AI Has Quietly Rewritten the Build-vs-Buy Rulebook. Most Boards Are Still Using the Old One.

For thirty years, "build or buy" had a settled answer. Building custom software was slow, expensive and risky, so the sane default was to buy a licence (Salesforce for the pipeline, Workday for the people, SAP for the rest) and adapt your process to the platform rather than the other way round. Whole procurement functions were built around that logic. It was rarely questioned because it was rarely wrong.

It is now wrong often enough to matter, and the reason isn't cheaper software. It's that AI has collapsed the cost of building something bespoke to the point where "custom" is no longer the expensive option it used to be.

What's actually changed, and what hasn't

The claim is not that SaaS is dying, or that every company should now build everything in-house. Most of what gets called the "AI kills SaaS" story overreaches, and a sceptical CFO is right to distrust it. The precise claim is narrower: the threshold at which a bespoke build beats a licensed platform has moved, function by function, and it has moved because AI has made two specific things cheap: gathering and structuring information, and producing a working first version of a tool. What it has not made cheap is judgement: the ability to know which version is actually right, defensible, and safe to bet the business on.

Sequoia's Julien Bek set out the underlying distinction in a widely discussed essay this year, "Services: The New Software," framing the opportunity as splitting any given piece of work into an "intelligence" component (data-gathering, benchmarking, first drafts) and a "judgement" component, meaning the actual call. His illustration is blunt: accounting software like QuickBooks costs on the order of $10,000 a year, but the accountant who uses it to close the books properly costs closer to $120,000. AI is rapidly commoditising the software layer of that comparison. It hasn't touched the second number, because the second number was never really about typing entries into a ledger.

The company that tried to prove the opposite, and what actually happened

The most useful evidence here isn't a study. It's a company that ran the experiment for real. In August 2024, Klarna's CEO told investors on an earnings call that the fintech had just shut down Salesforce and would shut down Workday within weeks, attributing it to "a combination of AI, standardisation and simplification". It was reported, understandably, as a company proving that platforms were obsolete.

What actually happened is more instructive. By the end of that year, reporting confirmed Klarna hadn't built Salesforce-equivalent or Workday-equivalent software from scratch at all. It had dropped those two platforms for a different SaaS provider (Deel, for HR), kept using Salesforce-owned Slack throughout, and layered AI on top of a still-largely-bought tech stack, while its AI customer-service assistant handled around two-thirds of support chats — genuinely well documented, though by May 2025 Klarna was rehiring human agents after its CEO conceded that cost had been "a too predominant evaluation factor" and quality had suffered. Salesforce's Marc Benioff pushed back publicly at the time, arguing there was "a broad misunderstanding of artificial intelligence and how it relates to data management and applications," and pressing Klarna to say what it was actually using to manage its employee, financial and customer information. He had a commercial interest in saying so, but on the specific point about governance, he wasn't wrong, and Klarna's own outcome quietly proved it.

That is the pattern worth learning from, not the headline. Klarna successfully replaced the cheap-to-replicate layer, routine customer interactions, with AI it built itself. It did not successfully replace the layers that carried real regulatory and governance weight. It bought those, just from someone else.

Why some things resist being built, even now

Bek's distinction extends naturally in another direction. Some functions won't be brought in-house no matter how good AI coding tools get, not because the technology can't do it, but because the client doesn't actually want to have done it themselves. Financial audit is the clean example: independence is the entire point, and the regulation requires an outside party regardless of how good your own numbers are. Management consulting sits in a softer version of the same category. A recommendation a board already wanted to make lands differently, and carries different accountability, when an independent expert has stress-tested it and put their name behind it. No amount of internal AI tooling replicates that, because the value was never purely in the analysis. It was in the analysis coming from somewhere else.

The practical test for a leadership team

This gives boards a genuinely useful filter, sharper than the old "can we afford to build it" question. Ask two things about any capability currently sitting on a SaaS invoice or a consulting statement of work. First: is the output well-defined and repeatable, the kind of thing where "good" has a checkable answer? If so, the make-versus-buy decision is now close to a coin toss on cost alone: build it cheaply, buy cheap tooling, or keep the existing licence; it barely matters, because AI has commoditised all three routes to roughly the same place. Second: does the value depend on independent judgement, external accountability, or a regulatory requirement that someone other than you signed off? If so, stop shopping for a platform or a DIY shortcut. That is exactly the category AI has not touched, and won't, because it isn't a knowledge problem, it's an accountability one.

This is the filter B2E applies to its own model, for the same reason. There is no value in selling a client a platform they could now assemble themselves in a fortnight, and no value in fielding a large standing team to do work that's become commoditised. What's left, and what's scarce, is flexible access to genuinely deep expertise, drawn from a wide bench rather than one partnership's payroll, deployed precisely where independent judgement is the actual product, and priced for exactly that long.

The build-versus-buy question hasn't disappeared. It's just being asked about a much smaller, much more expensive category of things than it used to be. The uncomfortable question for any leadership team is which side of that new line their current spending actually sits on, and whether anyone has checked.