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

Your 18-month custom build is already obsolete. You just don't know it yet.

I sit in meetings with Swedish CEOs who are still debating build vs buy software like it is 2019. They compare hourly rates. Jönköping vs Kraków vs Bangalore. They run spreadsheets on total cost of ownership over five years. They negotiate MSAs with dev shops billing $150-200/hr. And they are solving the wrong problem entirely.

The build vs buy framework is dead. Not dying. Dead. The real question in 2026 is: should I build this, or should I wait six months and get something better for a tenth of the cost?

Right now, McKinsey is spotlighting the Nordic software export surge. Swedish dev shops are busier and more expensive than ever. CapMan just sold a Swedish software developer to Bridgepoint and Polaris, which tells you PE firms see the build-side market peaking. London Post's 2026 ranking of top Swedish dev companies reads like a fine dining menu where every entrée costs the same and the kitchen is about to be automated.

And meanwhile, Sweden is launching free AI-ready robot programming courses for anyone with a laptop. The government itself is telegraphing what it expects to happen to labor costs.

If you are commissioning a large custom SaaS development project right now without accounting for the AGI trajectory, you are burning money.

The build side is broken

Let me be specific about what I mean.

The classic "build" argument went like this: we have unique business logic, no off-the-shelf product fits, we need custom software, we will own the IP, it will be our competitive advantage. Fine. That logic held when software was durable. When the thing you shipped in month 18 would still be state-of-the-art in month 36.

That is no longer the case. Custom-built software is now a depreciating asset that starts losing value the moment it ships. Not because the code rots. Because the capabilities available to everyone else accelerate faster than your internal team can iterate.

Look at what is trending on GitHub right now. OpenHands has over 72,000 stars. It is an AI-driven development platform that can generate, debug, and ship working code. everything-claude-code just crossed 173,000 stars. One hundred seventy-three thousand. It is a performance optimization system for AI coding agents. rtk, a single Rust binary, reduces LLM token consumption by 60-90% on common dev commands. These are not research projects. These are production tools being adopted right now.

The velocity of AI-assisted development means that whatever your team builds in 12 months, a two-person team with the right agent stack will replicate in 12 weeks. Possibly 12 days, depending on complexity. Your 18-month custom build is not a moat. It is a sand castle at high tide.

The buy side is collapsing too

Here is where it gets uncomfortable. The "buy" option is not safe either.

SaaS vendors themselves do not know if their product will exist by Q4. I talk to SaaS founders every week. The honest ones admit they are rebuilding their entire product around AI capabilities, and they do not know if the thing they ship next quarter will be the thing the market wants or if some new model release will make their approach irrelevant overnight.

Enterprise SaaS pricing is still based on seat licenses and annual contracts. But when an AI agent can do the work of five seats, the pricing model collapses. When your customer's intern can build a Streamlit app that does 80% of what your SaaS does, your value proposition collapses.

So "buy" is not safe. "Build" is not safe. Both sides of the old framework are unstable. That is why I say the framework itself is dead.

Sweden vs the world: we are getting this exactly wrong

From where I sit in Jönköping, the Swedish tech scene is doing something paradoxical. We are simultaneously producing world-class developers and clinging to development practices that the AI wave is about to flatten.

The Nordic software export surge is real. Swedish engineering talent is excellent. Our dev shops are competitive globally. But we are competing on craft at the exact moment when craft is being commoditized. It is like being the best blacksmith in 1910. Genuinely skilled. Genuinely irrelevant within a decade.

The US is moving faster on AI adoption. Not because American developers are better. They are not. But because the US venture ecosystem rewards speed and disposability. Build fast, ship fast, kill it if it does not work, build the next thing. Swedish companies still think in terms of building permanent infrastructure. Robust systems. (Sorry. Solid systems.) Five-year roadmaps. That mindset is now a liability.

Asia is moving even faster in some ways. Chinese AI development tools are iterating at a pace that European regulators cannot even track, let alone regulate. The EU AI Act is already struggling to keep up with models that shipped six months after the regulation was drafted.

Sweden's policy response? Free programming courses. Which is fine. Genuinely useful for individuals. But it does not address the structural question: what happens to the Swedish software development industry when AI agents can write, test, and deploy code at a fraction of the cost of a human team?

Swedish regulators are not prepared for this. The EU is not prepared for this. Nobody is, really. But the gap between the speed of AI capability growth and the speed of policy response is widening every quarter, and in Sweden we are adding bureaucracy while the US is stripping it away.

So what is the actual framework?

Here is what I tell the founders and CTOs who come to us at HEIMLANDR for SaaS development:

Stop thinking about build vs buy. Start thinking in terms of commitment horizons.

Zero to three months: Build only what you need to validate. This is MVP development in its purest form. Get something in front of users. Learn. Do not over-engineer. Do not write code that is "future-proof" because there is no such thing anymore.

Three to twelve months: Use AI agents aggressively. Build thin integration layers, not monoliths. Keep your architecture modular so you can swap components when better options appear. And they will appear. Every quarter.

Twelve months and beyond: This is where you stop and ask the hard question. Will this capability be commoditized before we finish building it? If the answer is "probably yes" or even "maybe," do not build it. Wait. Invest that capital in something AI cannot replicate: your data, your customer relationships, your domain expertise, your network effects.

The uncomfortable truth is that most custom software being commissioned right now falls into that last category. Most of it will be commoditized within 18 months. Most of it should not be built at all.

What should you actually build?

Not nothing. That is not my argument. The question is what.

Build things that compound. Data pipelines that make your proprietary data more valuable over time. AI agent systems that learn from your specific domain and customer base. Tokenized assets and smart contract infrastructure where your business logic becomes a protocol, not just an application.

Build things that are hard to replicate even with AI. If an AI agent can rebuild it from a prompt, it is not a competitive advantage. If it requires your unique data, your unique customer context, your unique regulatory position, then it has staying power.

Build thin. Build modular. Build with the assumption that every component will be replaced within 18 months. Make replacement cheap and painless. This is not technical debt tolerance. It is architectural realism.

Where this goes: 2027-2030

Let me paint the picture I see coming.

By late 2027, AI coding agents will be able to ship production-grade applications from natural language specifications. Not prototypes. Not demos. Production systems with proper error handling, security, testing, and deployment. We are already partway there with tools like OpenHands and the Claude Code ecosystem. The trajectory is clear.

By 2028, the concept of a "SaaS development company" in the traditional sense will be almost unrecognizable. Companies like ours at HEIMLANDR are already evolving from "we write code for you" to "we architect systems and deploy AI agents that write code." The value shifts from execution to judgment. From typing to thinking.

By 2029-2030, if AGI timelines hold (and I think the 2028-2030 window for something AGI-adjacent is plausible), the entire software industry restructures. Custom software becomes as cheap as electricity. The value is in the business logic, the data, and the human relationships, not the code itself.

Swedish companies that are commissioning 18-month custom builds right now will be maintaining legacy systems while their competitors spun up equivalent capabilities in a weekend. That is not hyperbole. That is the trajectory.

Regulators will be at least two years behind. EU policy will still be debating the AI Act implementation details while the market has already moved three generations ahead. Swedish companies that wait for regulatory clarity before acting will lose. You have to move now and adjust later.

What to look at

If you are a CTO or founder trying to navigate this, here is what I would put on your radar this week:

OpenHands (72,600+ stars). AI-driven development that is actually usable today. If you have not had your engineering team evaluate this for internal tooling, you are behind.

rtk (41,500+ stars). A single Rust binary that cuts LLM token costs by 60-90% on dev commands. This is the kind of infrastructure tooling that makes AI-assisted development economically viable at scale. Zero dependencies. Just works.

Daytona (72,300+ stars). Secure and elastic infrastructure specifically designed for running AI-generated code. When your agents are writing and deploying code, you need sandboxed, auditable execution environments. Daytona is solving that problem.

Breakit. If you are in the Swedish tech ecosystem, read Breakit daily. Their coverage of the AI transition in Nordic tech is the best source for understanding what is actually happening on the ground here, not what McKinsey thinks is happening.

The bottom line

Every week someone asks me: "Fredrik, should we build or buy?" And every week I tell them the same thing. You are asking the wrong question.

Ask instead: what is the half-life of this investment? If you build it today, how long before it is either commoditized or obsolete? If you buy it today, how confident are you that the vendor will still exist and still matter in 18 months?

Then act accordingly. Build small, build fast, build disposable. Own your data and your domain logic. Let AI agents handle the commodity code. And keep six months of your development budget in reserve for the thing that does not exist yet but will exist by November.

We are living through the most disruptive period in the history of software development. From Jönköping, from the middle of Sweden, I can see it clearly. The old frameworks are dead. The new ones are not fully formed yet. The only losing move is to pretend the game has not changed.

Build accordingly.

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.