Build vs Buy Is Dead. Now It's Build vs Rent vs Prompt

I read the Emergent Index 2026 report this week and had to stop halfway through. Not because it was wrong, but because it was answering a question nobody should still be asking. It compares custom software costs across regions like the biggest variable left in software is where you source your headcount. Clockwise Software published fresh ERP cost breakdowns the same week, still priced in hours and FTEs. Meanwhile every healthcare app and restaurant app cost guide floating around right now quotes 2023 numbers for 2026 problems. The entire industry is fighting the last war.

Here in Jönköping, I watch founders debate Ukrainian dev shops versus Indian agencies versus Swedish consultancies charging 1,400 SEK an hour like it's still the deciding factor. It isn't. That conversation assumes the bottleneck is labor cost and labor location. It's not anymore. The bottleneck is judgment: knowing what deserves to be built by hand at all.

Custom SaaS Development Just Got a Third Option

For twenty years the framing was simple. Build vs buy software. Either you hire developers and own your stack, or you license a vendor's SaaS and live inside their roadmap. That binary is dead. There's a third lane now and it's not "outsource cheaper," it's "prompt it into existence." An AI agent can scaffold your ERP logic, generate your CRUD layer, wire up your auth, and stand up a working MVP in an afternoon. Not a mockup. Working code, tested, deployable. I've watched our own team at HEIMLANDR do this for client discovery sprints. What used to be a two-week estimate from a dev shop is now a same-day proof of concept, and the client decides by dinner whether it's worth building for real.

So the decision tree isn't build vs buy anymore. It's:

• Build: hand-crafted, defensible, your moat. Worth the six-figure invoice because it's the reason customers choose you over the alternative.
• Rent: vertical SaaS, commodity infrastructure, things every competitor also has. Stripe, Auth0, generic CRM logic. Paying a dev shop to rebuild this is not ambition, it's insecurity.
• Prompt: scaffolded by an AI agent, validated fast, thrown away if it doesn't work, hardened if it does. This is where most MVP development should live in 2026.

The founders who win this decade won't be the ones who found the cheapest hourly rate. They'll be the ones who can look at their own product and say, with precision, "this 20% is our moat, everything else is a commodity I'd be an idiot to keep paying for." That precision is the actual skill now. Everything else is procurement.

Sweden Is Having the Wrong Argument

I'll say the uncomfortable thing: Swedish tech media and Swedish policy have not caught up to this at all. Breakit covers funding rounds. DI covers layoffs and AI anxiety in general terms. Nobody in the Swedish press is asking the actual 2026 question, which is what happens to the entire regional labor arbitrage argument when an agent can do in an afternoon what an offshore team used to bill forty hours for. Sweden's whole self-image around software has been built on "we're expensive but we're reliable and well-engineered, unlike the cheap offshore alternative." That story worked when the alternative was a junior dev in another timezone with worse code review discipline. It doesn't work when the alternative is a well-orchestrated agent stack that does the boilerplate faster and more consistently than either option, and does it for the cost of API tokens. Compare that to what's happening in the US right now. San Francisco founders aren't debating hourly rates by region anymore, they're debating which agent harness to standardize on. Look at what's trending on GitHub this month: ECC has 242,000 stars because it's solving agent performance and memory at the harness level, not the "should I hire a dev shop" level. That's the level the conversation has moved to. Sweden is still debating hourly rates while the US is debating orchestration architecture. EU regulation is even further behind. The AI Act was built around risk classification for AI systems making decisions about people. It has almost nothing to say about AI systems writing the software that runs your business. There's no framework for who's liable when an agent-scaffolded ERP module has a bug that costs a customer money. There's no framework for whether "AI-assisted" custom software needs different audit trails than human-written code. Sweden, and the EU broadly, is regulating the wrong layer of this stack entirely, and I don't see that changing in the next two years unless someone forces the issue with an actual incident.

What Nordic Founders Get Right

Credit where it's due. The Nordic instinct toward lean teams and skepticism of bloat is actually the correct instinct for this moment, it's just aimed at the wrong target. Swedish founders are naturally suspicious of overbuilt software and expensive vendor lock-in. That instinct, pointed at the build-vs-rent-vs-prompt decision instead of the offshore-vs-local labor decision, is exactly the right muscle. We just need to redirect it.

Where This Actually Goes

Here's the trajectory I'm watching, and I think it plays out over the next two to five years, not ten. As models move toward something closer to general reasoning capability, the "prompt" lane stops being for scaffolding and starts being for maintenance too. Right now an agent can generate your MVP but a human still has to own the production system, debug the edge cases, and make architectural calls. That gap closes. Not to zero, but close enough that the economics of keeping a dev team on retainer for commodity features stops making sense entirely, even for mid-size companies. What survives is the moat layer. The 20% that's genuinely differentiated: your pricing algorithm, your fraud model, your specific integration logic that nobody else has because nobody else has your data or your customer relationships. That's where hand-building stays valuable, possibly more valuable, because it's the only place left where craft actually matters and where the market will pay for it. The middle tier, the vertical SaaS companies charging enterprise prices for what is essentially a templated CRUD app with a nice UI, gets crushed. Not immediately. But a founder who can prompt their way to 80% of a competitor's feature set in a week is not going to keep paying that competitor's per-seat pricing. This is exactly the dynamic behind those inflated healthcare app and restaurant app cost guides everyone's citing right now. Those prices reflect a market that hasn't priced in what's already possible. That gap between real capability and real pricing is where the disruption comes from, and it's already open. For regulators, the honest answer is they're not ready and won't be in time. The EU moves in multi-year cycles. This shift is happening in quarters. My genuine expectation is that by 2028 we'll have a real liability crisis around agent-generated code in production financial or healthcare systems, and only then will Brussels move. Builders should not wait for that framework. Build your own audit discipline now, because nobody's going to hand you one.

What to Actually Look At

If you're a CTO trying to make this decision this quarter, don't start with a vendor call. Start with these: Graphify turns your existing codebase, including your docs, SQL schemas, and configs, into a queryable knowledge graph. Before you decide what to rebuild or replace, know what you actually have. Most build-vs-buy decisions are made on outdated mental models of the current stack. Fix that first. OpenHands is worth running an actual pilot with if you haven't. It's an honest test of how far AI-driven development has come, run it against a real internal ticket, not a toy problem, and see what comes back. rtk cuts LLM token consumption 60 to 90 percent on common dev commands. If you're scaling agent-assisted development across a team, cost efficiency at the token layer matters more than people think. This is infrastructure hygiene, not a nice-to-have. And if you want a structured way to evaluate what's actually differentiated in your stack versus commodity, that's a conversation, not a tool. It's the actual work of leadership right now. This is exactly what we do with clients at HEIMLANDR through our Rapid MVP process: we build the prompted version first, fast, and use it to draw the real line between moat and commodity before anyone commits six figures to the wrong 80%. If the moat is real, that's where our Fullstack Development team goes deep. If it's not, we tell you to stop paying for it.

What To Actually Do This Week

Stop comparing hourly rates across regions. That fight is over and nobody won it, the ground moved. Instead, take your current stack and force a hard sort: what is genuinely differentiated versus what is commodity dressed up as custom. Anything commodity, evaluate whether an agent can scaffold it faster than your current vendor can ship it. Anything genuinely differentiated, that's where you spend your real engineering budget and your real hours, and that's where you should be talking to a team that can build AI agents or ship custom SaaS with judgment about which parts deserve craft. If you're still budgeting for software development Sweden-style, per hour, per head, per region, you're optimizing a variable that's about to stop mattering. The variable that matters now is discernment.

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.