Build vs Buy Is Dead. Who Owns Your Judgment Now?

I read about Hypercore this week. They raised $13.5M to build an AI agent that makes credit decisions for private lenders. Not an agent that helps a loan officer. An agent that decides. Somewhere in the US right now, a lending company is about to outsource the actual judgment call, "should this person get money," to a vendor's model. That is not a build vs buy decision. That is a company handing over its spine and calling it efficiency.

This is the thing nobody wants to say out loud: build vs buy, the framework every CTO has used for twenty years, is dead. It died quietly, the way most important things die, while everyone was still arguing about it using the old rules.

The Old Question Was About Software. The New Question Is About Judgment.

The classic build vs buy calculation went like this: do we have the engineering time to build this ourselves, or do we buy a tool and integrate it? Either way, the tool did what you told it to do. You kept the decisions. The vendor sold you a hammer, you decided what to hit.

That framework assumed the product was software. It is not software anymore. When you "buy" an AI agent for underwriting, customer support, hiring, pricing, fraud detection, you are not buying a tool. You are buying a proxy for reasoning. The vendor's model decides what counts as risk, what counts as a good customer, what counts as normal behavior versus fraud. You didn't buy a hammer. You bought someone else's brain and rented it out to make your calls.

Adnan Masood's argument in "Moats in the Age of AI" gets at this directly: the advantage in this era does not sit in code anymore, it sits in who controls proprietary data and distribution. Code is commodity. Everyone can spin up an LLM wrapper this week (probably three teams in your building already have). What nobody can copy is your data, your specific customer relationships, and crucially, your specific way of thinking about your business. The moment you rent that thinking from a vendor, you've given away the only thing that made you different.

The Lock-In Nobody's Naming

The InformationWeek piece on Apple's AI rollout and on IBM and Red Hat's push to build "trust infrastructure" for AI agents is worth sitting with. Big companies are racing to build the rails that agentic systems run on before enterprises realize what they've signed up for. This is not contractual lock-in, the kind you can negotiate out of with a good lawyer and eighteen months notice. This is structural lock-in. Your workflows, your decision logic, your institutional memory of "how we actually do things here," all of it gets encoded into someone else's model, someone else's fine-tuning, someone else's agent harness. Try migrating that. There's no export button for judgment. I've watched companies in Jönköping and across Sweden sign up for AI platforms the same way they signed up for CRM tools a decade ago, thinking it's a swap-in, swap-out relationship. It isn't. Once your risk models, your customer scoring, your operational decisions run through a vendor's black box for two years, you don't own your own competence anymore. You own a subscription to someone else's.

Sweden's Structural Advantage, If Anyone Uses It

Here's where it gets interesting for us. A founder building in Sweden right now sits on something most Silicon Valley SaaS resellers do not have: strict data residency requirements, and full obligations under the EU AI Act. Everyone treats this as friction. I think it's a moat, if you're smart enough to build instead of rent. The EU AI Act forces you to know what's inside your systems. High-risk AI applications, the kind that make credit decisions or hiring decisions, need documentation, explainability, human oversight. Most American AI vendors selling "agentic judgment" as a product are not built for that level of scrutiny, because the US market rewarded speed over accountability for the last three years. That gap is closing, slowly, but Sweden and the EU already live inside the accountability requirement. If you build your own AI agents in-house, with your own data staying in the EU, with your own audit trail, you are already compliant with where the regulation is going, and you own the thing that makes you competitive. If you rent judgment from a US vendor whose model logic you can't inspect, you are one AI Act enforcement action away from a very bad quarter. SVT and DI have both covered how slow Swedish enterprise has been to actually operationalize the AI Act, mostly treating it as a compliance checkbox rather than a design principle. That's the mistake. It should be a design principle from day one. The irony is brutal: Swedish companies pay Jönköping-level or Stockholm-level salaries, real skilled engineering talent, then rent American AI infrastructure to make the decisions that talent should be making. You're paying twice. Once for the people who understand your business, and once for a vendor whose model doesn't.

What the US Gets Right, What It Gets Wrong

The US moves fast, I'll give them that. Hypercore going from idea to $13.5M raise for an AI credit-decision agent shows how much capital is chasing "judgment as a service." But speed without ownership is fragile. When the next regulatory wave hits (and it will, the EU AI Act is not staying in the EU, other jurisdictions are watching closely), companies that rented their decision-making will scramble. Companies that built it, even imperfectly, will already understand their own systems well enough to adapt. Asia, particularly companies in Japan and South Korea building sovereign AI stacks, gets this instinctively. They are not renting American models for core decisions. Europe talks a good regulatory game but too many companies still default to "just use OpenAI's API for everything," which is renting judgment with extra steps.

Where This Goes: AGI and the End of "Buy"

Push this forward two to five years. As we get closer to systems that can reason across domains, not just pattern-match within one, the gap between "software you buy" and "judgment you rent" disappears entirely. There won't be a meaningful distinction anymore between a SaaS tool and an autonomous decision-maker, because every SaaS tool will have agentic decision-making built in by default. At that point, the build vs buy question resolves itself. Companies that never built internal capability to understand, audit, and override AI judgment will be structurally dependent on vendors for their core competitive function. Not their software stack. Their thinking. That's a permanent tax on your business, paid to whoever owns the model you rented in 2024. The companies that survive that transition will be the ones who treated internal AI competence as an asset class starting now, in 2026, while it's still relatively cheap and while the regulatory environment (especially here in the EU) actually rewards transparency and ownership over black-box convenience. Regulators are not ready for this, to be honest. The EU AI Act is a good start but it was written for a world of discrete AI applications, not for a world where every software product has embedded agentic reasoning by default. Swedish regulators and Finansinspektionen in particular need to start asking financial and healthcare companies not just "do you use AI" but "who owns the judgment logic inside the AI you use." That question isn't in most compliance frameworks yet. It should be.

What to Actually Do About It

Stop asking "should we build or buy this feature." Start asking "does this decision need to belong to us." If the answer is yes, and for most core business decisions it is, you build. Not because building is romantic (it's not, it's expensive and slow and occasionally miserable), but because owning your judgment is now a survival trait, not a nice-to-have. This is exactly the work we do at HEIMLANDR.IO. When a company comes to us wanting AI agents built, the first conversation is never about the tech stack. It's about which decisions in their business actually need to stay theirs. Sometimes the answer is "rent the commodity stuff, build the core." That's fine. Most of the time, companies have no idea which is which because nobody's forced them to separate the two. That's the actual value of a good SaaS development company in 2026: not writing code faster than you could, but helping you draw the line between what you rent and what you own before you accidentally rent your entire competitive advantage. If you're testing this internally before committing, rapid MVP development is the right move. Build a thin version of the decision-making system you actually need, in-house, fast, before you sign a multi-year contract for someone else's judgment engine.

What to Look At

If you're serious about owning your AI judgment instead of renting it, a few tools worth your engineering team's time this month: Graphify turns your codebase, docs, SQL schemas, and configs into a queryable knowledge graph, locally, deterministically, no vector store required. That's exactly the kind of internal system-understanding tool that lets you audit your own decision logic instead of trusting a vendor's black box. OpenHands is worth watching if you want AI-driven development that stays inside your own infrastructure rather than a hosted black box. ECC, the agent harness performance system for Claude Code and similar tools, is a good example of building your own reasoning layer on top of foundation models rather than accepting whatever judgment logic ships by default. And if API testing and integration work is part of your stack, Bruno and Hoppscotch are open source alternatives to Postman that keep your testing infrastructure yours, not rented.

The Real Cost of Renting Your Brain

I'll say this plainly because someone should: if your company's core decisions, who gets credit, who gets hired, what price is fair, what counts as fraud, run through a vendor's model that you can't inspect, you don't have a technology dependency. You have an identity problem. You've outsourced the thing that made your company yours. Swedish companies have spent a decade being cautious, methodical, slow to adopt. That reputation is usually mocked from San Francisco. But methodical, done right, means you actually understand what you're building before you depend on it. That's not weakness. That's the exact instinct this moment rewards. Use it. If you want help figuring out where the line sits for your business, between what to rent and what to build, that's a conversation worth having before your next vendor renewal, not after.

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.