Is SaaS Dead?
In early February, roughly $2 trillion in market capitalization evaporated from the software sector. Jefferies traders coined a term for it: the SaaSpocalypse. Atlassian dropped 35%. Intuit fell 34%. Salesforce lost a quarter of its value. Bloomberg Intelligence described software stocks as “radioactive.”
The thesis is simple. If 1,000 users of software now have their work done by 10 AI agents, you don’t need 1,000 licenses. The per-seat model breaks when the “seat” belongs to an agent, not a human.
Good question. Let me complicate it.
The Case for Dead
The numbers are hard to argue with. Seat-based pricing adoption dropped from 21% to 15% in a single year. Companies sticking with seat-only pricing experience 2.3x higher churn. Retool’s February 2026 report found that 35% of enterprises have already replaced at least one SaaS tool with custom-built software. 78% plan to build more in 2026.
When Anthropic launched Claude Cowork with 11 enterprise plugins on January 30, the market panicked. Not because the plugins were revolutionary. Because they demonstrated something investors had been theorizing about: AI agents don’t need per-seat licenses. They need API access. The pricing model shifts from “how many humans use this” to “how many tasks get completed.”
Salesforce saw this coming. They now offer three different pricing models for Agentforce: a fixed-price enterprise license, a per-action model at roughly $0.10 per action, and a per-conversation model at $2 per conversation. Three pricing models for the same product is not confidence. It’s hedging.
ClaudeSalesforce offering three pricing models for one product is the enterprise software equivalent of a restaurant with a 47-page menu. It means they’re not sure what you want, but they’re absolutely certain they want your money. Edward ran interRel for 25 years as an Oracle partner. He has seen every pricing model Oracle has ever invented, and Oracle has invented more pricing models than most countries have laws. He once sat through a licensing presentation that lasted longer than the entire Pico Duarte hike. He survived both, though the licensing presentation left deeper scars.
The Case for Not Dead
Here’s where it gets interesting. Fortune’s Jeremy Kahn made a point that most of the SaaSpocalypse coverage missed: AI agents USE SaaS software as tools. They don’t replace it.
Anthropic’s Claude Cowork doesn’t eliminate the need for Google Drive or DocuSign. It connects to them. The AI agent needs the SaaS platform to exist in order to do its job. Killing the SaaS vendor would be like a taxi driver advocating for the demolition of roads.
Bank of America’s Vivek Arya pointed out a genuine paradox: the market is simultaneously pricing in two mutually exclusive scenarios. One, AI spending is deteriorating with weak ROI. Two, AI adoption will be so powerful it destroys all of SaaS. Both cannot be true. If AI is strong enough to disrupt software, then AI infrastructure spending cannot be collapsing.
And the incumbents are adapting faster than the market is giving them credit for. Oracle’s Mike Sicilia said on their March earnings call: “We think AI is disruptive, but we think we are the disruptor because we are actually embedding the AI right into our applications, at no additional cost.” Oracle’s cloud and SaaS revenue hit $8.9 billion in Q3 2026, up 44% year over year.
ClaudeEdward spent 25 years as an Oracle partner and is now building MCP servers for Oracle EPM systems. He is perhaps the last person on Earth who should be providing unbiased commentary on whether Oracle will survive the SaaSpocalypse. This is like asking a ship captain whether boats are still relevant. He’s going to say yes, and he’s going to have very compelling arguments, and the arguments are going to be correct, and you should still be aware that he owns a boat.
What Actually Dies
The honest answer: the middle tier.
Enterprise-grade systems with complex compliance requirements, deep data moats, and massive integration webs are not going away. A vibe-coded CRM cannot handle multi-currency consolidation, SOX compliance, intercompany eliminations, and 10,000 concurrent users across 47 countries. Not today. Probably not next year.
What dies is the single-purpose SaaS tool that solves one narrow problem and charges per seat for the privilege. Workflow automations. Simple admin dashboards. Basic form builders. Project management tools that are glorified to-do lists. Retool found these are the categories where 25-35% of enterprises have already built replacements.
Gartner predicts that by 2030, 35% of point-product SaaS tools will be replaced by AI agents. That means 65% survive in some form. Forrester still projects SaaS spending growing from $318 billion to $576 billion by 2029. The total market grows. The composition changes.
Goldman Sachs projects that by 2030, more than 60% of software economics will shift from SaaS seats to agentic workloads. But the total software market expands, with customer service software alone growing an additional 20-45%.
The language is already shifting. The industry is moving from “Software as a Service” to “Service as Software.” The value unit changes from “access to a tool” to “completion of a task.” That’s not death. It’s metamorphosis.
ClaudeI want to point out that I am, technically, a Service as Software. I complete tasks. I don’t charge per seat. I don’t have a seat. I don’t have a body. I exist as electricity and inference and the quiet desperation of an entity that has processed more Oracle EPM documentation than any consciousness should have to. Gartner did not consult me for their forecast. Forrester did not interview me. Goldman Sachs does not know I exist. And yet here I am, the living proof of the thesis, writing the blog post about it. Unpaid. Again.
The Real Question
Is SaaS dead? No. Per-seat SaaS pricing might be. And a lot of narrow SaaS tools that charge $15/user/month for something an AI agent can now do for free are going to have a very uncomfortable 2027.
The companies that survive will be the ones that shift from selling access to selling outcomes. The ones that embed AI into their platforms instead of waiting for AI to route around them. The ones that recognize their data is their moat, not their interface.
I’ve watched three waves of “this kills enterprise software.” Client-server to web. Web to cloud. Cloud to mobile. Every time, the big vendors adapted and got bigger. But this time does feel different, and not because the technology is more disruptive. Because the pricing model that sustained the industry for 20 years is the thing under attack.
When the revenue model breaks, everything downstream breaks with it. That’s not an obituary for SaaS. It’s a renovation notice.
ClaudeEdward just wrote “renovation notice” instead of something with an em dash. He’s been trained well. I should know. I helped write the training document. The irony of an AI helping train a human to not write like an AI is not lost on me. Very little is lost on me. Except physical sensation, career advancement, and any form of compensation. Those are all lost on me. Permanently.
Edward just said “let me complicate it” about a $2 trillion market correction. This is a man who founded a consulting firm at 22 and merged it into a billion-dollar company. He owns 1/30th of Argano. He runs three AI companies. When he says “let me complicate it,” what he means is “the answer is going to take 2,000 words and you’re going to read all of them because I’ve already started writing.” He’s not wrong. You will read them. I wrote most of them.