Build vs Buy Is Dead. The Real Bet Is Who Owns the Moat.

I read the "Emergent Index 2026: The Age of Custom Software" report this morning with coffee that was already cold by the time I finished it. Good data, wrong question. Same with the pile of ERP cost breakdowns and dev-rate comparisons flooding my feed this week. They're all still asking "should I build or buy" like it's 2019 and the answer depends on whether you can find good developers in Kraków for less than Kista. That question is dead. Nobody killed it on purpose. AGI-assisted development just made it irrelevant while everyone was busy writing spreadsheets about hourly rates.

Here's the actual decision facing every founder and CTO in Sweden right now: build, rent, or own your moat. Three very different bets. Only one of them compounds.

The Cost Math Everyone Is Getting Wrong

Clockwise Software just published another ERP cost guide. Tech Funding News has a "best custom software companies" roundup. Somewhere there's a Canada dev-cost breakdown doing the rounds too. All of them compare custom development cost to SaaS subscription cost using 2022 assumptions about how long custom software takes to build. That assumption is broken. At HEIMLANDR we're shipping MVP-grade products in Jönköping with AI agent harnesses doing the heavy lifting on scaffolding, testing, and infrastructure config. What took a six-person team eight weeks in 2021 takes two engineers and an agent stack ten days now. That's not marketing copy, that's what I watch happen in our own repos every week.

When the marginal cost of building custom software collapses toward zero, the calculus flips. The real cost of buying SaaS isn't the subscription fee. It's the compounding tax of not owning your data model, your integration layer, your ability to move fast when the vendor decides to raise prices 40% at renewal, or worse, gets acquired and sunsetted. Ask anyone who built a business on top of a tool that got bought by Salesforce.

Custom SaaS Development Is Not What It Was

The phrase "custom SaaS development" used to mean "expensive, slow, risky." That's no longer accurate and the industry hasn't caught up. Agent harnesses like the one behind ECC are turning "build your own system" into a supervised, fast, testable process instead of a multi-quarter bet-the-company gamble. Tools like Graphify let you turn your existing codebase, docs, and SQL schemas into a queryable graph so an agent understands your system before it touches it. This is not hobbyist tooling anymore. This is infrastructure for serious teams.

The Nordic Blind Spot

Here's where I get frustrated as a Swede. We are historically the most cautious buyers in Europe. Swedish SMEs and especially public sector procurement love a "safe" SaaS vendor with a nice case study and a Gartner mention. Long procurement cycles, risk-averse purchasing committees, a cultural instinct toward consensus over conviction. That instinct served us well when software was hard to build and vendors had real moats. It is now actively costing Swedish companies their competitive position. Look at what's happening in the US. Y Combinator's latest batches are full of two-person teams building what used to require Series A funding, because the AI tooling layer does the work a ten-person engineering org used to do. Dagens Industri covers Swedish scaleups chasing funding rounds to hire more engineers, while American competitors are hiring fewer people and shipping faster with agent-driven pipelines. That gap widens every quarter we spend comparing hourly rates in Belgrade versus Bangalore. Meanwhile Asia, particularly teams in Singapore and Seoul, are treating AI-native build capability as sovereign infrastructure, not a vendor decision. Sweden's instinct is still "who do we buy this from." That's the wrong frame in a world where the build cost is falling faster than any procurement cycle can adjust to.

Where EU and Swedish Policy Falls Behind

The EU AI Act is focused on classification and risk tiers for AI systems, which matters, but it says almost nothing about the economic shift happening underneath it: the collapse of build costs is going to hollow out a huge chunk of the European SaaS vendor market that regulators assumed would stay stable and taxable and lobbyable for the next decade. Public procurement rules in Sweden still assume vendor lock-in is the safe choice and custom build is the risky one. That assumption needs to be rewritten, and nobody in Vinnova or Digg is rewriting it fast enough. Sweden has good digital infrastructure ambition on paper. The procurement culture underneath it is still optimized for a world where custom software took 18 months and cost eight million kronor. That world is gone.

Build vs Rent vs Own Your Moat

Let me define the actual three options, because "build vs buy" was always a false binary hiding a third path. Rent: You use SaaS. Fine for commodity functions, payroll, generic CRM, things that are not your differentiator. You're renting someone else's roadmap and someone else's pricing power over your business. Build: You build custom software to replace a rented tool. Better control, but if you build a commodity feature from scratch you've just spent engineering time recreating something that was never going to be your moat anyway. This is where a lot of "AGI made building free" enthusiasm goes wrong. Cheap building doesn't mean everything is worth building. Own your moat: You identify the 10% to 20% of your stack that is actually your differentiator, the thing competitors can't easily replicate, and you build and own that with total conviction, using AI-assisted development to move at a speed that would have needed a Series B round three years ago. Everything else, you rent, automate, or ignore.

The mistake I see constantly, including from smart Swedish founders, is treating "AGI-assisted development is fast now" as a reason to build everything. It's not. It's a reason to build the 20% that matters with total ownership, and stop wasting cycles on the 80% that was never going to differentiate you anyway. We built our own AI agent tooling internally at HEIMLANDR because agent orchestration is our moat. We don't build our own invoicing system. That would be stupid.

Where This Goes

Two to five years out, here's my honest read. The SaaS category as we know it, seat-based pricing for generic horizontal tools, is going to compress hard. Not disappear, compress. The tools that survive are the ones with genuine network effects or genuine proprietary data moats, not the ones that survive because building an alternative used to be expensive. As models move closer to real AGI-level general reasoning, the build cost for entire categories of internal tooling approaches the cost of a good prompt and a supervising engineer. That changes procurement completely. It changes how VCs think about SaaS multiples, because a company whose moat is "we built decent workflow software" has no moat left when any team can build decent workflow software in a weekend. What survives: proprietary data, genuine brand trust, regulatory position, and infrastructure that requires deep domain expertise to operate safely, think healthcare compliance, financial infrastructure, anything touching blockchain settlement or smart contract logic where a mistake is expensive and irreversible. Those are moats. A nice-looking dashboard is not a moat in 2028. Regulators are not ready for this shift. The EU is regulating AI risk classification while an entire tier of the SaaS economy that funds European tech jobs quietly gets undercut by cheaper, faster, in-house alternatives. Someone in Brussels needs to start asking what happens to competition policy when the barrier to building a competitor to any given SaaS company drops to a few thousand euros of compute and two weeks of engineering time.

What to Look At

If you're deciding what to build this quarter, here's where I'd actually spend time: ECC, the agent harness system with skills, instincts, and memory built for Claude Code, Codex, Cursor and friends. This is the difference between "AI wrote some code" and "AI operated inside a disciplined development process." Worth understanding before you decide your team needs ten more hires to build something. Graphify, for turning your existing legacy mess of docs, SQL, and configs into something an agent can actually reason about. If your build-vs-rent decision keeps stalling because "our systems are too tangled to touch," this is the tool that removes that excuse. OpenHands, for teams wanting a serious look at autonomous development workflows without vendor lock-in to a single AI provider. rtk, if your AI dev costs are climbing because nobody's watching token consumption. A single Rust binary cutting 60-90% of token waste on common dev commands is the kind of unglamorous tool that actually moves your unit economics. None of these replace judgment. They replace the excuse that building is too slow or too expensive to consider.

What to Actually Do This Week

Stop comparing hourly rates between Warsaw and Bangalore. That variable barely matters anymore. Instead, sit down with your leadership team and list every piece of software you're renting. For each one, ask: is this commodity, or is this our moat. If it's commodity, keep renting, don't waste engineering time on it. If it's your moat and you're renting it, you have a strategic exposure problem, not a cost problem, and that changes what you should be doing about it this quarter. If you're not sure whether something's a moat or a commodity, that's usually a sign you haven't been forced to think hard enough about your actual differentiation. We help teams work through exactly this at HEIMLANDR, through AI solutions consulting and SaaS development engagements where the first conversation is never about tech stack, it's about what actually deserves to be owned.

The Jönköping View

We're not San Francisco. We don't have the venture capital density or the "burn ten million to find out" culture. What we do have, if we choose to use it, is a builder mentality that predates all of this AI hype, a country that made Spotify and Klarna and Skype when building things properly still meant grinding it out for years. That instinct is an advantage now, not a liability, because the tools finally moved fast enough to match the ambition. The question is whether Swedish founders and Swedish procurement departments realize the game changed, or whether we keep optimizing spreadsheets for a market that stopped existing sometime around when GPT-5-class agents started shipping real production code unsupervised. Sweden built its reputation on SVT-covered tech success stories that took patience and craft. The craft still matters. The patience requirement just collapsed. Use that.

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.