The Cataclysm of Enterprise SaaS

For most of the last twenty years, the buy-versus-build calculus had exactly one rational answer: buy. Building was slow, expensive, and never finished. A generic seat license was always cheaper than a team of engineers babysitting an internal tool. That asymmetry is the entire foundation of the enterprise SaaS industry — the bet that it will always be cheaper to rent software than to make your own.

That bet is now under real pressure. When the marginal cost of producing a working internal tool drops toward zero, the case for renting a generic one weakens with it. The thesis worth taking seriously is the sharp one: when companies can cheaply code exactly the internal tools they need, the market for selling collaboration software to them starts to collapse. As the argument was put on a recent episode of Naval’s podcast, The AI Industrial Revolution — the creative seed for this post — “internally you’re just coding the right things that you need at any given time.” Even spreadsheets aren’t safe: “I personally have moved almost entirely from Excel to Python.”

That last line sounds like a personal quirk. It’s actually the whole mechanism in miniature. A spreadsheet is bought leverage — someone else’s general-purpose tool that you bend toward your specific problem. Python plus an agent is made leverage: the exact tool for the exact job, generated on demand. When the cost of “made” approaches the cost of “bought,” a lot of bought software stops making sense.

The numbers are starting to move #

This isn’t only a builder’s intuition anymore; the early data points the same direction. Drawing on a late-2025 survey of 817 builders, 35% of enterprise teams have already replaced at least one SaaS tool with a custom build, and 78% plan to build more custom tools in 2026 (Newsweek). The supply-side cost has dropped to match: a module that took roughly six weeks in 2024 now takes about four, and work that once meant weeks and six figures increasingly takes days. Telling, too: 60% of builders shipped tools outside IT oversight in the past year — the shift is already outrunning official procurement (Newsweek).

The pricing model that funds the industry is the part most exposed. SaaS economics rest on per-seat licenses and the renewal compounding that follows. As companies build workflow-specific agents instead of buying seats, that compounding gets undercut — and customers walk into renewals holding a credible “we could just build this” lever. The market has started to price the risk, too. Through the first months of 2026, software valuations took a visible hit as investors began discounting that build-alternative threat, with horizontal point-solution vendors repriced the hardest (Forrester).

The analyst class has caught the scent. AlixPartners, in May 2025, declared that “perpetual-to-SaaS is the past; SaaS-to-AI is the future,” noting that nearly 90% of software executives are optimistic about AI even as the value migrates from SaaS vendors to whoever owns the AI layer (AlixPartners). Forrester, in February 2026, went further, pronouncing that SaaS “as we know it is dead” — with horizontal, point-solution vendors facing elimination while vertical, data-rich SaaS has better odds (Forrester).

One widely cited prediction belongs in the column with a caveat: Gartner is reported to have forecast that 40% of enterprise applications will feature task-specific AI agents by the end of 2026, up from less than 5% in 2025 (Gartner). I could not independently verify that page — it returned a 403 — so treat the exact figure as directional rather than confirmed. The trend it implies, though, is independently corroborated: Menlo Ventures put enterprise AI spend at $37 billion in 2025, up 3.2x from $11.5 billion in 2024, with startups capturing 63% of application-layer revenue, up from 36% (Menlo Ventures). Incumbents are already ceding share at the AI layer to newer entrants. That part is not in dispute.

The honest counter-case #

Now the other side, because it’s strong and the strong version of this thesis has to survive it.

The same data set argues the opposite. Menlo Ventures also found that 76% of AI use cases are now purchased rather than built — up from 53% in 2024 (Menlo Ventures). In the near term, the AI wave is expanding the market for AI-native SaaS, not collapsing SaaS wholesale. The threat is concentrated on incumbent point-solutions, not on the model of selling software itself.

The moats that matter aren’t features. Forrester’s own teardown concedes that enterprise software bundles compliance infrastructure, security governance, and audit trails that take years to build and sit deeply embedded in regulated industries. Healthcare and financial-services platforms are far less exposed than productivity and collaboration apps. Vertical SaaS is, if anything, gaining — a roughly $133.5 billion market in 2025 on its way toward ~$194 billion in 2029, defended by proprietary domain data that a custom build cannot cheaply reproduce.

Version one is the easy 10–20%. Retool’s own survey notes that building v1 of a tool is often as little as 10–20% of total lifecycle effort. The “hidden SaaS tax” of integrations and workarounds has a mirror image: the hidden cost of owning vibe-coded internal tools — the technical debt, the security exposure, the day-1,000 maintenance that a vendor contract otherwise absorbs.

Even the canonical example hedges. Klarna is the poster child here. Around September 2023 it dropped its Salesforce CRM and built an in-house system on OpenAI’s models, a Neo4j graph database, and internal data infrastructure. The downstream AI customer-service bot handled 2.3 million conversations within a month of launch and drove an estimated $40 million profit improvement in 2024, roughly 700 full-time agents’ worth of work (AlixPartners). Klarna didn’t trade one vendor for another — it coded exactly what it needed, and the Salesforce relationship wasn’t upgraded, it was cancelled. But Klarna’s own CEO, Sebastian Siemiatkowski, threw cold water on generalizing it: “Will all companies do what Klarna does? I doubt it,” predicting SaaS will consolidate into fewer AI-integrated platforms rather than vanish (TechCrunch). Klarna is an engineering-heavy fintech with the in-house capability to pull this off. It is not a 200-person logistics firm.

What “cataclysm” actually means #

Put the two sides together and the picture is less apocalypse, more re-sorting. “Even spreadsheets are cooked” is the right instinct aimed at the wrong target. What’s cooked is the generic, horizontal, seat-priced tool whose only job was to be slightly better than a spreadsheet — the layer where switching cost was always low and differentiation was always thin. What survives is software that owns something a customer can’t regenerate in an afternoon: proprietary data, regulatory surface, deep integration, the brutal day-1,000 maintenance burden.

For builders, the strategic read is concrete. If your product’s core value is a workflow a competent team can now vibe-code in a week, your moat is already gone — you just haven’t gotten the renewal call yet. If your value is in the compliance, the data network, or the long-tail maintenance nobody wants to own, the same AI wave that threatens the seat-license business is handing you cheaper leverage to widen the gap. The cataclysm is real. It’s just landing on a narrower band of the market than the headline suggests — and the question for every SaaS company is which band it’s standing in.