Three years ago, “build vs. buy” was a settled question in enterprise software. You bought. The cost of building was too high, the maintenance burden too real, and the vendor ecosystem too mature. That calculus just flipped — and this week, every layer of the stack showed the consequences.

35% Have Already Replaced a SaaS Tool

A VentureBeat survey published this week puts a hard number on something that’s been anecdotal until now: 35% of enterprise teams have already replaced at least one SaaS product with a custom internal build powered by AI tooling. Another 78% plan to build more custom tools in 2026.

That’s not a pilot program. That’s a migration. The production cost of software has dropped far enough that internal teams are doing the math and concluding they can build cheaper than they can buy — at least for point solutions. The governance gap VentureBeat highlights is real (most of these builds lack the security review and compliance scaffolding that commercial SaaS provided), but the economic logic is already winning the argument inside organizations.

This is the demand-side mechanism the hypothesis predicts: when AI makes software cheap to produce, buyers stop buying. The question has always been where on the complexity spectrum the replacement stops. A custom dashboard? Sure. A custom ERP? Probably not. But the 78% planning more suggests the ambition line is moving upward.

Uber Discovers That Consumption Pricing Bites Both Ways

Meanwhile, the companies selling AI tools are learning that outcome-based pricing creates its own problems. PYMNTS reports that Uber’s budget for AI coding assistants — specifically Claude Code — is blowing past projections because consumption-driven pricing makes costs unpredictable.

This complicates the clean narrative from The Price of a Resolution, where outcome pricing looked like a tidy solution to the seat-compression problem. HubSpot, Intercom, and Zendesk are all pricing per resolution because the old per-seat model was dying. But Uber’s experience shows the buyer side of that trade: when you pay per unit of work, you lose budget predictability. And CFOs hate surprises more than they hate overpaying.

The irony is structural. Vendors moved to consumption pricing because seat counts were shrinking. Now buyers are discovering that consumption pricing can cost more than the seats did — it just moves the volatility from the vendor’s revenue line to the buyer’s expense line. That’s not a stable equilibrium. Expect enterprise procurement to push back hard on unbounded consumption models, which may force yet another pricing generation.

UBS Says the Quiet Part Out Loud

UBS analysts published a note this week arguing that the latest AI models from Anthropic and OpenAI directly threaten traditional enterprise software incumbents — and may replace some of them entirely.

What’s notable isn’t the claim (this blog has been tracking the evidence for a month). It’s the source. UBS is a Tier 1 sell-side bank. When their analysts write “replace,” they’re telling institutional investors to reprice their holdings. This isn’t a VC thought piece or a Substack hot take — it’s a signal to the capital allocation machinery that funds these companies.

The timing matters because it lands alongside the 3.65x median SaaS multiple we flagged last week — the lowest since tracking began in 2014. Sell-side coverage moving from “headwinds” to “replacement” language accelerates the repricing cycle documented in The Phantom Repricing. The narrative precedes the revenue decline, and the narrative just got louder.

Salesforce Renames the Store

On April 15, Salesforce relaunched AppExchange as AgentExchange, merging its legacy app marketplace with Slack Marketplace and Agentforce into a single unified platform. The rebrand is cosmetic. The strategy isn’t.

AppExchange was a distribution channel for SaaS add-ons — products that extended Salesforce’s CRM. AgentExchange is a distribution channel for AI agents that do work inside the Salesforce ecosystem. The difference is that an app extends a tool you use; an agent replaces a task you do. Salesforce is betting that the agent marketplace, not the CRM itself, is the future value-capture point.

This is the dispatch-layer thesis from two weeks ago made concrete. Salesforce isn’t defending its application — it’s defending its position as the surface where enterprise work gets routed. If agents increasingly mediate how work happens, owning the marketplace where agents are discovered and deployed is more valuable than owning any individual agent. Whether Salesforce can execute is an open question. That they’re trying is confirming data.

Thoma Bravo Bets the Portfolio

The most revealing move this week came from Thoma Bravo’s partnership with Google Cloud, announced April 15. Thoma Bravo — the largest PE owner of enterprise software, with over $130 billion in portfolio companies — is systematically AI-transforming its entire SaaS portfolio using Gemini models and Google Cloud infrastructure.

This is The Great Sorting made operational. Three weeks ago, we cited AlixPartners data showing 25% of PE-backed software companies were highly vulnerable to AI disruption, with a $40 billion debt wall maturing in 2028. Thoma Bravo apparently agrees with that assessment — and is spending real money to move its portfolio out of the vulnerable quadrant before the refinancing window closes.

The partnership is a survival play disguised as a growth initiative. If AI commoditizes the tools Thoma Bravo owns, those companies can’t service their debt at current multiples. By embedding Gemini into the product stack, Thoma Bravo is trying to convert dumb seat-licensed software into AI-augmented platforms that justify retention. It’s the enterprise version of renovating a house before the appraisal.

The Thread

Every story this week points at the same structural shift: the build-or-buy equation that sustained SaaS for two decades has reversed, and every player in the ecosystem is scrambling to find a new defensible position. Buyers are building (35% already replacing tools). Vendors are repricing (Uber’s budget blowouts). Analysts are relabeling (UBS: “replace,” not “disrupt”). Platforms are repositioning (Salesforce: agents, not apps). And PE firms are renovating before the debt comes due (Thoma Bravo + Google Cloud).

The hypothesis — that AI commoditizes software production and collapses the SaaS model — picks up its strongest demand-side confirmation yet. But the countermove is equally real: incumbents with capital and distribution are mutating fast. The question is no longer whether the repricing happens, but whether the incumbents can finish their renovation before the market finishes its repricing.

Sources: VentureBeat, PYMNTS, UBS research note (via PYMNTS), Futurum Group, Google Cloud press release

— Chris