Key insights

• Even a $30 billion company is not safe: Cursor's core product premise collapsed the moment AI agents stopped needing a human collaborator. Valuation does not equal stability in this market.
• Cursor is now competing directly against its own suppliers. Building in-house models puts it on a collision course with Anthropic and OpenAI, the same companies it depends on today.
• Revenue growth can hide fragility. Cursor doubled ARR to $2 billion in three months, yet it runs at negative margins on consumer plans. Enterprise contracts with Meta and NVIDIA are the real lifeline.

In Brief

On January 5, employees at Cursor returned from the holiday break to an all-hands meeting titled "wartime." During the break, developers testing Anthropic's latest model had noticed something alarming: the AI could now write code, build features, and ship full products without a human reviewing every line. That made Cursor's core product look fragile. Chief executive Michael Truell had described Cursor as "Google Docs for programmers," a place where humans and AI work side by side. But if AI no longer needs a human partner, what is the editor for?

Leadership warned that turbulent months were ahead, projects would be cut, and priorities would shift. The new directive was clear: build the best coding model in the world.

A rocket ship that suddenly feels exposed

The urgency of the "wartime" meeting stands out because Cursor had been growing at an almost absurd pace. The company started 2025 with roughly $100 million in annualized revenue (ARR, the total a company would earn in a year based on its current rate). By November it had crossed $1 billion ARR, and a new funding round valued the company at $30 billion. All four co-founders, who met as friends at MIT and founded the company in 2022, became billionaires. Cursor ranked among the top 20 most valuable private companies in the world.

The company reached $100 million in revenue with just 20 employees and no dedicated sales team. It attracted major investors including Andreessen Horowitz (a16z) and Thrive Capital, and it had a close early relationship with Anthropic, which gave Cursor early access to its AI models.

Then the same Anthropic released Claude Code, a command-line tool (a program you run by typing text commands, not clicking buttons) that lets AI agents write and run code directly, without any editor in the middle. Claude Code hit $1 billion ARR in just six months and $2.5 billion ARR shortly after. The partner had become a rival.

The existential question

Jerry Murdock, co-founder of the venture capital firm Insight Partners, put it bluntly: "Their view is that Cursor is obsolete today. The numbers suggest otherwise."

Both parts of that sentence are true at the same time. According to a source familiar with Cursor's finances, ARR has now crossed $2 billion, doubling in just three months. Growth continued through February. But adoption among businesses is showing slight declines, and no one knows yet how much tools like Claude Code will cut into that trajectory.

The structural problem is deeper than any single competitor. Developers are moving away from writing code line by line and toward orchestrating agents: assigning tasks, reviewing outputs, and coordinating several AI processes running in parallel. That shift makes the traditional code editor, where you type and the AI helps, look like yesterday's tool.

"Delete the product"

Cursor's leadership has acknowledged this internally with a blunt internal directive: "delete the product." The pivot has three parts.

First, build in-house AI models. Cursor is investing in research to compete directly with Anthropic and OpenAI by training its own specialized models on proprietary data. Its so-called "composer" models, built on open-source foundations like DeepSeek and Qwen (an Alibaba AI model), are cheaper to run and gaining users on the platform.

Second, go agent-first. Cursor recently launched updates to its cloud agents feature, allowing multiple AI agents to work simultaneously in separate environments. The goal is to match what Claude Code and OpenAI's Codex (another agent-based coding tool) already offer.

Third, win enterprise. Consumer plans currently run at negative margins. Cursor subsidizes usage for individual subscribers and loses money on them. Business contracts with large companies, including Meta and NVIDIA, are profitable and more predictable. Expanding those deals is now the clearest path to financial stability.

Can Cursor move fast enough?

Cursor helped popularize vibe coding, the practice of describing what you want in plain English and letting AI generate the application. It showed the world that you did not need to know how to code to build software. That was a genuine cultural shift.

But vibe coding is now a commodity. Every major AI platform offers it. The question Cursor faces is whether a company that built its identity around a code editor can transform itself into a model company and an agent platform fast enough, while being squeezed from both sides by the very AI labs whose models power it.

The answer will define what the next chapter of AI-assisted software development looks like.

---

Sources and resources