When OpenAI released its frontier AI models, the industry discourse quickly shifted to AI ‘eating’ startups, particularly SaaS companies. Despite these concerns, one of the startups to have thrived in an AI-first landscape was Cursor.
According to research from Sacra, the AI-enabled coding platform reached $100 million in annual recurring revenue (ARR) within 21 months of its inception. It took Cursor only 12 months to reach this figure from $1 million ARR, making it the fastest-growing SaaS company of all time. The report added that Cursor reached these figures with roughly 360,000 individual developers who paid $20 – $40 monthly.
Anysphere, its parent company, was founded in 2022 by MIT students and friends Michael Truell, Sualeg Asif, Arvid Lunnemark, and Aman Sanger. Their vision was to build an integrated development environment (IDE) that was “AI native”, and they launched Cursor in January 2023 on top of Microsoft’s Visual Studio Code.
Founders ‘Obsessed with the Problem of Better AI Coding’
Cursor began with the ambition to take over the competition. While Microsoft had launched a GitHub Copilot tool for AI-assisted coding in 2021, Cursor wanted to differentiate itself by integrating the latest available AI models, providing a superior user experience, and regularly implementing new features and capabilities.
In September 2023, the company raised an initial $8 million in funding from OpenAI’s startup fund. In August 2024, Anysphere raised $60 million funding in Series A from Andreessen Horowitz (a16z), Thrive Capital, and other investors.
“Our belief is that Cursor, distinctly among AI coding tools, has simply gotten it right,” said a16z in a blog post. “They [Cursor’s founders] are obsessed with the problem of better AI coding, and are laser-focused on building a great developer experience,” the VC firm added.
The company also recently raised $105 million in Series B funding led by Thrive Capital, a16z, and other existing investors.
A significant win for Anysphere was that its users turned into Cursor evangelists in no time. The product was predominantly aimed at individual developers, unlike traditional SaaS makers that prioritised enterprise sales. Cursor also offered a free tier to try out the platform, besides a $20/month Pro subscription and $40/month/user for Enterprise.
Besides individual users, AI experts, like former OpenAI researcher Andrej Karpathy, have been vocal about their love for Cursor.
Cursor was also clear on understanding all the problems with existing AI-enabled coding solutions, and the team, on multiple occasions, published a comprehensive list of problems it intended to solve.
Initially, Cursor functioned as an AI-powered autocomplete tool. However, its capabilities quickly expanded to include more autonomous features capable of handling complex coding problems with the launch of Composer, an AI agent mode. Besides, developers could chat with the code database, retrieve real-time information from the web, and use AI models to understand the code base.
Eventually, the platform got ahead of the competition.
Cursor also positions itself as an extension of the human engineer, but not to merely replace one. “Using a combination of AI and human ingenuity, they [the hybrid engineers] will out-smart and out-engineer the best pure-AI system,” the company claims. This strategy resonated with users, who preferred Cursor’s workflow over Devin’s (an end-to-end autonomous software development tool).
Comparing it to Devin, Will Brown, an AI researcher at Morgan Stanley, said on X, “I think I slightly prefer the ‘pair programming’ workflow of Cursor agent, which is way more hands-on. You’re reviewing the code in real-time, plus [you] can give suggestions more easily.”
Will Cursor Remain Undefeated?
This isn’t to say Cursor is free from competition. Microsoft’s GitHub Copilot is estimated to earn more than $300 million in revenue annually. Sacra also noted that GitHub Copilot has an “enormous advantage”, thanks to its tight-knit integration in Microsoft’s ecosystem, as well as its existing enterprise relationships and security compliances.
Other tools like Vercel’s v0, Bolt.new, and Coedium’s Windsurf competed directly against Cursor. Carl Rannaberg, a software engineer, stated in his review that Cursor “remains the go-to tool for day-to-day coding tasks” thanks to its familiar code editor environment. On the other hand, v0 excels in rapid interface prototyping, and Bolt.new “shines in full stack prototyping and quick project setups”.
However, Coedium’s Windsurf has emerged as the biggest threat to Cursor. Recently, several developers have been reportedly transitioning to Windsurf.
“Windsurf outperforms Cursor when working within an existing large-scale codebase. Cursor struggles to maintain context as projects grow,” said Corey Sutton, an engineer in a blog post comparing both platforms. He also said that Windsurf was better at managing API modifications, while Cursor struggles to maintain consistency in TypeScript-heavy environments.
At the end of 2024, Sacra estimates Codeium to have earned $12 million in revenue.
Besides standalone products, Cursor also faces competition from frontier model makers. Recently, Anthriopic released its coding agent, the Claude Code. A developer on Reddit compared the two, and said that the code quality of Claude Code “blows Cursor out of the water”.
As revealed by the founders in a podcast episode with Lex Fridman, Cursor faces significant challenges related to both infrastructure and model management. As usage grows, system components like caching and databases encounter unexpected issues.
Cursor processes code in the cloud, but it must also stay in sync with the local environment. “One of the technical challenges is always making sure that the local index, and the local code base state is the same as the state that is on the server,” said Sualeh.
However, Cursor has grown exponentially over the past few months. The founders also said that the ceiling is ‘incredibly’ high for innovation and that Cursor’s advantage comes not just from integrating newer models but also from the depth at which it’s integrated into the tool. “You don’t realise [how the AI models] are working for you in every facet of the product, as well as the really thoughtful UX with every single feature,” said Sanger.
