The cost of writing software is collapsing, and most SaaS businesses are built on the assumption that it cannot. The market has started to notice.
In the days after Anthropic shipped Claude Cowork plugins and Claude Opus 4.6, nearly a trillion dollars in software market cap disappeared. Salesforce fell 26% year-to-date, the second-worst stock in the Dow. Intuit, Adobe, Workday, Autodesk, and ServiceNow all dropped. LegalZoom fell 20%. The S&P 500 Software & Services Index fell 20% from its October peak.
Analysts were raising earnings estimates while the sell-off happened. The near-term numbers looked fine. The market is not pricing in a bad quarter; it is pricing in the possibility that these companies do not exist in five years.
Why This Is Structural
Normal bad news expires. Missed earnings recover. Regulatory risk gets priced in.
This sell-off concerns the SaaS model itself. That model rests on one assumption: building software is hard and expensive, so customers pay recurring fees for pre-built tools. AI removes the assumption. One person can now produce in days what used to take a team six months. In some cases, agents do the work directly and the tool the human used to operate is no longer needed. Fewer humans in the loop means seat-based pricing fails.
Newspapers Already Showed Us
Old newspapers had expensive distribution. Printing presses, physical delivery, and geographic monopolies kept margins high and competition local. A paper in Seoul did not compete with the New York Times.
The internet dropped distribution cost to zero. Free distribution reached every publisher equally, along with bloggers and anyone else with something to say. Geographic monopolies vanished. Advertising moved to platforms. Most papers died.
Old software had the same structure. Coding was expensive, engineers were scarce, and switching costs locked customers in. The pattern was simple: find a business function, write an app, hire a sales team, IPO.
AI is doing to coding what the internet did to distribution, and the paradox is the same. Cheap code helps any individual company in isolation, but every competitor gets the same advantage, and products commoditize fast.
What Survives
Not every newspaper died. The New York Times and the Wall Street Journal survived by offering something the internet could not replicate. Software companies face the same test.
The dominant SaaS model has been the system of record: a tool that stores the data, charges per seat, and locks in the workflow. Salesforce does not close deals. Workday does not run payroll. They hold information while humans do the work. That model worked when building an alternative was expensive. It no longer is.
Value is shifting from systems that record work to systems that do work. A task that costs $200 an hour for a specialist can be done by an agent for pennies. A company that captures even a fraction of that spread has a durable business. One that still charges seat licenses for database access does not.
The common pushback is that today's agentic AI looks rough. Disruptive technologies enter through gaps that incumbents do not serve well. What matters is whether the technology is cheap enough and meets real demand. When both conditions hold, polish comes later. Dismissing agentic AI for looking scrappy now repeats the mistake newspapers made about blogs in 2004.
References
- Ben Thompson, "Microsoft and Software Survival," Stratechery, February 3, 2026.
- Ben Thompson, "The AI Unbundling," Stratechery, September 15, 2022.
- Ben Thompson, "Economic Power in the Age of Abundance," Stratechery, 2014.
- Anthropic, Claude Cowork product page, 2026.
- Anthropic, Claude Opus 4.6 announcement, February 5, 2026.