It’s turning into another bad week for software stocks. Last year the sector underperformed other parts of the tech landscape like semiconductors, and that has continued in 2026. On Tuesday, enterprise names like Salesforce, Workday, and Atlassian each fell at least 4%.
The slide continued on Wednesday, with the iShares Expanded Tech-Software Sector ETF down 1.4%. The tech-heavy Nasdaq Composite was down 1%.
“Software is sucking wind,” Jefferies software analyst Brent Thill told Barron’s. “People are just like, ‘I can’t take the pain anymore.’”
The brewing narrative is that artificial-intelligence coding tools will drain the moats that these companies have worked so hard to fill over the past decades. AI coders are already providing advanced capabilities, and they will only improve with time.
Investors are betting that at some point companies will be able to make their own customized versions of these applications at a lower cost, bypassing software subscriptions.
The immediate catalyst for Tuesday’s selloff was the release of a new AI-agent called Cowork from AI start-up Anthropic. Agents are software that can automate a complex series of tasks from a simple prompt. Anthropic has already had success with its coding agent, Claude Code, and hopes to bring that sort of tool to a more general audience. Cowork’s launch demos were rudimentary, but it isn’t hard to see where these things could go, replacing software functions and making some of them obsolete.
If AI chat becomes the primary interface for using software as many people imagine will happen, developers may need to rethink what software looks like and how it is delivered in that new world. It could look more like Cowork than today’s graphical interfaces.
Not every company will survive that transition. The current environment mirrors the shift to cloud software over two decades ago.
In the 1990s, Siebel Systems dominated customer relations management software, known as CRM. Its business model was to sell packaged software to large enterprise customers, who would host it on their own servers. There was a lot of expense up front, including a big payment to Siebel, large internal IT costs for deployment, and often an army of outside consultants. Implementation could take more than a year. The model relied on big upgrade cycles, which heaped on more expense.
Then along came Salesforce with an entirely different model—software delivered inside a browser. There was much reduced initial cost to customers, and a much simpler subscription-based sales model. Since the software was all hosted by Salesforce, updates were continuous, most of which went unnoticed by users.
Initially, Salesforce went after small and medium-size customers, and Siebel dismissed it as a toy. But eventually Salesforce’s cloud model won over the large Siebel customers with a product that combined ease of implementation, ease of use, and predictable expenses. In a marketing campaign, Salesforce called it “the end of software.” The rise of AI has some investors singing the same tune.