Fordel StudiosOn February 3, 2026, Anthropic pushed eleven plugins to GitHub under the product name Claude Cowork. The plugins targeted enterprise workflows directly — legal research, compliance review, sales preparation, meeting summarisation, contract drafting. By end of trading that day, roughly $285 billion had been erased from software sector market capitalisation. The S&P 500 Software & Services Index fell nearly 13% — its worst single-day performance on record.
Jeffrey Favuzza, an equity trader at Jefferies, named the event SaaSpocalypse. By mid-February, the total sector losses had widened to approximately $1 trillion. The WisdomTree Cloud Computing Fund (WCLD) was down around 20%.
The question that followed — "Is this a market panic or a structural collapse?" — turned out to be the wrong question. Both things are true at once. The panic was real and partly irrational. The structural problem it surfaced is also real and, in the long run, more consequential.
The Mechanism: Why Per-Seat Pricing Is the Exposed Nerve
SaaS companies charge per seat — per human user. It is a pricing model built on the assumption that every worker who derives value from the software holds a licence. That assumption held for roughly thirty years.
AI agents do not hold licences. One agent can do the work of ten to fifteen employees — data entry, status updates, report generation, ticket routing, compliance checks, contract review. The same workflow continues. The seat count does not. Revenue per account shrinks even when the product itself remains in use.
This is the first mechanism, and it is the one that drove the February selloff. Investors did not suddenly believe that Salesforce or Workday software stops working. They recalculated what the revenue ceiling looks like when the human headcount those tools are priced against starts declining.
Three Mechanisms, Not One
The Seat Problem
The immediate mechanism described above. Agents do not buy licences. As enterprise headcount flattens or falls in certain functions, per-seat revenues compress. The software may stay in place as a backend system; the human interface layer — and the billing attached to it — contracts.
Gross Margin Erosion
SaaS companies historically operated at 75–85% gross margins. That margin justified the valuation multiples. To remain competitive in an agentic world, these companies now pay significant fees to hyperscalers — Azure, AWS, Google Cloud — to run AI features on top of their platforms. The infrastructure costs that SaaS avoided by selling software are now coming back through the AI layer. The margin story that underpinned a decade of valuation is under structural pressure.
Vibe Coding and Custom Displacement
A third mechanism is slower but potentially more disruptive over a five-year horizon. AI coding tools have made custom software meaningfully accessible to smaller teams and to engineers who previously could not build at speed. The argument — popularised by The SaaS CFO as the "vibe coding threat" — is that horizontal SaaS tools solving generic problems (scheduling, basic CRM, project tracking, form management) are now buildable internally in days rather than months. Why pay $500/month per seat for a tool that does 80% of what you need when an AI-assisted internal build can do 100% in a week?
Who Got Hit, and by How Much
| Company | Impact | Response |
|---|---|---|
| Salesforce | Down ~28% at peak. Spent $1.2B acquiring AI startups. Pivoting to "agentic work units" (AWU) as a new pricing metric. Raised FY2030 revenue target to $63B. | Marc Benioff appeared on stage with Dario Amodei on Feb 24 to stabilise sentiment. "This isn't our first SaaSpocalypse." |
| Workday | Shares hit 52-week low of ~$117, down ~10% despite beating Q4 earnings estimates. | No major public pivot announced. Continued investing in AI features. |
| Adobe | Down 16–22% in the first two months of 2026. Navigating a CEO transition simultaneously. | Limited public response during the selloff period. |
| HubSpot | Down ~20% during the peak SaaSpocalypse weeks. | No major structural pivot announced publicly. |
| ServiceNow | Described by analysts as a "resilient outlier." "Now Assist" platform crossed $600M ACV in late 2025. | Focused on workflow automation framing rather than "creative assistant" positioning. On track for $1B ACV this year. |
| Thomson Reuters / LegalZoom | Hit directly by Claude Cowork's legal automation plugins. | No significant public counter-announcement during the period. |
The pattern across the losers is consistent: companies whose value proposition lives primarily at the human-interface layer got repriced. ServiceNow — the outlier — has spent years positioning itself as workflow infrastructure rather than user-facing software. That framing proved resilient precisely because workflow automation and AI agents are complementary, not competitive.
The Counter-Argument (and Where It Is Correct)
Several credible voices called the selloff an overreaction, and they are not wrong. The strongest counter-arguments:
- Enterprise SaaS encodes institutional knowledge — compliance history, approval hierarchies, audit trails, multi-year integration layers. An AI agent cannot replicate this context without the underlying system. Systems of record are not under threat from agents; they are what agents need to function.
- Remaining Performance Obligations (contracted future revenue) at major SaaS players continued to show double-digit growth through February. Enterprise customers did not cancel contracts. Investors panicked; procurement departments did not.
- Both OpenAI and Anthropic CEOs confirmed their organisations use Slack. The companies building the most capable AI agents are themselves SaaS customers.
- Gartner's 2026 forecast: enterprise software spend rising to $1.43 trillion, up 14.7% year-over-year, with GenAI as the primary driver.
- Scott Galloway publicly said the selloff was "farcical" and that he was buying SaaS stocks — arguing companies like Salesforce and Adobe were trading at decade-low free cash flow multiples.
The nuance that most analysis missed: the selloff was a valuation correction, not a revenue collapse. The market was repricing future revenue multiples based on a revised ceiling, not repricing current contracts. Those are different problems on very different timelines.
“The SaaSpocalypse was a stock market event, not a customer event. Investors repriced the future. Customers kept renewing.”
What Actually Survives
Forrester published the clearest framework: the distinction is not between "SaaS" and "not SaaS" — it is between horizontal point-solution SaaS and vertical or workflow-embedded SaaS.
| Category | Risk Level | Why | Examples |
|---|---|---|---|
| Horizontal point-solution SaaS | High | Generic features, no proprietary data moat, replaceable by AI agents or internal builds | Scheduling tools, basic form builders, simple project trackers, generic survey platforms |
| Horizontal enterprise platforms (CRM, ERP, HRIS) | Medium | Systems of record with deep integration layers — threatened at the interface, not the data layer | Salesforce, Workday, SAP, Oracle |
| Vertical / domain-specific SaaS | Low | Proprietary industry workflows, regulatory context, compliance history that generic AI cannot replicate | Healthcare EMR, legal matter management, financial compliance platforms |
| Workflow infrastructure | Low | AI agents need orchestration layers — these become more valuable as agentic workloads grow | ServiceNow, workflow automation, integration platforms |
Forrester projects vertical SaaS growing from $133.5B (2025) to $194B (2029). The category that is at genuine structural risk — horizontal point-solution SaaS — is also the category that was most generously valued coming into 2026. The correction there is not irrational.
The Pricing Model Transformation
Separate from the valuation question, a genuine structural shift in SaaS pricing is underway and appears irreversible regardless of how the stock market performs.
Zendesk has begun experimenting with outcomes-based pricing for AI agents — charging per resolved support ticket rather than per agent seat. They have described the work as "a work in progress." Salesforce is piloting "agentic work units" — pricing by the task completed rather than the licence held.
This transition is not comfortable for SaaS finance teams. Outcome-based pricing introduces revenue variability that per-seat models do not have. It also creates a measurement problem: what counts as a "resolved" ticket, a "completed" sales task, a "finished" compliance review? The industry is working this out in real time.
What This Means If You Are Building Software
The SaaSpocalypse is a useful forcing function for a question that builders should have been asking for the past two years: what problem does this software actually solve, and is that problem better solved by a dedicated agent than a tool humans log into?
The answers are not uniform. Enterprise software that stores organisational history, enforces compliance logic, and provides auditable records for regulators is not going away. It is, if anything, more valuable in an agentic world — because agents need trustworthy data to operate against. The SAP that survives 2030 is less a human interface and more an authoritative backend.
The software that is genuinely at risk is software whose primary value is helping humans perform tasks that agents now perform faster, cheaper, and without error fatigue. If the majority of your product's interactions are task execution — data entry, status tracking, report generation, routing, scheduling — the structural pressure is real.
The opportunity created by all of this is in the gap between what general-purpose AI agents can do with generic context and what purpose-built AI agents can do with rich, domain-specific context. Claude Cowork with its eleven generic plugins is not the ceiling of what is possible. It is closer to the floor. The teams building agents that understand a specific business's data, approval chains, regulatory context, and operational history are building things that generic agents cannot replicate.
The Private Credit Risk No One Is Discussing
One underreported dimension of the SaaSpocalypse: UBS estimates that private credit default rates could hit 13% if AI disruption to software revenue accelerates — more than three times the projected high-yield default rate. Many VC-backed SaaS companies carry private credit at valuations predicated on per-seat revenue growth. As that growth story gets repriced, the debt structures underneath become fragile.
No venture-backed SaaS unicorn filed a new IPO in 2026, according to Crunchbase. The IPO window that SaaS companies were waiting for has not opened. It may not open on the same terms it existed before February 2026. The exit path for a generation of VC-backed horizontal SaaS businesses just got materially harder.
The Stabilisation and What Comes Next
On February 24, 2026, Dario Amodei and Marc Benioff appeared on stage together. Amodei articulated what has since been called the Human-in-the-Loop doctrine — AI as augmentation, not replacement, agents that amplify human capacity rather than eliminate human roles. The markets began recovering.
The recovery is real, and the doctrine is sensible as far as it goes. But it does not change the underlying repricing. SaaS companies that built their value on per-seat revenue from tasks that AI now performs better will need a different value proposition, a different pricing model, or both. The stage appearance bought time. It did not reverse the structural argument.
Gartner's 2030 forecast — enterprise software spend growing 40%+ with GenAI as the primary driver — suggests the aggregate market is not shrinking. It is transforming. The money that would have gone into thirty-seat Salesforce licences is going somewhere. The question for everyone building software in 2026 is whether they are building something that captures that shift or something that the shift is passing through.