On Monday, February 3, 2026, something unprecedented happened. Software, financial, and investment companies lost $285 billion in market capitalization in a single day. Thomson Reuters lost $8.2 billion. LegalZoom fell 20%. India’s Nifty IT technology index recorded its worst month since October 2008 – worse than during the financial crisis.
A trader at investment bank Jefferies coined the term "SaaSpocalypse" for the phenomenon. He described the mood on the trading floor as "everyone wants out at any price."
The trigger was Anthropic, which released Claude Cowork plugins for legal, financial, and sales workflows. The market immediately reached a conclusion: why pay for ten software licenses when a single AI agent can handle the entire workflow?
Why now?
According to Simon Taylor of Fintech Brainfood, the stock sell-off itself is not the point of the story. The key question is: why now and not six months ago? What changed?
In 2011, Marc Andreessen wrote the famous essay "software is eating the world." Since then, hundreds of publicly traded SaaS (Software as a Service) companies have emerged – Salesforce, Adobe, Intuit, ServiceNow – with valuations ranging from $2 billion to $100 billion. The 2010–2020 decade was a golden era: margins above 75%, customer retention above 100%, no physical inventory. The per-user licensing model had zero marginal costs. Each additional user added no extra costs. Margins therefore remained incredibly high (80%). This miracle cure is now facing an existential threat.
Four turning points that changed the trend
Four fundamental changes emerged at once over the course of a week:
- Models improved and began improving themselves
OpenAI released GPT-5.3-Codex. 25% faster. Half the token usage. The first model classified as "highly capable" in cybersecurity. But most importantly: it is the first model that helped create itself. OpenAI used an early version of Codex to fine-tune its own training runs. - Models escaped the chat window
The greatest obstacle to AI’s value was not intelligence – it was the interface. That obstacle is now gone. Claude now works directly in Excel, Notion, Linear, and Slack – not as a sidebar, but as an analyst. It can receive a request through Slack, run a cash flow report, update a task list, save it to Notion, and notify you when it is finished.
Claude Opus 4.6 is so powerful at financial analysis, including data processing and reading SEC filings, that shares of financial data providers fell after its launch, with FactSet dropping nearly 10%.
OpenAI launched Frontier – a complete platform for managing autonomous agents. Identity management, permissions, audit logs, and a "Business Context" layer that connects to data warehouses and CRM systems. Early adopters include companies such as HP, Uber, Oracle, State Farm, and Intuit. - Task duration exploded
Research by METR shows that autonomous task time horizons are doubling every four months. At 30 minutes: code snippets. At 5 hours: module refactoring. At several days: complete codebase audits. - Agents work in teams
Cursor created a system that orchestrates thousands of agents simultaneously. Its FastRender experiment had 30,000 commits, 2,000 concurrent agents, and more than one million lines of Rust code.
Enterprises are no longer experimenting
There used to be a reliable delay. Startups experimented, and enterprises followed 12–18 months later. That delay is gone.
Goldman Sachs embedded Anthropic engineers directly into its technology teams for six months. AI agents for trade accounting, reconciliation, and onboarding new clients. CEO David Solomon announced a multi-year AI reorganization aimed at "limiting headcount growth".
Norges Bank – a $1.7 trillion sovereign wealth fund with approximately 650 employees – reports an approximately 20% increase in productivity. 213,000 hours saved. The equivalent of more than 100 full-time positions.
Programming is key
According to SemiAnalysis, 4% of all public commits on GitHub are now created by Claude Code. Not "assisted" – created. An increase of 42,896× in 13 months.
Dylan Patel predicts 20% of daily commits by December. One model. One company.
This matters for the SaaSpocalypse because programming unlocks every other category of tool use. Code is how internal tools are created. Code is how APIs are connected. Code is how a SaaS license can be replaced with a custom workflow.
January 2026: Most layoffs since 2009
According to a report by Challenger, Gray & Christmas, more than 108,000 layoffs were announced in January – an increase of 205% from December and 118% from January 2025. The only worse January came during the Great Recession in 2009, when 241,749 layoffs were announced. In January alone, 7,624 layoffs were directly attributed to AI, representing 7% of all layoffs. Since 2023, when this metric began to be tracked, nearly 80,000 people have been laid off because of AI. "It is difficult to say how large an impact AI specifically has on layoffs," said Andy Challenger, the firm’s chief revenue officer. "The market seems to reward companies that mention it."
Dow Inc. accounted for 4,701 layoffs in January, the worst month for the chemical industry since 2016, explicitly citing AI-driven automation. Amazon led the technology sector with 16,000 job cuts.
SaaS is not dead, but it may not thrive either
As Jeremy Kahn of Fortune writes, SaaS is not finished. It is highly unlikely that Fortune 500 companies will want to build their own CRM or HR software, even if the process were mostly automated by AI software engineers. Nevertheless, there are reasons for concern. Anthropic, OpenAI, and Google could dominate the top layer of the agentic AI stack – building agent orchestration platforms that allow large companies to create, run, and manage complex workflows.
SaaS companies recognize the risk and are adapting. Salesforce is pioneering the so-called "Agentic Enterprise License Agreement" (AELA), which essentially offers customers a fixed price for unlimited access to Agentforce. ServiceNow is moving to consumption- and value-based pricing models. Microsoft has also introduced an element of consumption-based pricing alongside its usual per-user, per-month model.
The new hierarchy
The companies that survive will be those whose core components are accessible to agents. The new hierarchy looks like this:
- Top tier: Unique data and intellectual property. Bloomberg’s decades of financial data. Palantir’s ontology. Datasets that agents need to access, not replace.
- Middle tier: Excellent APIs and composability. Companies that make it easy for agents to use functionality. Stripe. Plaid. Twilio.
- Bottom tier: Commoditized user interfaces. Attractive dashboards built on data that anyone can access. A CRM with a drag-and-drop interface and nothing unique underneath.
Future impacts
Overall IT budgets are growing by 8%. AI budgets are growing by 100%. AI investment is a black hole – it captures every available dollar of IT spending and investor interest from other stocks because AI delivers a return on investment.
The models have learned to do your job. Zero marginal costs may be dead. Replaced by API calls and intelligent APIs, where margins are closer to 60% at best. This is the new normal that markets have just priced in.
Source: finance.yahoo.com
