Cursor CEO explains how his AI coding assistant will compete with OpenAI and Anthropic

Why Cursor's CEO believes OpenAI, Anthropic competition won't crush his startup | TechCrunch

Anysphere, the company that makes AI coding assistant darling Cursor, isn't thinking about an IPO any time soon, its co-founder CEO Michael Truell said on stage Monday at Fortune's AI Brainstorming conference. After reaching $1 billion in annualized revenue in November, and raising $2.3 billion at a $29.3 billion valuation last month, Truell said his company is instead focused on building out more features. For instance, he noted that Cursor's home-grown LLMs were geared to support specific products. Cursor also confirmed the existance of those models in November when it touted, "Our in-house models now generate more code than almost any other LLMs in the world." His comments about the models came up when the founder was asked how he plans to compete with the LLM makers that he relies on when the major ones -- OpenAI, Anthropic -- have their own AI coding offerings. Truell likened their coding products to "a concept car" whereas his product is a production automobile. "It would be like taking an engine and a concept car around it instead of a whole end-to-end car that was manufactured," Truell said. "What we do is we take the best intelligence that the market has to offer from many different providers. And we also do our own product-specific models in places. We take that, we build it together and integrate it then also build the best tool and end UX for working with AI." Cursor's dependence on its competitors -- and its need to build its own LLMs -- has been a subject of speculation among VCs in Silicon Valley since earlier this year when OpenAI reportedly looked at Anysphere as an acquisition target. Anysphere turned the idea down. (This was around the same time that Windsurf's OpenAI deal also didn't materialize, with the founder eventually joining Google). The issue, investors told TechCrunch, was that AI coding editors were losing money thanks to high costs they paid to the model makers. In Cursor's case, instead of selling, it adjusted pricing to a usage model in July, directly passing along the API fees that model makers charge to its users. This change from an all-inclusive subscription fee (and the surprise big bills some customers faced) caused an uproar among some of its users. On Monday, when asked about the pricing kerfuffle, Truell said, "When we started Cursor, you would turn to Cursor for a quick JavaScript question and now you're turning to it to do hours of work for you. So the pricing model had to shift for us and others in the space. That means shifting more towards a consumption model," he said. Truell added that one of the tools the company is working on is cloud-computing-like cost-management tools, which lets enterprises monitor their total usage and keep tabs on the bills their engineers are running up. "We have a whole team internally dedicated to enterprise engineering and building things like spend controls and billing groups and visibility," he said. Additionally, he said Cursor is focused on two major areas for the next year. One is handling more complex agentic functions. "We want you to take end-to-end tasks, ones that are concise to specify but then are really hard to do, and have them entirely be done by Cursor. An example is a bug fix," Truell explained. He particularly wants Cursor to be able to fix the kinds of bugs that might be easy to describe but take "weeks of someone's time, thousands of times running the code" to handle. "We want Cursor to do that, end-to-end," he said. The other area he named, but didn't explain with much detail, was the idea of "thinking about teams as the atomic unit that we serve," he said. This must be in contrast to serving individual coders, and a hint to how well its enterprise business is going. In addition to cost-monitoring 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 is being used by some customers to analyze every pull-request, be it written by AI or human. (A pull-request is when a programmer submits code for review before it is merged into the main project). "So you'll see us start to help teams more as a whole," with more features like that, he promised. Meanwhile, big competitors are also all gearing up for the complex-task agentic world. Amazon just released a coding tool it promises can already run for days on end. Just this week, the AI power players, including Anthropic, OpenAI, Microsoft, AWS and many others, launched a new consortium under the Linux Foundation to develop open source agentic interoperability standards. They even contributed some of their key projects, like Anthropic's wildly popular Model Context Protocol (MCP). His plans for the year likely won't put Anysphere firmly ahead of Cursor's main model-maker competitors. They should, however, keep the company in the race.

Cursor CEO says no IPO planned as revenue hits $1B

At Fortune's AI Brainstorm conference, Anysphere CEO Michael Truell announced that the company behind the AI coding assistant Cursor has no plans for an initial public offering in the near term, prioritizing feature development after achieving $1 billion in annualized revenue in November and securing $2.3 billion in funding at a $29.3 billion valuation the previous month. Truell emphasized that Anysphere's current efforts center on enhancing Cursor's capabilities rather than preparing for public market entry. The company's rapid financial growth underscores its position in the AI coding sector, where annualized revenue represents the projected yearly income based on recent performance metrics. This milestone, reached in November, reflects the strong adoption of Cursor among developers seeking AI-assisted programming tools. The recent funding round, completed last month, involved investments from various venture capital firms, boosting the company's resources for ongoing innovation without the immediate pressures of going public. One key aspect of Cursor's development involves its proprietary large language models, designed specifically to support individual products within the platform. These home-grown models enable tailored functionalities that integrate seamlessly with the overall system. In a November blog post, Anysphere stated, "Our in-house models now generate more code than almost any other LLMs in the world." This claim highlights the efficiency and output capacity of these models compared to external alternatives, positioning Cursor as a leader in code generation volume. During the conference session, Truell addressed strategies for competing against major large language model providers like OpenAI and Anthropic, which offer their own AI coding products. He described these competitors' offerings as akin to experimental prototypes, contrasting them with Cursor's fully realized application. Truell said, "It would be like taking an engine and a concept car around it instead of a whole end-to-end car that was manufactured." He further elaborated, "What we do is we take the best intelligence that the market has to offer from many different providers. And we also do our own product-specific models in places. We take that, we build it together and integrate it, then also build the best tool and end UX for working with AI." This approach combines external AI intelligence with internal models to create a comprehensive user experience, including optimized tools and interface design for developer interactions. Anysphere's reliance on these competitors for certain components has fueled discussions in Silicon Valley's venture capital community since earlier this year. Reports indicated that OpenAI explored acquiring Anysphere as a potential target, but the company declined the overture. This decision occurred around the same period when another AI coding firm, Windsurf, saw its potential deal with OpenAI fall through, leading its founder to join Google. Such events illustrate the consolidation trends in the AI sector, where startups face choices between independence and integration with larger entities. Investors have raised concerns about the profitability of AI coding editors, according to TechCrunch, pointing to substantial losses incurred from high application programming interface costs charged by model providers. In response to these financial pressures, Anysphere transitioned Cursor's pricing structure in July from a flat subscription fee to a usage-based model. This shift directly passes API fees from model makers onto users, aiming to align costs with actual consumption. The change resulted in unexpected high bills for some customers, prompting widespread complaints and debates within the user community about transparency and affordability. Addressing the pricing controversy at the conference, Truell explained the rationale behind the adjustment. He stated, "When we started Cursor, you would turn to Cursor for a quick JavaScript question and now you're turning to it to do hours of work for you. So the pricing model had to shift for us and others in the space. That means shifting more towards a consumption model." This evolution reflects how user behavior has progressed from brief queries to extended, resource-intensive tasks, necessitating a model that scales with usage patterns across the industry. To mitigate enterprise concerns over these costs, Anysphere is developing advanced management tools reminiscent of cloud computing services. These include features for monitoring total usage, implementing spend controls, creating billing groups, and providing visibility dashboards. Truell noted, "We have a whole team internally dedicated to enterprise engineering and building things like spend controls and billing groups and visibility." This dedicated team focuses on equipping businesses with the means to track and regulate AI tool expenditures, ensuring better financial oversight for engineering teams deploying Cursor at scale. Looking ahead, Truell outlined two primary focus areas for Cursor over the next year. The first involves advancing agentic functions to manage more complex operations autonomously. He described the goal as, "We want you to take end-to-end tasks, ones that are concise to specify but then are really hard to do, and have them entirely be done by Cursor. An example is a bug fix." Specifically, Truell aims for Cursor to resolve bugs that are straightforward to identify yet demand extensive effort, such as "weeks of someone's time, thousands of times running the code" to address. He affirmed, "We want Cursor to do that, end-to-end." This capability would automate processes currently requiring prolonged manual intervention, streamlining software maintenance workflows. The second priority shifts emphasis toward serving teams rather than individual developers. Truell referred to this as "thinking about teams as the atomic unit that we serve," indicating a strategic pivot to collective usage models. This approach builds on the growing enterprise adoption of Cursor. Beyond code writing, the company plans to expand into additional phases of the software development life cycle. For instance, Cursor's code review product already assists some customers in examining every pull request, whether generated by AI or humans. A pull request represents a submission of code changes for peer evaluation prior to integration into the primary codebase. Truell promised, "So you'll see us start to help teams more as a whole," with expanded features supporting group dynamics and collaborative processes. In parallel, major industry players are advancing similar agentic technologies. Amazon has introduced a coding tool capable of operating continuously for days, demonstrating endurance in handling prolonged tasks. Separately, this week a consortium formed under the Linux Foundation, involving Anthropic, OpenAI, Microsoft, AWS, and additional participants, launched initiatives for open-source agentic interoperability standards. Contributions include projects such as Anthropic's Model Context Protocol, which facilitates standardized interactions among AI systems.