"SaaS Is Dead" Is the Wrong Bet, Here's the Steelman, and Why It Still Loses

Table of Contents

• Where the "SaaS Is Dead" Thesis Comes From
• The Steelman: Building the Strongest Case for Death
• Seats Were Always a Proxy, and the Proxy Is Breaking
• Agents Collapse the Workflow the SaaS Used to Host
• The Buyer Stops Buying Software
• The Rebuttal: Why the Obituary Is Premature
• Agents Need Something to Stand On
• Distribution and Trust Don't Transfer
• "Dead" Confuses the Pricing Model with the Product
• What Actually Happens: Repricing, Not Death
• Insights Most People Overlook
• References

Where the "SaaS Is Dead" Thesis Comes From

The slogan went viral in 2024 and 2025, but the underlying observation is older than the agents. SaaS spent two decades selling access to software measured in seats, pay per person who logs in. That model worked because software was a tool that humans operated. The unit of value was a human plus a license.

Agentic AI-as-a-service breaks that arithmetic at the root. If you can buy a sales-development agent that drafts, sends, and follows up on outreach autonomously, you are no longer buying ten seats of a sales-engagement tool for ten reps. You're buying an outcome, meetings booked, and paying per task or per result. Microsoft's own framing of agents as "the new apps" and Satya Nadella's much-quoted line that agents could collapse the business-logic tier of SaaS gave the thesis a brand-name megaphone. When the CEO of the largest enterprise-software vendor on earth muses publicly that the SaaS model might dissolve, the idea stops being a Twitter take and starts being a board-meeting question.

So the thesis deserves a real steelman, not a strawman. Let me build the strongest version before I take it apart.

The Steelman: Building the Strongest Case for Death

Seats Were Always a Proxy, and the Proxy Is Breaking

Here's the uncomfortable truth seat-based pricing never wanted you to notice: nobody actually wanted the seat. Companies bought seats because that's how value got metered, not because a login had inherent worth. The seat was a proxy for "a person doing work inside this tool."

Agents sever the link between headcount and software consumption. When a single agent handles the workload of a ten-person team, the customer needs one agent, not ten seats, and the vendor's revenue per account can fall by 80 to 90 percent overnight if pricing doesn't change. This is the core of why, as we cover in #286, When one agent replaces a ten-seat team, SaaS pricing breaks, the entire revenue architecture of seat-based SaaS is structurally exposed. It's not a marketing problem. It's a unit-economics problem baked into the billing model.

Investors have noticed. The metric that used to signal health, net revenue retention driven by seat expansion, is exactly the metric now under threat, which is why seat-count shrinkage has become a number SaaS boards quietly track. The bull case for most SaaS companies assumed seats only go up. Agents invert that assumption.

Agents Collapse the Workflow the SaaS Used to Host

The second pillar of the steelman is sharper than pricing. SaaS products are, fundamentally, hosted workflows. A support tool hosts the ticket-resolution workflow. A CRM hosts the deal-progression workflow. The software's job is to give humans a structured place to do the steps.

An agent doesn't need the structured place. It does the steps. If an agent reads the inbound email, checks the order system, issues the refund, and writes the reply, the support tool's carefully designed ticket queue becomes scaffolding nobody climbs. The dashboard that displayed the work is irrelevant when the work just happens. This is the pattern that genuinely threatens incumbents, the disappearing dashboard, the agent that acts instead of displays.

Andreessen Horowitz has argued that AI is poised to compress the application and services layers, with agents capturing value that used to flow to software licenses and to the humans operating that software, a thesis laid out in a16z's analysis of how AI reshapes the software stack. If you believe the workflow is the product, and agents perform workflows, the conclusion writes itself.

The Buyer Stops Buying Software

The third pillar is about the buyer, not the technology. Procurement is shifting from "buy software, then hire people to run it" toward "buy the outcome." A CFO doesn't care whether a refund was processed by a Zendesk seat or a refund agent, they care about cost per resolved ticket. When the purchase becomes an outcome, the line item moves from the software budget to the labor budget.

That reframing is existential for SaaS positioning. Software vendors have spent twenty years convincing buyers that software is a capital-efficient investment that scales without headcount. Agents let buyers reframe the same spend as opex labor, and labor budgets are far larger than software budgets, but they're also evaluated on cost-per-outcome, where a cheaper agent always wins. The vendor loses its premium and its pricing power in one move.

That's the steelman. Seats are a broken proxy, agents eat the workflow, and the buyer stops buying software. Taken together, it's a coherent and genuinely uncomfortable case. Now here's why it still doesn't add up to "dead."

The Rebuttal: Why the Obituary Is Premature

Agents Need Something to Stand On

The first crack in the death thesis is the most fundamental: agents are not self-sufficient. An agent that issues a refund still needs a system that holds the order, the payment record, the customer history, the inventory state, and the audit trail. That system is SaaS. The agent is a brilliant operator standing on top of a foundation it did not build and cannot easily replace.

This is the system-of-record versus system-of-action distinction, and it matters enormously. The agent is the system of action. The SaaS database is the system of record. You can swap the actor far more easily than you can swap the record, migrating a company's entire transactional history, integrations, and compliance state is a multi-year project that almost nobody undertakes voluntarily. The incumbent's data gravity is a moat agents can't easily cross, and it's why Salesforce can bolt Agentforce onto its existing data layer and charge for it, rather than being disintermediated by an outside agent that has no access to that data in the first place.

McKinsey's research on enterprise AI adoption keeps surfacing the same bottleneck: the hard part isn't the model, it's the data plumbing, governance, and integration that production agents require, work that overwhelmingly runs through existing SaaS systems, as detailed in McKinsey's state of AI in the enterprise. Agents don't kill that infrastructure. They depend on it.

Distribution and Trust Don't Transfer

The second rebuttal is unglamorous but decisive: incumbents already sit inside the enterprise. They have the contracts, the security reviews, the procurement relationships, the SOC 2 reports, the data-processing agreements, and the IT department's grudging trust. A new agent startup has to win all of that from zero.

Enterprises do not rip out a vendor that survived three years of security audits to adopt an unproven agent that promises to do the same job cheaper. Distribution beats capability more often than founders want to admit. The incumbent's real advantage isn't its software, it's the painfully accumulated permission to operate inside regulated, risk-averse organizations. When ServiceNow or Salesforce ships an agent, it ships through a channel that already reaches the buyer. The startup agent ships through a cold email.

This is why "we're AI-native" became table stakes overnight rather than a winning differentiator. Once every incumbent adds competent agents, native-ness stops being a wedge. The thin-wrapper startups that bet on capability alone discover that capability is the easy part to copy and distribution is the hard part to acquire.

"Dead" Confuses the Pricing Model with the Product

The third rebuttal is the one the slogan most cleverly hides. "SaaS is dead" smuggles a switch: it observes that seat-based pricing is breaking, then declares that software-as-a-service is dying. Those are different claims.

Seat-based pricing is one monetization model. The delivery model, software hosted in the cloud, multi-tenant, continuously updated, accessed over the network, is completely orthogonal to how you bill for it. An agent sold per-outcome is still, architecturally, software-as-a-service. It runs in someone's cloud, it's multi-tenant, it updates continuously. Agentic AI-as-a-service is not the antithesis of SaaS. It is SaaS with a different meter.

When you separate the two claims, the death thesis shrinks to something true but narrow: seat-based pricing is dying. That's correct, and it's a big deal, but it's a repricing, not an extinction. The companies that confuse "our pricing model is obsolete" with "our category is dead" will make the wrong strategic choice, either freezing in denial or burning the business down in panic.

What Actually Happens: Repricing, Not Death

So strip away the slogan and what's the honest forecast? Three things happen at once, and they're uneven.

First, the strong incumbents reprice and survive. Salesforce, ServiceNow, Microsoft, and the better vertical-SaaS players bolt agents onto their data and distribution, shift from seats toward consumption and outcome metrics, and defend the system of record. They take a revenue hit on seat expansion and try to recapture it on agent consumption. Some succeed; the ones with genuine data moats mostly do.

Second, the thin middle gets gutted. Categories that were purely workflow scaffolding with no proprietary data, generic form builders, simple automation layers, single-function tools that exist only to give humans a place to click, get absorbed into agents. RPA is the clearest casualty; it was always a brittle automation of human clicks, and agents do that more robustly. Workflow-automation incumbents face the same squeeze from the other direction.

Third, net-new GaaS-native companies win the greenfield. Where there's no incumbent data moat and the workflow is genuinely automatable end-to-end, agent-native startups can build outcome-priced businesses from scratch and win categories outright. This is the real disruption, not the death of SaaS, but the birth of a parallel category that competes on a different axis.

The five-year picture is a transition, not a funeral. SaaS valuations are being rewritten, Gartner's tracking of the broader software market shift toward AI-augmented and consumption-based models confirms the repricing is already underway, per Gartner's forecast on enterprise software and AI spending. The seat-based bull case is over. The category is not.

Insights Most People Overlook

The death thesis is loudest where the data moat is weakest. Notice who's most bullish on "SaaS is dead": foundation-model companies, agent startups, and VCs holding agent-native portfolios. They're not wrong, but they're talking their book. The thesis is strongest precisely in the categories they can attack, thin-data, pure-workflow tools, and conveniently quiet about the system-of-record incumbents they can't dislodge. The slogan generalizes a real local truth into a false global one.

Outcome pricing can be worse for the buyer, not just the vendor. Everyone assumes per-outcome pricing favors the customer. But outcome pricing hands the vendor a claim on the upside of the work. A seat costs the same whether the rep books one meeting or fifty. An agent priced per booked meeting captures more as it gets better, and the customer loses the deflationary benefit they'd get from a flat license over a tool that improves for free. Smart procurement teams will fight outcome pricing, not embrace it, once they do the math on a high-volume workflow.

The most disruptable SaaS isn't the cheapest, it's the most workflow-shaped. People assume agents kill low-end tools first. The real predictor of disruption is how much of the product is "a place for humans to do steps" versus "a place where proprietary state lives." A pricey but data-rich vertical tool is safer than a cheap but generic workflow app. The disruption axis is data-density, not price.

Incumbents may cannibalize themselves slower than startups can attack, and that's the actual risk. The incumbent's dilemma is real: repricing from seats to consumption means voluntarily shrinking near-term revenue, which public-company CFOs hate. The death thesis underrates how this internal incentive problem, not the technology, is what actually kills incumbents. The threat isn't that agents are better. It's that incumbents won't reprice fast enough to compete with their own agent-native rivals.

"Agent of record" is the next lock-in, and it's stickier than seats ever were. The company whose agent owns the customer relationship and accumulates the operational context becomes the new switching-cost anchor. We spent a decade worrying about data lock-in. The next lock-in is the agent that knows how your business runs. That's a deeper moat than any seat-based contract, and it means the post-SaaS world may be less competitive, not more.

References

• The Economic Case for Generative AI and Foundation Models, Andreessen Horowitz (a16z)
• The State of AI, McKinsey & Company (QuantumBlack)
• Gartner Forecasts Worldwide IT Spending, Gartner Newsroom