The Software Sell-Off Tests the Value of Subscription Models
On Monday, February 23, 2026, the software sector experienced one of those days that change conversations in investment committees and boardrooms. The iShares Expanded Tech-Software Sector ETF (IGV) fell 4.75% and closed at its lowest level since November 28, 2023; it was its worst session since February 5, 2026, when it declined 4.97%. According to Dow Jones Market Data, cited by MarketWatch that Monday, the ETF's components lost approximately $223.75 billion in market capitalization that day. This is not lost revenue or an equivalent reduction in the assets of the fund itself.
What stood out was not just the percentage. It was the composition of the decline. MarketWatch reported drops exceeding 9% in IBM, Datadog, CrowdStrike and Zebra during that session. Other figures belong to an earlier period: in its February 13 analysis, deVere put Thomson Reuters' year-to-date decline at 28% as of February 11 and Adobe's decline at 19% over the month preceding that report. It also mentioned a 13% decline in Microsoft without unambiguously defining its individual measurement window. These figures do not all describe the movement on the 23rd.
MarketWatch linked the unease that day to a Citrini Research scenario and several Anthropic announcements. On February 22, Citrini published a hypothetical exercise, not a prediction: automation would enable job cuts and economic pressure would encourage further AI investment, feeding additional cuts. The market report also covered Claude Code Security, aimed at detecting vulnerabilities, and messages about COBOL modernization. These are elements of the information environment; their coincidence does not demonstrate how much of the decline each one caused.
My reading of this episode goes beyond fear of AI. I propose examining it as a possible value correction: which part of a subscription pays for a durable product and which part remunerates packaged human work. This is a strategic hypothesis, not a demonstrated cause for every company whose shares fell or a denial of the element of panic.
A Metric Alone Does Not Explain the Business Model
The headline of CNBC's February 25 analysis focused on a valuation metric associated with software stock selling. That framing raises a relevant financial question, but by itself it cannot identify the mechanism behind the decline or demonstrate a regime change. The question here is different: what economically sustains the recurring business that the market values?
When a sector falls indiscriminately, as JPMorgan characterized the selling in a note reported by Business Insider and cited by deVere on February 13, it is important to distinguish that opinion from a universal explanation. My interpretation is that, for years, the narrative of predictable software became a mental shortcut: recurring contracts, automatic renewal, attractive margins and compound growth. Getting from there to the assumption that all SaaS is defensive, deserves high multiples and turns any correction into an opportunity requires a leap that needs evidence.
AI challenges that shortcut. Not because it replaces everyone, but because it can dismantle the boundary between tool and work. If a new automation layer reduces the marginal effort of tasks that once justified licenses, seats or modules, investors are no longer buying recurrence alone: they are assessing a company's ability to sustain its price as the cost of performing the task falls. I do not assume that this cost is zero or that integration and control disappear.
On February 23, MarketWatch reported Karl Keirstead of UBS commenting on enterprise software's sensitivity to news from Anthropic and OpenAI. Jordan Klein of Mizuho described interest in buying the dips, but conditional on stocks no longer retreating with each AI announcement. My interpretation of those opinions is a risk: losing narrative control of the value proposition when customers or investors no longer know which part will withstand automation.
When that happens, valuation can shift from what a company delivers today to what it might stop delivering tomorrow. In that uncertainty, a metric that promises to distinguish durability from fragility gains influence. This does not turn a particular metric into a proven cause of the movement or replace business analysis.
The Real Pressure: SaaS That Relied on Human Friction
The Citrini scenario invites us to examine a risk, not treat it as a realized outcome: how much value captured by office, operations and corporate-function software depends on human friction. I do not mean inefficiency as a moral insult, but processes organized around people, controls, reviews, requests, approvals and layers. The exposed proportion must be assessed in each business; the scenario does not measure it for the whole sector.
If a customer automates tasks and reduces administrative, legal or IT headcount, it may also need fewer per-user licenses and fewer manual processes. This is not inevitable: adoption can expand other uses. The commercial question is whether a suite continues to justify its price through measurable outcomes as the buyer's organization changes, rather than depending solely on its workforce size.
The announcements mentioned allow that question to be asked by activity, without confusing an announced capability with demonstrated substitution.
- With Claude Code Security, one commercial hypothesis is that buyers may begin comparing certain functions sold as a platform with capabilities purchased as a service or through an API. The stock market reaction does not prove that the product replaces cybersecurity leaders or performs all their functions.
- With COBOL modernization, the issue is legacy systems: part of their maintenance price may reflect scarce talent and accumulated complexity. Reducing that scarcity could change the service's valuation, although the real work of migration, validation and continuity is more complex than a headline.
- In legal automation, deVere linked Anthropic's developments to pressure on Thomson Reuters in its February 13 analysis. My reading is a risk of charging a toll for access to processes that buyers can handle another way, not proof that the entire offering lacks technological superiority.
MarketWatch also reported that Jefferies downgraded its recommendations on Workday, DocuSign, Monday.com and Freshworks to hold. My strategic reading of that selection concerns exposure in repeatable tasks, document workflows, coordination and administration. I do not attribute this entire interpretation to Jefferies or turn a stock recommendation into certainty about those companies' futures.
Here is what many executives do not want to hear: copying features is not enough as a defensive strategy. If everyone rushes to add AI to their product plans without changing the value proposition, they can homogenize their offerings and teach customers to negotiate. This is a competitive risk, not inevitable collapse.
The New Axis of Differentiation: From Software as a Tool to Software as an Outcome Guarantee
Alongside broad selling, there were differentiated preferences. MarketWatch reported Brent Thill of Jefferies favoring Intuit, Procore, Atlassian and Salesforce. These were one analyst's opinions at that time, not market consensus or investment recommendations from Sustainabl. The contrast matters here for the business-model durability it proposes assessing, not as a promise about those companies' performance.
From this perspective, defending a valuation premium requires more than offering additional features. I propose examining three capabilities through financial evidence; none alone guarantees that the market will pay that premium.
First, the ability to turn fixed costs into variable costs where it makes economic sense. If delivering value depends on intensive consulting, lengthy implementations or recurring human support, we need to measure how much of that work is necessary and how much can be redesigned. AI could reduce internal friction, but it also brings usage, oversight and error costs. Margins must be verified, not assumed simply because automation is added.
Second, resilience to compression in per-user licenses. If value is tied to the number of people and the customer reduces headcount, annual recurrence does not guarantee the same future demand. Charging for outcomes, reduced risk or time saved may better align incentives, but requires reliable attribution and control over delivery costs. It does not ensure that revenue holds up with smaller teams.
Third, control of the integration point. In an email cited by MarketWatch, Gil Luria of D.A. Davidson interpreted Anthropic as prioritizing API traffic over dominating application categories. This is his interpretation, not an official statement of Anthropic's objectives. For supplier strategy, it raises the question of where data, permissions, compliance and decisions connect, and what distinctive value an application retains at that point.
There were also reservations about the alarm. MarketWatch reported Peter Weed of Bernstein's view that writing code had not been cybersecurity's bottleneck. Separately, deVere cited JPMorgan's expectation that AI would be more complementary than substitutive in the near term. These are not assertions that substitution is impossible. They allow me to state my thesis precisely: even if it does not replace products today, AI may raise the standard of evidence for their future value.
The Winning Move Requires Elimination and Reduction, Not More Modules
An established company's typical response may be to expand its suite, add copilots, accumulate dashboards and promise that everything is now intelligent. That reaction is understandable, but it can also create excessive functionality that drives away people who do not want to adopt such a complex solution.
When a category fears disruption, defending product identity with greater complexity can raise adoption costs just when buyers want less friction, less dependence, less training and less consulting. That tension deserves a commercial test, not an automatic response of adding features.
The strategic proposal for this cycle is not simply to add AI. It is to redesign the value curve through uncomfortable decisions, validating which ones make sense for each customer.
- Eliminate parts of the model that exist to justify prices rather than deliver outcomes: configuration layers only specialists understand, redundant reports and catalogs of features nobody uses.
- Reduce dependence on professional services that do not add proportionate value. Not because services are bad, but because human assistance must justify its price rather than be taken for granted as a structural advantage.
- Increase impact traceability: metrics linking usage to real savings, avoided risk, faster compliance or shorter sales cycles. If customers cannot defend spending internally with evidence, the risk of cuts grows.
- Create an offering where customers experience the economic problem: verifiable performance commitments, outcome-aligned pricing and simple integration. The test is whether it makes work easier to execute without forcing customers to buy unnecessary complexity.
This is the uncomfortable part for senior leadership: competitive pressure can expose the economics of the company's own model, not just a rival's strength. The distinction between a scalable product and packaged hours is a management question that must be answered with delivery, margin and demand data; a falling share price does not settle it.
Leading in 2026 Means Validating New Demand, Not Defending Old Demand
Writing on February 23, MarketWatch identified Anthropic's event scheduled for Tuesday the 24th as the next information milestone. That expectation belonged to the day before: when this article was published on February 26, 2026, the date had already passed. No unverified event outcomes are asserted here; the strategic point is that new announcements can add volatility, not that they make it predictable.
But the decisive test takes place away from the screen: validating whether new demand exists for a different proposition, not a cosmetically altered version of the previous one. Trying to sustain multiples through feature expansion alone may end up training buyers to demand more for less.
The way forward I propose is practical: cut what customers do not value, redesign pricing around outcomes and demonstrate it through repeatable sales, not presentations. Leadership should not burn capital fighting over crumbs in a saturated market; it should eliminate what is irrelevant and validate new demand through real purchase commitments.
