want to work with.

The future belongs to companies that know the difference between their actual competitive moat and the commodity software surrounding it. Figure out which is which. Then build only the moat.

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.