AI vs SaaS: Sell first, ask later—is the market "halfway right"?

```

Nearly one trillion dollars of enterprise software stocks were sold off following the launch of Anthropic’s new products, exposing the market’s excessive panic over AI threats.

Barclays points out that investors have overlooked a key technological distinction: AI tools are indeed eroding the application layer business of SaaS companies, but they are not yet capable of shaking the underlying “system of record” infrastructure—which is precisely the core moat for enterprises like Salesforce and SAP.

Anthropic’s release last week of products such as Claude Cowork became the last straw for enterprise software stocks. Customer relationship management software Salesforce and financial management software Workday have collectively fallen more than 40% over the past 12 months.

Behind this panic selling is investors’ vague understanding of AI’s boundaries. The market widely believes that new generation AI tools from companies like Anthropic and OpenAI will completely replace traditional SaaS software, causing the value of established firms to drop to zero.

According to Barclays' February 10 report “Software Is Not Dead, Just Changing,” this “one-size-fits-all” logic does not apply to most enterprise software companies.

What AI Can and Cannot Do

The essential advantage of generative AI lies in pattern recognition and “first draft generation,” but its probabilistic nature also introduces fundamental limitations. AI excels at tasks that require pattern extraction from massive data sets, like natural language processing and code writing, but struggles in scenarios demanding absolute accuracy.

According to the Barclays report, traditional software operates on deterministic rules, where the same input necessarily produces the same output. AI software is inherently probabilistic, operating through learned behaviors rather than hard-coded logic, and cannot guarantee consistent output each time.

This means AI operates at a higher level of abstraction as software, rather than being a substitute for traditional software.

This technical characteristic defines the boundaries of AI’s applicability. In error-tolerant scenarios, such as knowledge work and content generation, AI can replace or even surpass traditional SaaS applications; but in fields like billing, compliance auditing, and business rule enforcement that require “the one correct answer”, AI is currently not up to the task.

Independent analyst Benedict Evans points out that successful SaaS products stem from mapping unique organizational problems into workflows, then encoding them as software. These custom business rules, accumulated over years, are the foundation of infrastructure for banks, hospitals, and retailers, and the bedrock of companies like Epic Systems and Oracle.

The “System of Record” Layer Is Hard to Replace

The Barclays report explicitly states that three categories of enterprise software companies have been incorrectly priced during the selloff and deserve investors’ renewed attention.

First is the System of Record type of company. For example, Salesforce as a customer relationship management system holds the organization’s “single source of truth” for customers and revenue—deal progress, discount approvals, commission calculations, revenue forecasts, all are core data that require deterministic answers.

SAP is even more secure in its status. As the system of record for enterprise finance, SAP CEO Christian Klein emphasized, during the January earnings call, that advanced generative AI models cannot handle the key business data and workflows on which companies depend for survival.

Barclays believes SAP is stickier than Salesforce due to the irreplaceability of financial truth. Workday holds a similar role in HR and payroll.

AI will not only fail to replace these systems, but will actually increase their importance. AI agents will create even more data touchpoints, making the complexity these systems must handle even greater. The Barclays research note says, “This means the importance of these systems increases, rather than their value going to zero—contrary to market opinion.”

Data Tools and AI Compute Sector Also Misjudged by Market

Besides system of record companies, Barclays reports two other categories of investment opportunities misjudged by the market.

Second are beneficiaries of AI agents. AI brings increased demand for code and data, with tools like JFrog (FROG) for managing software artifact versions and security, and data vendors like Snowflake (SNOW) and MongoDB (MDB), likely seeing increased usage as AI expands.

Third are AI compute providers. This is where the market logic reaches its greatest contradiction. If AI is powerful enough to disrupt the entire software industry, compute demand should logically surge—and yet companies like Oracle and CoreWeave have suffered heavy losses in the selloff. “There’s clearly a problem here that needs deeper study. The market’s sentiment is excessively pessimistic,” Barclays analysts write.

Where the Selloff Was Justified: Squeezed Profits at the Application Layer

There is some justification for the market’s panic. SaaS companies built on top of database infrastructure have long struggled at the application layer: clunky interfaces, lack of intuitiveness, inflated pricing, and sometimes even security vulnerabilities. Worse, customers often get locked into poor-quality systems due to high switching costs.

Matt Stoller, research director for the American Economic Liberties Project, writes: “The US software industry model is shaped around monopoly, delivering poor quality and bad security at high prices.” He described a 2016 meeting with community bankers who lambasted their niche software vendors as “expensive” and “terrible.”

In 2024, Swedish fintech company Klarna stopped using Salesforce and Workday, opting instead for products from smaller SaaS companies like Deel and Neo4j, and used the AI coding tool Cursor to build a more modern application layer atop them.

This reveals the real threat path AI poses to SaaS: Clients don’t simply replace SaaS with AI tools, but use AI to build their own apps, squeezing out expensive interface layers while keeping the underlying data.

The Repricing of Software Stocks Will Continue

This market correction is necessary for the enterprise software application layer. SaaS companies have long enjoyed high valuation multiples because they control both infrastructure and interface. If Anthropic and OpenAI’s technologies can overlay the system of record, SaaS companies’ pricing power will begin to erode.

Barclays’ research note sums up: “This means the application layer of bloated enterprise software may see the end of easy profits.” But this is not the doomsday for the entire industry. The key is to distinguish between companies that rely on application-layer profits and those whose value is rooted in the irreplaceable system of record layer.

SAP’s statement during the January earnings call reflects the confidence of system of record vendors. Other SaaS executives are also pushing back against bearish views. But the market needs time to digest these technical nuances and discern true disruptive threats from exaggerated panic.

The current indiscriminate selloff suggests that investors in credit markets and elsewhere—who have limited prior understanding of the software sector—are making decisions based on the most extreme forecasts.

As understanding deepens regarding the boundaries of AI capability and SaaS business models, the market may reprice companies mistakenly classified as “AI victims.” But for those long reliant on extracting high fees from subpar application layers, valuation compression may have only just begun.

Risk Warning and DisclaimerThe market has risks; investments require caution. This article does not constitute personal investment advice, nor does it take into account any user’s specific investment objectives, financial situation, or needs. Users should consider whether any opinions, views or conclusions in this article suit their particular circumstances. Investment decisions based on this article are made at your own risk. ```