Updated July 3, 2026.

For twenty years, “build vs. buy” had a boring answer: buy. Why sink engineers into a CRM or a project tracker when someone else already sells a polished one for a few dollars a seat? That logic built the entire software-as-a-service industry. In 2026, it stopped being obvious.

Call it the SaaSpocalypse. In early 2026, roughly 300 billion of market value evaporated once investors did the math on what AI agents actually replace. The thing that made SaaS printing-press profitable, charging per human seat, only works when you need a lot of humans clicking around in the software. Agents don’t click. They just do the task.

What actually broke

The per-seat model assumed people. A team of thirty needed thirty licenses across a stack of tools, and every tool billed accordingly. Now a handful of agents can run the research, the drafting, and the coordination that used to keep those thirty people (and their thirty subscriptions) busy. Fewer humans in the loop means fewer seats to sell, and a pricing model that quietly assumed otherwise starts to crack.

At the same time, building your own tool got dramatically cheaper. Not long ago, “build” meant a real engineering team and a real budget. Now someone who can describe what they want in plain language can stand up a working internal tool with something like Claude Code in an afternoon. The wall between “we could buy this” and “we could just build this ourselves” got a lot shorter.

And once building is cheap, most SaaS features look awfully replicable. If your product’s moat was a tidy interface around a database, an agent can reproduce it faster than your sales team can renew the contract. The competitive ground shifts from having the feature to how well the whole thing integrates and holds up in real work.

The incumbents say: not so fast

Here’s where it gets interesting, because the people running the biggest SaaS businesses think the obituary is premature. Oracle NetSuite’s Evan Goldberg, who has been in this game since he founded NetSuite in 1998, called the SaaSpocalypse averted in a July 2026 interview, and his reasoning is hard to wave away.

His point: an agent can write the code, but it can’t write the knowledge. “The code’s not very hard to build,” Goldberg said. “The knowledge is a lot harder to build.” NetSuite’s real asset isn’t its features; it’s decades of accumulated understanding of how thousands of companies actually run their back offices, most of it living in the heads of an ecosystem, never written down where a model could scrape it. You can clone a dashboard in an afternoon. You can’t clone that.

He also flips the in-house-build romance on its head. Building everything yourself means becoming a technology provider to yourself, and, as he put it, that’s not why most businesses got into business. There’s a real cost to reinventing plumbing that a vendor has already pressure-tested across thousands of customers.

The catch in his own argument, and he says this plainly, is the word “if.” A SaaS company that fails to build AI into its product would “absolutely be in danger of being replaced.” So the incumbents aren’t safe because SaaS is safe. They’re safe if they embed agents fast enough to defend the knowledge moat. That’s a much narrower claim than “nothing changed.”

From software to systems

So who’s right? Both, mostly. The commodity middle of SaaS, the tidy interface wrapped around a database, is genuinely exposed. The defensible core, the hard-won knowledge and the deep integrations, is not. What’s really happening is a move from software-as-a-service toward systems-as-a-service: value delivered through AI-native agents and paid for by outcomes, not by how many people you sit in front of a dashboard.

That reframes build vs. buy entirely. You don’t buy software to avoid building; you buy the knowledge and integration you can’t reproduce, and you build the thin, specific layers where owning the outcome matters. This is exactly why we build the tools we run our own work on, and buy the ones where a vendor’s accumulated expertise beats anything we’d reinvent. The advantage isn’t owning software or renting it. It’s owning the judgment about which is which, and wiring them together so the whole thing keeps working when the landscape shifts again.

If your build-vs-buy spreadsheet still defaults to “buy” out of habit, it’s running on 2019 assumptions. But if it now defaults to “build it all ourselves,” that’s a 2026 overcorrection. The real question isn’t build or buy. It’s which outcomes you want to own outright, and which are backed by knowledge you’d be foolish to try to rebuild from scratch.