Cursor's $2B Revenue Can't Mask Its Existential Pivot

Cursor, the AI coding startup that helped define the modern editor-plus-agent boom, does not look like a company in immediate collapse. It looks like a company trying to outrun the next market it helped create. Fortune this week framed the tension in stark terms: Cursor says 64 percent of Fortune 500 companies use it on its enterprise page, while its homepage uses the looser claim that it is trusted by over half of the Fortune 500, and Fortune reported that the company crossed $2 billion in annualized revenue in February even as customers and investors were openly gaming out whether Anthropic Claude Code could displace it (Fortune; Bloomberg on revenue; Bloomberg on fundraising talks; Cursor enterprise page; Cursor homepage).

The more interesting detail is that Cursor, the company behind the editor, is already arguing this is no longer mainly an editor story. In a February essay, Michael Truell, Cursor chief executive, wrote that the company is no longer primarily about writing code and is now helping developers build the factory that creates their software through fleets of agents running in the cloud (Cursor blog). That is not the language of a startup defending autocomplete. It is the language of a company trying to escape its own category before the category gets commoditized.

Cursor's own product materials back that up. The company says more than one-third of the pull requests it merges internally are now created by autonomous cloud agents, and in a separate launch post said more than 30 percent of its merged PRs come from agents operating in cloud sandboxes that can build, test, and produce review artifacts like logs, screenshots, and video (Cursor blog; Cursor product post). Its documentation says those cloud agents run in isolated virtual machines, can be launched from the web, desktop, Slack, GitHub, Linear, or an API, support Model Context Protocol servers, and are billed with API-style usage pricing rather than as a flat editor subscription (Cursor docs). Cursor's homepage now pitches the same shift in plainer marketing language: agents work autonomously and run in parallel on their own computers (Cursor website).

That matters because the competitive threat from Anthropic is credible precisely where the market is moving. When Anthropic launched Claude 3.7 Sonnet in February 2025, it introduced Claude Code as a terminal-based coding agent that could search code, edit files, run tests, use command-line tools, and handle large engineering tasks from the shell (Anthropic launch post). Anthropic has since pushed Claude Code into Team and Enterprise plans with premium seats, spend controls, usage analytics, managed policy settings, and a Compliance API (Anthropic enterprise post). In other words, Anthropic is not just selling a better model and hoping developers figure it out. It is turning coding agents into a governed enterprise product.

That helps explain why the mood around Cursor changed so quickly. Fortune reported that some startups were decoupling from Cursor, that key talent including the company head of engineering had left, and that investors were taking Claude Code seriously as the central threat (Fortune). Earlier coverage and newsletter chatter pointed in the same direction, though with thinner visibility from this session: Forbes' Anna Tong had already flagged Cursor's strategic response to the pressure, and The Information reported a competitive shift among Notion engineers, but both were blocked here, so those specifics should be treated as attribution, not independently verified fact.

Still, the easy Cursor is dead line misses what Cursor itself is building toward. Its recent research and product writing points toward self-driving codebases, with agents that do not just suggest code inside an integrated development environment, but manage long-running software tasks across multiple surfaces and eventually through deployment workflows (Cursor research post; Cursor product post). The docs even note that Cloud Agents were previously called Background Agents, a small rename that feels like a tell: this is a company re-centering the product around orchestration, not around the editor pane (Cursor docs).

The caveat is that much of the strongest evidence for Cursor's strategic shift comes from Cursor itself. The company's Fortune 500 penetration claims are self-reported, and even its own public materials vary a bit in wording: the homepage says over half, while the enterprise page says 64 percent (Cursor website; Cursor enterprise page). The fundraising and revenue numbers still rest on attributed reporting from Fortune and Bloomberg, not documents we could independently inspect here (Fortune; Bloomberg on revenue; Bloomberg on fundraising talks). And some of the competitive evidence remains frustratingly anecdotal or paywalled. Even Anthropic's current Claude Code product page was messy to fetch cleanly in this session after redirecting to claude.com, which is its own reminder that the product layer in AI still changes faster than the documentation around it.

But the deeper story is already visible. Cursor helped teach developers to code with AI inside the editor. Now the market is asking a harsher question: who owns the control plane once software work spills out of the editor and into terminal agents, cloud sandboxes, Slack threads, GitHub comments, and metered autonomous runs? Cursor has enough revenue, adoption, and product ambition to stay in that fight. What it may not have anymore is the luxury of defining the fight on its own terms.