The Software Crash is Not Panic: It’s a Repricing of Models That Sold Hours Disguised as Subscriptions
On February 23, 2026, the software sector experienced a day that changed conversations in investment committees and boardrooms. The iShares Expanded Tech-Software Sector ETF (IGV) fell 4.75%, closing at its lowest level since November 28, 2023; it was its worst session since February 5, 2026, when it dropped 4.97%. In just one day, according to Dow Jones Market Data cited by CNBC, the decline erased approximately $223.75 billion from the market capitalization of the ETF’s components.
What was notable was not just the percentage drop but the composition of the hit. Names from the S&P 500 linked to software led the declines: IBM, Datadog, CrowdStrike, and Zebra, all with drops exceeding 9% during the session, according to the same report. The fallout was not confined to “pure software”: Thomson Reuters had accumulated a 28% drop year-to-date by mid-February; Microsoft was down 13% recently; Adobe had fallen 19% in the past month, per the analysis referenced by CNBC.
The market sought immediate scapegoats and found them in two triggers. Firstly, a post from Citrini Research over the weekend of February 21-22 suggested a hypothetical scenario—not a prediction—of productivity gains through AI leading to white-collar job cuts and a damaging cycle where each dollar saved on personnel funds the AI enabling the next round of layoffs. Secondly, a series of announcements from Anthropic including the launch of Claude Code Security for vulnerability scanning (pressuring cybersecurity), messages about COBOL modernization (sensitive for IBM), and an upcoming event on February 24 unveiling 2026 capabilities.
I don’t interpret this episode as “fear of AI.” I view it as something more structural: a value correction. The market stopped assuming every subscription was a defensive revenue machine and began to question what part of the software was truly a product and what part was human labor neatly packaged.
The “Arcane” Metric is the Symptom, Not the Disease
CNBC framed the movement around an unusual way of valuing stocks, an “arcane” metric that supposedly amplified the sell-off. That angle is useful for explaining price mechanics, but it is insufficient to explain the regime shift.
When a sector falls indiscriminately, as analysts quoted in the report described, the narrative of a “technical metric” is often the match, not the forest. The forest is this: for years, software was marketed to the market as the ideal asset due to its predictability. Recurring contracts, automatic renewals, attractive margins, compounded growth. That narrative became a mental shortcut: if it’s SaaS, then it’s defensive. If it’s defensive, then it deserves high multiples. If it deserves high multiples, then any correction is an opportunity.
AI disrupts this shortcut. Not because it “replaces everyone,” but because it destroys the boundary between tool and labor. If a new layer of automation reduces the marginal effort needed to perform tasks that previously justified licenses, seats, or modules, the investor is no longer just buying recurrence; they are buying the company’s ability to maintain its price when the cost of performing the task trends towards zero.
That’s why the reaction to headlines is so violent. UBS, as cited by CNBC, observed that enterprise software became extremely sensitive to news from Anthropic and OpenAI. Mizuho, also mentioned, captured the psychological state of capital demand: there is interest in buying on dips, but first, they need to see that stocks stop falling with each new AI announcement. That sentence is an X-ray of the issue: the industry lost narrative control over its value proposition.
When an industry loses narrative control, the market stops valuing “what it is” and starts valuing “what it could stop being.” That’s where rare metrics gain power: in environments of uncertainty, any instrument that promises to distinguish durability from fragility becomes a weapon.
The Real Pressure: SaaS That Relied on Human Friction
The post from Citrini Research acted as a catalyst because it articulated something many companies avoid admitting in their own P&L: much of the value captured by office, operations, and corporate function software relies on human friction. I’m not referring to “inefficiency” as a moral insult; I’m speaking of an economic reality: processes designed around people, controls, reviews, tickets, approvals, and layers.
As automation progresses, the first impact is not the disappearance of software, but rather the reconfiguration of the buyer. If a client reduces headcount in administrative, legal, or IT functions, they also decrease the number of seats, the volume of manual workflows, and their appetite for massive suites. Software that isn’t tied to a measurable outcome starts to be viewed as a tax.
Anthropic’s sequence illustrates this across verticals, not through theory.
- With Claude Code Security, the market punishes cybersecurity. Not necessarily because the product replaces a leader, but because it sends a message: certain functions that were monetized as platforms may now, in the eyes of the CFO, seem like “capabilities” bought as a service or consumed as an API.
- With the COBOL modernization narrative, a sensitive fiber is touched: legacy systems. Legacy always represented a rent due to talent scarcity and accumulated complexity. If a way emerges to reduce that scarcity, the value is repriced, even if the real path is longer and more complex than a headline may suggest.
- With legal automation-oriented plug-ins, the blow to Thomson Reuters—as cited in deVere's analysis in the report—shows a pattern: the market fears that “software” in certain industries was, in reality, a toll for access to processes, not a sustained technological superiority.
Meanwhile, Jefferies downgraded its rating on Workday, DocuSign, Monday.com, and Freshworks to hold. This is not a moral judgment on those companies; it’s a recognition of exposure: products that operate close to repeatable tasks, document flows, operational coordination, and management. If AI becomes a cross-sectional layer, that ground turns into a price and distribution war.
And here comes what many executives don’t want to hear: copying features is no longer a defensive strategy; it is a collapse accelerant. If everyone rushes to “put AI” in their roadmaps, they end up homogenizing the offering and teaching the customer to negotiate.
The New Axis of Differentiation: From Software as Tool to Software as Result Guarantee
As the market sells off en masse, lists of “durable models” also appeared. In the CNBC report, names like Intuit, Procore, Atlassian, and Salesforce were mentioned as better positioned due to AI adoption and perceived resilience. This contrast matters less for the brands and more for the type of promise conveyed.
Models that will sustain premium valuations are not those with the most features but those that can demonstrate three things with financial coolness.
First, the capacity to convert fixed costs into variable ones. In software, this isn’t a slogan; it’s architecture. If your value delivery depends on heavy consulting, endless implementations, or human support as a crutch, your operating margin is a sandcastle. Paradoxically, AI should allow you to deliver the same result with less internal friction. If that doesn’t happen, the market assumes your “AI” is just marketing.
Second, resilience to seat compression. The conversation around per-user licenses is a relic when the client begins purchasing results by process. If the value is tied to the number of people, the client can “optimize,” and your ARR becomes a misleading indicator. Conversely, if you monetize by result, reduced risk, or time saved, your model holds even with smaller teams.
Third, control of the integration point. D.A. Davidson, cited by CNBC, suggests that Anthropic is focused on traffic through APIs more than dominating categories. That phrase reveals the battleground: whoever controls the point where work is decided—where data, permissions, compliance, and decisions connect—captures value. Those who merely provide “another app” end up on a shelf of plug-ins.
This also explains why some analysts downplayed the drama. Bernstein, quoted by CNBC, stated that writing cybersecurity software hasn’t been a bottleneck. JP Morgan, according to the report, characterized the sell-off as indiscriminate and viewed AI more as additive than substitutive. Both may be right in the technical detail yet live with the underlying thesis: while AI may not replace products today, it has already changed the standard of evidence that the market demands to value tomorrow.
The Winning Move Requires Elimination and Reduction, Not Adding Modules
The typical response from incumbents will be to expand suites, add copilots, accumulate dashboards, and promise that everything is now “intelligent.” That reaction is understandable, but it is precisely the type of over-service that creates non-customers.
In my experience, when a category panics over disruption, product teams tend to defend their identity with more complexity. The result is a value proposition that raises the cost of adoption just when the buyer wants the opposite: less friction, less dependency, less training, and fewer consultants.
The most powerful strategic move in enterprise software for this cycle is not “adding AI.” It’s redesigning the value curve through unpopular decisions.
- Eliminate components of the model that exist to justify price, not to deliver results: configuration layers that specialists only understand, redundant reports, and function catalogs that no one uses.
- Reduce reliance on professional services as a growth engine. Not because services are bad, but because the market is ceasing to pay for human accompaniment as a structural advantage.
- Increase the traceability of impact: metrics that connect usage with real savings, avoided risk, accelerated compliance, and shortened sales cycles. If the customer cannot internally justify the expenditure with evidence, the spending gets cut.
- Create an offering that resides where the customer feels economic pain: performance guarantees, pricing packages aligned with outcomes, and integration so straightforward that the client’s AI does not need “your suite” to execute the work.
This is the uncomfortable part for C-Level executives: many software companies are not being attacked by a competitor, but by the exposure of their own economics. The market is re-evaluating who was selling products and who was selling hours disguised.
Leading in 2026 Means Validating New Demand, Not Defending Old Demand
The next informational milestone, according to CNBC, was the Anthropic event on February 24, 2026, about 2026 capabilities. Such moments will continue to create short-term volatility, as the market is hypersensitive to any signals suggesting substitution.
However, the real game is played away from the screen. It is played in the field where demand is validated for a distinct proposal, not for a rehashed version of the previous one. Leadership teams that try to maintain multiples with the same functionality expansion playbook will train the buyer to demand more for less.
The real exit is strategic and brutally practical: cut what the customer does not value, redesign pricing around results, and demonstrate it with repeatable sales, not presentations. The leadership that matters does not burn capital fighting over crumbs in a saturated market; it eliminates the irrelevant to build its own demand, validated in the field with real purchase commitments.
