Why Cursor’s CEO thinks OpenAI and human competition won’t crush his startup

Anysphere, the company that makes the AI ​​programming assistant Darling Cursor, is not considering an IPO anytime soon, its founding CEO Michael Troll said. He said On stage Monday at Fortune’s AI Brainstorming conference.

After arrival $1 billion in annual revenue In November it was lifted $2.3 billion at a valuation of $29.3 billion Last month, Trowell said his company was instead focusing on building more features.

For example, he noted that Cursor’s local LLM programs were geared toward supporting specific products. Cursor also confirmed the existence of these models in November when it announced that “our internal models now generate more code than almost any other MBA in the world.”

His comments about models came when the founder was asked about his plans to compete with the LLM makers he relies on when the bigger companies — OpenAI, Anthropic — have their own AI coding offerings.

Trowell likened their development products to a “concept car,” while his product is a production car.

“It would be like taking an engine and a concept car around it instead of a complete car that was built,” Troll said. “What we do is we take the best information that the market can offer from many different providers. We also design our own models in certain places. We take that, build it together and integrate it, and then we also build the best tool and finalize the user experience for working with AI.”

Cursor’s dependence on its competitors — and its need to build its own LLM programs — has been the subject of speculation among Silicon Valley venture capital firms since earlier this year when OpenAI has reportedly looked at Anysphere As an acquisition target. Anysphere rejected the idea. (This was around the same time Windsurf’s OpenAI deal also did not materializewith the founder eventually joining Google.)

The problem, investors told TechCrunch, is that it comes down to it AI coding editors were losing money Thanks to the high costs they paid to model makers. In Cursor’s case, instead of selling, it adjusted pricing based on its usage model in July, passing on the API fees charged by model makers directly to its users. This change caused extensive subscription fees (and the sudden large bills some customers faced). Uproar among some of its users.

On Monday, when asked about the disruption in pricing, Troell said: “When we started Cursor, you were turning to Cursor to get a quick JavaScript question, and now you’re turning to Cursor to do hours of work for you. So the pricing model has had to shift for us and others in the space. And that means shifting more toward a consumption model.”

Trowell added that one of the tools the company is working on is cost management tools similar to cloud computing, which allow companies to monitor their overall usage and keep track of the invoices paid by their engineers.

“We have an entire team dedicated internally to enterprise architecture and building things like spend control, billing groups, and visibility,” he said.

In addition, he said the index focuses on two key areas for the coming year. The first is to handle more complex proxy functions.

“We want you to take end-to-end tasks, the ones that are brief to define but then are very difficult to do, and have them all done by Cursor. One example of this is bug fixing,” Truell explained.

He especially wants Cursor to be able to fix the types of bugs that might be easy to describe but would take “weeks of someone’s time, and thousands of code runs” to deal with. “We want Cursor to do that, from start to finish,” he said.

Another area he mentioned, but didn’t explain in a great deal of detail, was the idea of ​​”thinking of teams as the atomic unit that we serve,” he said. This should be a contrast to the service of individual programmers, and a hint at how successful the organization’s business is.

In addition to the cost control features, Truell said he wants Cursor to handle more parts of the software development lifecycle outside of writing code. He pointed to Cursor’s code review product as an example, which he said some customers use to analyze every pull request, whether written by AI or human. (A pull request is when a programmer submits code for review before merging it into the main project.)

“So you’ll see us start helping teams in general,” with more features like that, he promised.

Meanwhile, major competitors are also preparing for the world of agents with complex missions. Amazon just released a programming tool that promises it can do just that Already running for several days in a row.

Just this week, AI companies are backing players, including Anthropic, OpenAI, Microsoft, AWS, Many otherslaunched a new consortium under the Linux Foundation to develop open source interoperability standards. They even contributed some of their own major projects, such as the very popular Anthropic Model Context Protocol (Multilateral consultative process).

Its plans for this year likely won’t put Anysphere ahead of Cursor’s main model maker competitors. However, they should keep the company in the race.