AI was the opportunity; now it's also the threat

For three years, AI was the stock market's golden child. Every earnings call featured some version of the same pitch: we're integrating AI, we're deploying AI, we're building the future with AI. The message to investors was simple. AI is the opportunity. Adopt it or get left behind.

In the past two weeks, that story has inverted completely.

Anthropic's launch of Claude Cowork and its agentic plugins triggered what traders at Jefferies are now calling the "SaaSpocalypse," an indiscriminate selloff that has erased more than $800 billion in market value from the software sector in six trading sessions. The S&P 500 Software Index dropped 13% in five trading sessions. The iShares Expanded Tech Software Sector ETF is down nearly 30% from its September high. Salesforce is down 29% year-to-date. ServiceNow has lost $115 billion in market cap since late January. Adobe, Microsoft, SAP, and Oracle have collectively shed hundreds of billions more. The IPO window has frozen shut. Private equity firms are hiring consultants to stress-test their portfolios for AI vulnerability. Hedge funds have piled into short positions.

The question for CEOs is no longer "How do we adopt AI?" It is "Is AI about to eat our business?"

The screwdriver vs. the newspaper

Two competing narratives are fighting for control of the conversation right now, and which one you believe will shape how you lead through the next several years.

The first comes from Jensen Huang, who called the panic "the most illogical thing in the world." His argument is that AI will use and enhance existing software, not replace it. Software is a screwdriver. AI is the hand that picks it up. You don't throw away the screwdriver because the hand got stronger. Arm Holdings CEO Rene Haas echoed the sentiment, calling the market reaction "micro-hysteria." Bank of America published a note arguing the selloff is built on two mutually exclusive premises: that AI spending will collapse because it's not delivering returns, and simultaneously, that AI will be so effective it renders all existing software obsolete. You can believe one of those things. You cannot logically believe both.

The second narrative is darker and more structural. Goldman Sachs analysts have compared the software sector's trajectory to that of newspapers in the early 2000s, when disruption began as a slow leak and then became a flood. Microsoft CEO Satya Nadella himself laid the groundwork for this argument over a year ago, saying that business applications would "all collapse in the agent era" because AI agents won't discriminate between backends. They'll just update multiple databases, and all the logic will live in the AI layer. When the CEO of the world's largest software company tells you that traditional business applications are structurally vulnerable, it is worth taking seriously.

The math behind this second narrative is straightforward and unforgiving. If 10 AI agents can do the work of 100 sales reps, you don't need 100 Salesforce seats anymore. You need 10. That is a 90% reduction in seat revenue for the same work output. The per-seat licensing model, which generated predictable, recurring revenue for a generation of software companies, suddenly appears to have a structural ceiling. And it is not just the pricing model. IT budgets are being actively redirected. CIO surveys show budget growth decelerating, with funds moving away from application software and toward the massive compute costs of AI infrastructure.

The self-destruct paradox

Here is the part that should keep every software CEO up at night: the companies trying hardest to survive are accelerating their own disruption. Salesforce is aggressively pushing its Agentforce platform. Adobe is embedding generative AI across its creative suite. ServiceNow is building AI into its workflow automation. But the more successful those AI features become at automating work, the fewer human users need the software, and the fewer seats those companies sell. It is a self-destruct paradox. Success in AI undermines the business model that pays the bills.

This is not theoretical. It is already showing up in the numbers. The narrative through 2024 and early 2025 was that AI would augment incumbents and make their products stickier. By late 2025, the reality on the ground was that the pace at which incumbents were adding AI lagged the pace at which native AI tools were beginning to replace them.

Jason Lemkin, whose SaaStr blog has been a barometer for the industry for over a decade, offered a characteristically blunt assessment. The crash, he argues, is not really about AI killing SaaS overnight. Nobody is vibe-coding a replacement for their Salesforce instance in Replit. Building a version one of any product is maybe two percent of the actual work of running enterprise software. The crash is the market finally pricing in a deceleration that has been building for years. Revenue growth was already slowing. Net retention rates were already stalling. AI just made the reckoning urgent. The era of easy SaaS growth is over.

How agents actually kill software

It is worth understanding the mechanics of this threat, because it is not the same story that has been told about AI for the past three years.

The old story was about copilots. AI sits beside a human, suggests a line of code, drafts an email, summarizes a meeting. The human stays in control. The software stays necessary. The seat stays billable. That story was comfortable for incumbents because it meant AI made their products better without threatening the underlying business model.

The new story is about agents. An AI agent does not assist a human using software. It uses the software itself, or it bypasses it entirely. It reads the database directly, executes the workflow, updates the records, and moves on to the next task. The human who used to sit in front of the CRM, project management tool, or legal review platform is now optional. And when the human becomes optional, so does the per-seat license that paid for their access.

This is what Satya Nadella was describing when he said business applications are "essentially CRUD databases with a bunch of business logic" and that in the agent era, all the logic moves to the AI layer. Agents will not discriminate between backends. They will update multiple databases simultaneously, and the value will live in the intelligence tier, not in any individual application's interface.

The implications cascade. Bessemer Venture Partners frames it bluntly: vertical AI is not competing for IT budgets. It is competing for labor budgets. Traditional vertical SaaS captures a fraction of Fortune 500 IT spend. Vertical AI taps directly into the labor line of a P&L. When an AI agent can do the work of a junior associate at a law firm, a first-year analyst at a bank, or a customer service rep at a SaaS company, the disruption is not to the software budget. It is to headcount. And when headcount shrinks, so does the total addressable market for every tool those workers used.

Consider the practical examples already emerging. In customer service, AI agents can handle ticket volumes that used to require a full team, cutting service-desk costs by 30% or more in early deployments. In sales, agents can handle prospecting, qualification, and follow-up sequences that previously justified dozens of seats on a CRM. In legal, AI can review contracts, flag risk, and generate redlines faster than a team of associates. Every one of those use cases represents not just a productivity gain but a seat reduction. The more capable the agent, the fewer humans in the loop, the fewer licenses sold.

When the model companies come for your market

And then there is the question nobody in SaaS wants to ask out loud: what happens when AI model companies stop being platforms and become products?

This past weekend, OpenAI hired Peter Steinberger, the creator of OpenClaw, the open-source AI agent that racked up 180,000 GitHub stars and became the fastest-growing consumer AI tool in history. Sam Altman said Steinberger will "drive the next generation of personal agents" and that the technology will "quickly become core to our product offerings." Steinberger himself predicts that agents like OpenClaw will eliminate 80% of current apps. His reasoning is simple: every app is just a slow API now. Why open a separate tool for food delivery, calendar management, or expense tracking when your agent already knows your context, your preferences, and your schedule?

This is the move that should alarm every software CEO. It is one thing when AI makes your product more efficient. It is another thing entirely when the companies building the AI models start building the products themselves. In January, both OpenAI and Anthropic announced HIPAA-compliant life sciences tools that compete directly with Veeva and Salesforce's healthcare offerings. These are not generic AI features. They are vertical products targeting specific markets that SaaS companies spent years and billions of dollars building out.

The pattern is familiar if you have been in tech long enough. Google started as a search engine, then became an email provider, a document suite, a cloud platform, a phone operating system. Amazon started as a bookstore, then became everything. When platform companies decide to move up the stack into applications, incumbents rarely survive the transition. The question is whether Anthropic, OpenAI, Google, and Meta will follow the same playbook. The early signals suggest they will. And unlike previous platform shifts, this one could move much faster because the marginal cost of building an AI-native application on top of your own model is close to zero.

What the market is actually pricing in

Bain & Company published an analysis that cuts through some of the noise. Enterprise customers, they found, actually prefer to buy AI-enabled solutions from their incumbent vendors. They trust them, they know the security posture, they believe they will be around long-term. But most incumbents have not yet delivered compelling AI offerings. And even where they can, many have not figured out how they would get paid for them.

This is the real crisis. The market is not pricing in the death of software. It is pricing in the possibility that the transition from selling seats to selling outcomes will take longer, cost more, and destroy more value than anyone wants to admit. The companies that figure it out will lead the next cycle. The companies that do not will join a long list of industries that saw disruption coming and still could not move fast enough.

Some categories look better positioned than others. Usage-based companies like Datadog and MongoDB function as something like tax collectors on AI activity. As agents proliferate, they generate more API calls, more database queries, more compute cycles. Companies that charge by usage rather than by seat effectively benefit when AI does more work. Mission-critical infrastructure providers like Oracle and ServiceNow, whose products are deeply embedded in customer workflows, have what one analyst called a sustained "right to earn." Their data moats and operational entrenchment make them more likely to coexist with AI than be replaced by it.

Then there is an emerging category that does not yet fully exist: AI orchestration. The software that manages, secures, audits, and governs the AI agents themselves. As businesses shift from humans using software to AI agents using software on behalf of humans, the companies that provide the security and governance layer for those agents will become the new must-have line item in the IT budget.

The leadership question

I live in Utah, one of the world's SaaS capitals. Silicon Slopes is home to Qualtrics, Pluralsight, Domo, Podium, and hundreds of other software companies built on the per-seat subscription model that the market is now questioning. This is not an abstract financial story here. It is a daily conversation at every dinner, every board meeting, every coffee between founders and VCs along the Wasatch Front. People I know, people I work with, are watching their companies lose a third of their market value in weeks. At the Silicon Slopes Summit earlier this month, the subtext of nearly every panel was the same: what do we do now?

What I have learned from a decade in this community is that moments like this reveal character. Not strategy. Character.

The instinct when your stock drops 25% in a month is to do one of two things: panic and slash everything in sight, or go on CNBC and insist everything is fine. Neither response works. The first destroys your capacity to adapt. The second destroys your credibility when the next quarter comes in soft.

The leaders who survive disruption are not the ones who predict the future correctly. They are the ones who build organizations capable of adapting to whatever future arrives. That means doing a few things that are uncomfortable but necessary.

First, assess the threat honestly. Not every company is equally exposed. If your revenue comes primarily from per-seat licensing of features that AI can plausibly automate, you are in the direct path. If your product sits at the infrastructure or data layer, you have more time. The worst thing you can do is pretend the distinction does not matter.

Second, stop confusing AI adoption with AI strategy. Every company in America has adopted AI to some degree. Very few have actually rethought their business model in response to it. There is a difference between adding a chatbot to your product and reconceiving how you create and capture value in a world where autonomous agents do an increasing share of the work.

Third, be honest with your board and your team about what you do not know. The Fortune analysis of this selloff quoted one investor who captured the real fear: "What's scarier than uncertainty?" The temptation in uncertain moments is to project false confidence. The better move is to name the uncertainty, define the scenarios, and make sure your organization can execute under any of them.

This is personal leadership in its most essential form. Not having the answers, but having the honesty to say you are working on finding them, and the discipline to build an organization that can turn on a dime when the answers become clear.

The mantra has changed

For the past decade, the technology industry operated under a single mantra: software is eating the world. Build software, sell subscriptions, grow recurring revenue, and watch the multiples expand. That playbook produced some of the most valuable companies in history.

In 2026, the mantra has changed. AI is eating the software budget. The question is whether you are the one holding the fork or sitting on the plate.

The market will sort itself out. Some of this selloff is rational repricing. Some of it is panic. The software sector is almost certainly oversold in the short term. But the structural questions underneath the volatility are real, and they are not going away. The companies that thrive will be led by people who can hold two truths simultaneously: that their current business still works today and that it may not work three years from now. The ability to live in that tension, to plan for the present while building for a fundamentally different future, is what separates leaders from administrators.

The tool has become the competitor. Act accordingly.