The iShares Expanded Tech-Software ETF (IGV) has fallen more than 21% year-to-date, erasing an estimated $2 trillion in market capitalization from B2B software companies. Analysts at multiple firms are calling it the “SaaSpocalypse,” and the thesis is straightforward: AI agents don’t pay per seat, and neither do the human employees whose jobs they’re replacing.
Atlassian has become the poster child for the correction. The stock fell 36% in February alone, according to S&P Global Market Intelligence data cited by The Motley Fool. By late February, Seeking Alpha reported the stock was down more than 50% year-to-date. The company is now more than 80% below its pandemic-era peak.
The Earnings Paradox
Atlassian’s collapse is notable because the company actually beat expectations. Second-quarter revenue rose 23% to $1.59 billion, topping the consensus of $1.54 billion. Adjusted earnings per share hit $1.22, above the $1.14 estimate. None of it mattered.
The problem is structural, not cyclical. Atlassian posted a $47.7 million GAAP operating loss driven by $452.6 million in share-based compensation, more than a third of total revenue. The company repurchases shares to offset dilution, but the cash drain compounds the vulnerability at exactly the moment investors are questioning whether Jira boards will matter when AI agents manage projects autonomously.
Seat Compression Hits the Application Layer
MarketMinute’s analysis, published March 24, describes a phenomenon it calls “seat compression”: enterprises deploying AI agents are reducing total software license counts because a single agent can perform the work of multiple human seat holders. The publication reports that for every autonomous agent deployed, companies are cutting human seat requirements at a roughly 1:5 ratio.
Atlassian is not alone. Monday.com, Workday, Asana, and Adobe have all faced significant drawdowns, with some application-layer companies falling 40-50% from their 2025 peaks. The S&P 500 Software index has dropped 19%, with over 90% of its components posting negative returns year-to-date.
The Companies That Are Surviving
The sell-off is not uniform. Infrastructure companies have held up because AI agents still need data storage, processing, and observability, regardless of how many human seats they replace. Snowflake and Datadog have remained relatively resilient, and Palantir has weathered the storm by positioning itself as a platform for AI integration rather than a per-user tool.
The companies attempting to bridge the gap are the ones pivoting to outcome-based pricing. Salesforce and ServiceNow have begun charging for tasks completed rather than seats occupied, an implicit acknowledgment that the per-seat model is broken. Salesforce stumbled first with a rare revenue miss in late 2025, sparking what MarketMinute called a “Salesforce Contagion” across the sector.
Why This Matters Now
The timing is significant. Oracle shipped 22 agentic AI applications into Fusion Cloud ERP on March 24, explicitly declaring the copilot era over. NVIDIA launched its Agent Toolkit at GTC. Anthropic shipped Claude computer-use the same day. Every major platform vendor is racing to put autonomous agents into production workflows.
For every enterprise that deploys those agents successfully, the question becomes: how many Jira seats, Workday licenses, and Monday.com subscriptions do we actually still need? The market’s answer, at least as of March 2026, is fewer than anyone expected.