Cloud Cost & Pricing Transparency
The AI Infrastructure Stack · Part 1 of 3
The software-as-a-service selloff of early 2026 has been brutal, clinical, and entirely predictable, if you were paying attention. Salesforce, ServiceNow, Oracle, Adobe, HubSpot: week after week, the names read like a who’s who of enterprise software, and the direction has been one way. Software stocks shed more than $300 billion in combined market value in a single week in February alone. The SaaS ETF benchmark dropped nearly 20% year-to-date. Analysts have taken to calling it the “SaaSpocalypse.”
Wall Street, characteristically, is treating this like a revelation. It isn’t one. The business model crack was always there. AI agents just made it impossible to ignore.
Russ Artzt has seen this movie before. The co-founder of CA Technologies, one of the largest enterprise software companies ever built, and former executive chairman and head of R&D at RingLead, acquired by ZoomInfo, Artzt spent decades navigating the shifts from mainframe to client-server to cloud. In a recent conversation with DataStorage.com, he laid out exactly why the per-seat model is breaking, who’s responding well, and why the companies sitting still are in serious trouble.
To understand why the reckoning is happening now, you have to understand why per-seat pricing was so elegant in the first place. Artzt lived it firsthand at CA, and then watched Salesforce turn it into a religion.
“A flat fee: here’s what the platform costs, here’s what the app costs. Not based on the software itself, but on how many users you have using that app,” he said. “And that user model just changed the business model.”
The genius of the user model wasn’t just pricing mechanics. It was what it did to revenue recognition. Under the old perpetual license model, a company like CA could close a $100,000 deal and book it immediately. Under SaaS, that same revenue gets spread across months and quarters. On the surface, that sounds worse. In practice, it turned out to be far better. “You can predict revenue for companies,” Artzt said. “It keeps going and keeps going and multiplies.” Analysts loved it because it was predictable. Wall Street built entire valuation frameworks around it. Salesforce, which Artzt noted became “one of the biggest SaaS platforms” by pushing the model hardest in the early 2000s, helped turn per-seat from a billing choice into the defining logic of a generation of enterprise software.
The model proliferated fast. CA adopted it. Oracle adopted it. Microsoft adopted it. ServiceNow adopted it. “Everybody did it,” Artzt said. And for two decades, everybody was right to. But the architecture that made per-seat so durable is now the same thing making it so hard to escape. Based upon the application’s architecture, some apps will need to be rewritten from scratch and others will be re-architected to work with new AI agents, and knowing which path your product sits on is the first decision every SaaS company has to make right now.
Here’s the problem with per-seat pricing when agents enter the picture: agents don’t need seats.
The logic of per-seat is that value scales with the number of people using the software. More users, more value delivered, more revenue justified. AI agents invert that assumption entirely. Enterprises are already reporting “seat compression” during contract renewals: companies that once required 500 licenses discovering they can achieve the same output with 50, because autonomous agents are doing the remaining work around the clock.
Artzt put it plainly: “You can’t charge them by user. You have to charge them for the platform or the app and have a fixed charge.”
The market has noticed. According to Growth Unhinged’s 2025 State of B2B Monetization report, seat-based pricing dropped from 21% to 15% of companies in just 12 months, while hybrid pricing surged from 27% to 41%. Bain’s Technology Report 2025 put the challenge in stark terms: “If an agent replaces a human task, customers will expect to pay based on outcomes, not log-ons.”
The transition isn’t just theoretical. Intercom made the switch in 2023, abandoning per-seat pricing for a per-resolution model, and saw 40% higher adoption within six months. SaaStr has noted that its own organization is already downgrading seat counts at vendors now that they have more than a dozen AI agents running in production. The seat math just doesn’t work anymore.
“You can’t charge them by user. You have to charge them for the platform or the app.” (Russ Artzt)
The pricing disruption is inseparable from a deeper shift in how software looks and works. SaaS built its kingdom on dashboards, workflows, and carefully designed user interfaces. AI agents change that picture, but not in the way most people assume.
“When you’re running an AI agent, you’re actually seeing two things on screen at once,” Artzt explained. “You see the agent’s own interface, the app itself, and alongside it, the chat interface connected to that agent. The user is interacting with both simultaneously.”
That coexistence is the key point. This isn’t chat replacing the app. It’s chat running alongside it, giving users a way to query, instruct, and interact in natural language while still having full visibility into what the agent is doing. Traditional SaaS gave users one interface to master, and they had to learn it on the software’s terms: navigate menus, configure settings, run queries. Agentic software inverts that relationship. The software meets the user where they are, and the conversation becomes the control layer.
That shift is already reshaping how enterprises evaluate tooling. Deloitte’s 2025 Tech Value survey found 57% of respondents were allocating between 21% and 50% of their annual digital transformation budgets to AI automation. The buying center is shifting from “which SaaS tool does this?” to “which agent handles this workflow?”
Not all incumbents are caught flat-footed. Artzt was direct about who has navigated the transition well, and who still has serious work to do.
On Google: “They’ve integrated search with AI. It’s not pretty, you still see Google Search, you hit an AI button, but you know what? It works. And it’s powerful.” Google’s approach, integrating Gemini across Search, Gmail, and its Workspace suite while launching net-new AI products like Notebook LM and Vertex AI, earns Artzt’s nod as the most coherent response in the market. “I think Google has done it the best.”
On ServiceNow: Artzt called out their approach as a model for how solid legacy apps should respond. “I would re-architect my app, not rewrite it, but re-architect it to make use of all the new agents that I write.” The stock has been punished anyway, down roughly 45% from its 2025 peaks, though its underlying business, Q4 revenue up 21% year-over-year with $15.5 billion projected for 2026, tells a different story than its share price.
On Salesforce: Salesforce’s Agentforce push has been aggressive, but the market isn’t convinced. The company has lost roughly 40% from its 2025 highs, and CEO Marc Benioff’s attempts to reframe the AI threat as an opportunity (“We’ve got all the customers’ data”) have not yet moved investors. At $500 per seat per month for enterprise tiers, the seat compression math is punishing.
On Microsoft: Artzt’s read is that Copilot is a decent step, but not a sufficient one. “It’s much smarter, more intuitive” than old-style help documentation, he allowed, but it falls short of a genuine agentic architecture. “I think eventually Microsoft’s going to have to go further.”
For technology leaders navigating this, whether they’re evaluating their own software stack or running a SaaS company, Artzt offers a framework that maps to business reality rather than analyst idealism.
The first question is whether your existing app is worth saving. “Is my app worth saving? Or is it old? Is it buggy? Does it cause a lot of problems? Am I up every night supporting customers because it’s falling apart?” If the answer to those questions is yes, the only honest path is a rewrite. Stop patching a broken foundation.
If the app is solid, reliable, well-used, and generating real revenue, then the move is integration first, not a full rebuild. “I would re-architect and then integrate to start,” Artzt said, pointing to ServiceNow as the model to emulate: keep what works, expose APIs, and wire new agents into the existing platform rather than replacing it wholesale.
But he’s clear that integration is a bridge, not a destination. “Eventually, I think everybody’s got to rewrite.”
The talent implication is real and immediate. SaaS companies will need to run parallel tracks: keeping existing development and support staff on legacy applications that still generate billions in revenue, while simultaneously hiring or retraining a separate cohort for agent development. Artzt drew the parallel explicitly to CA’s mainframe-to-client-server transition in the 1990s. He split his teams and gave people a choice. The ones who moved to the new architecture built the products that drove the next decade of growth. The ones who stayed on legacy systems kept the lights on. Both were needed. The AI talent market has, predictably, not made this easy: the demand for skilled AI engineers is running well ahead of supply, with compensation reaching seven figures for top-tier architects.
The SaaS business model was a twenty-year run that reshaped enterprise computing. It isn’t dead. The systems of record (the CRMs, ERPs, HCMs) aren’t being ripped out next quarter. But the unit economics that justified their valuations are being renegotiated, and the companies that insist on charging for seats in a world where agents don’t need them are going to find that conversation increasingly uncomfortable.
“The software companies are not dead,” Artzt said. “They’re very much alive. But the companies that don’t do it and just leave things alone are likely to slow down and stagnate.”
That’s not a eulogy. It’s a warning. The reckoning isn’t about AI replacing software. It’s about the industry repricing itself for a world where the work gets done differently, the interface looks different, and the user who justifies the seat may, increasingly, not be a human at all.
Part 2 of this series examines what the agentic software shift means for the infrastructure layer beneath it, and why storage is becoming the anchor of the AI stack. Coming next: “Agentic Software Will Explode Storage and Compute Demand: Why Storage Becomes the Anchor”
Russ Artzt is co-founder of CA Technologies and former executive chairman and head of R&D at RingLead, acquired by ZoomInfo. He speaks with DataStorage.com regularly on AI infrastructure, enterprise software strategy, and the evolving data stack. Connect with him on LinkedIn.
