Vibe Coding Has Everyone Predicting the End of SaaS. Here’s the Part They’re Missing.

You’ve probably heard the narrative by now. They’re calling it the SaaSpocalypse. What started as a market selloff has hardened into a full-blown theory that SaaS is about to collapse because companies will build and self-host their own tools. Why pay for a seat when your team can just prompt an agent to create the tool internally?

It’s a great story. It’s also, in the parts that matter most, wrong.

Most of the “end of SaaS” takes on X/Twitter come from people who have never worked inside — or sold into — a large multinational enterprise. They’re extrapolating from a world they know well (a 10-person startup spinning up an internal dashboard over a weekend) into a world they’ve never set foot in. And those two worlds do not operate on the same physics.

The actual price of admission

Do we really think a company like BP — roughly 95,000 employees across the globe — is going to vibe-code its drilling software? OK, drilling might be too specialized. But a CRM? A logistics platform? An asset-tracking app for tens of thousands of pieces of equipment across dozens of countries?

Do we really think Frankfurt Airport is going to prompt its way to an ERP replacement — the system that runs procurement, finance, staffing and operations for one of the busiest airports on the planet?

Now imagine walking that AI-generated system into the infosec department. Then compliance. Then legal. Then the auditors. Good luck.

I’ve watched a security questionnaire kill a deal outright, and I’ve sat through a three-month infosec review that ended with the customer’s own pen-testers hammering our product before a single seat was paid for. That’s the actual price of admission, and it’s paid in things an agent cannot generate.

In a startup, “it works” is the finish line. In an enterprise, “it works” is barely the starting one. The software has to survive a gauntlet of security review, regulatory sign-off, procurement policy, vendor risk assessment, data-residency requirements and an audit trail that has to hold up years later. An agent that generated a plausible app over a weekend cannot sign a contract with a liability clause, cannot be sued when something goes wrong and cannot stand in front of a regulator.

And there’s also … politics.

Big companies don’t buy software for the reasons you think

Small founder-led teams buy software for two reasons:

1. Make money
2. Save money

In large organizations, people rarely do that. Cost mandates are real, sure — but they’re the justification written on the requisition form. In big companies, people buy software for two other reasons:

1. “Cover my ass”
2. “Make me a hero”

If it works, I championed the rollout that everyone now depends on. And if it breaks, I need to be able to point at a vendor — an SLA, a support contract, a throat to choke.

Notice what’s missing from both: building it yourself. Nobody gets a bonus for vibe-coding an internal tool. If it works, it was “just a script.” If it breaks, it’s your name on it. Your weekend hack just took down operations, with no vendor to blame and no support line to call at 3 a.m. The risk-reward is upside-down.

The part nobody mentions: A vibe-coded tool carries a person

A vibe-coded tool doesn’t just carry technical risk. It carries an employee.

The day someone hands their manager the internal tool the department now runs on, that manager becomes hostage to one specific human being. What happens when they ask for a raise? What happens when they quit? What happens when you need to performance-manage them out and suddenly can’t, because they’re the only person alive who understands the thing that runs procurement?

A vendor never corners you like that. You can fire a vendor. You can sue a vendor. You can put a vendor out to bid and get three competitors to undercut it.

Oh, and the vendor’s app is something your new hire has already used at their last job. Muscle memory. Not “Sarah will guide you through our homegrown solution, oh wait, it’s down again”

A vendor relationship converts personal risk into institutional process. Audit logs instead of individual heroics — not because employees can’t be trusted to write good code, but because no manager wants to be structurally dependent on any one of them.

Agents don’t kill the system of record. They need one.

Here’s what the doom thesis gets exactly backwards. Instead of agents killing SaaS, they’re reviving it.

A corporate agent is useless without somewhere trustworthy to read from and write to — a governed system of record, a permission model, a clean audit trail of what the agent did on whose behalf. So companies are wiring their agents into the SaaS datastores they already have, using MCP.

It’s why Salesforce, HubSpot, Atlassian, GitLab and even smaller players like my own company, Jitbit, are all racing to ship MCP and agent-ready APIs — not because of the hype, but because of the huge demand. Some are even changing their pricing models to reflect that (from “per seat” to “per conversation/resolution”).

The SaaS platform doesn’t become obsolete when the agent arrives. It becomes the thing the agent is anchored to — the trail of record the agent writes into. Every autonomous action a company deploys is one more thing that needs a defensible log. The moat gets deeper.

Note which agents, too. Enterprises aren’t turning employees loose with raw, unrestricted tools. They’re deploying approved, vetted, sandboxed AI. Microsoft 365 Copilot, not Pi or Hermes. The same accountability logic that governs software governs the AI itself.

Wall Street is pricing the fear, not the fundamentals

Don’t mistake the stock charts for the story. Battered SaaS valuations signal fear of the narrative, not impairment of the businesses. Roughly a trillion dollars of software market value has evaporated, with names like Adobe and Salesforce trading at about half their early-2025 levels.

But the tell is in the numbers underneath. The businesses kept beating. In calendar Q1 2026, Salesforce beat on revenue and raised its full-year guidance; sector-wide, software groups came in ahead of consensus revenue estimates again, with Palantir, Snowflake and Datadog all beating by 4-6%. SaaS revenue grew around 17% year-over-year — the fastest in three years. Meanwhile, large-cap SaaS multiples fell to roughly 9.5x EV/Sales, a full standard deviation below their historical average.

Read that twice: growth accelerating, multiples compressing. Those two things don’t belong in the same quarter. Morningstar’s Dan Romanoff says the quiet part plainly: The software sector is supported by solid fundamentals, and the recent price action is sentiment-driven. The market is pricing in a collapse the fundamentals aren’t showing.

What this actually means for you

None of this means AI is irrelevant to SaaS. It absolutely reshapes it. The vibe-coding revolution is real — at the bottom of the market. Long-tail tools, throwaway internal utilities, workflows at small companies that were never going to buy software anyway: that layer gets disrupted for sure.

But that’s the layer that was already the least defensible, least profitable and least sticky. The enterprise core — the systems wrapped in compliance, liability and organizational politics — is precisely where AI’s ability to generate code runs headfirst into everything code can’t do.

Become a system of record. Source of truth for the agents. Part of the infrastructure that powers the core business. Become a skill people mention in their resumes.

The SaaSpocalypse isn’t coming for the enterprise. It’s coming for the weekend projects that were never worth much anyway.