The offices of Anysphere, the company behind the Cursor code editor, occupy two floors of a building in San Francisco’s South of Market district that, in the manner of all consequential technology companies, manages to be simultaneously nondescript and charged with the particular energy of people who believe they are building something that will change how a fundamental category of human work is performed. They are not wrong. Cursor, the AI-native code editor that Anysphere launched in early 2024, has crossed one million daily active users and signed fifty thousand business customers, including Stripe, Figma, Shopify, and a substantial fraction of the companies whose software infrastructure constitutes the invisible architecture of modern commerce. The product’s growth curve does not resemble the gradual adoption pattern of a developer tool; it resembles the vertical ascent of a platform that has found the precise intersection of technical capability and market need.

Composer 2, the next generation of Cursor’s AI-assisted coding interface, is currently in internal testing and represents what the company’s founders describe as a transition from AI-assisted coding to AI-native software development — a distinction that sounds like marketing language until one observes the system in operation. Where the current Composer allows a developer to describe a change in natural language and receive a code modification that they review and accept or reject, Composer 2 operates across entire codebases, maintaining context about project architecture, dependency relationships, testing patterns, and organizational coding conventions in a manner that approaches, without yet reaching, the contextual understanding of a senior engineer who has worked on a project for years. The system does not merely autocomplete; it architects.

The competitive landscape that Cursor navigates has become, in the past twelve months, one of the most fiercely contested territories in the technology industry. GitHub Copilot, backed by the combined resources of Microsoft and OpenAI, remains the market’s incumbent and claims more than two million paying subscribers, though industry analysts note that its growth rate has decelerated as Cursor and other competitors have eroded its technical advantage. Amazon’s CodeWhisperer has been integrated into the AWS development ecosystem with the patient persistence that characterizes Amazon’s approach to adjacent markets. Windsurf, born from the Codeium project, has attracted a devoted following among developers who prefer its approach to multi-file editing. And Devin, the much-publicized autonomous coding agent from Cognition Labs, has moved from demonstration to production deployment at a handful of large enterprises, though its fully autonomous approach has generated as much skepticism as enthusiasm among working developers who remain protective of their agency in the coding process.

The stakes of this competition extend far beyond the immediate market for developer tools, which is itself substantial — estimated at approximately eighteen billion dollars annually and growing at a rate that makes traditional software market projections look quaint. Whoever owns the developer workflow — the environment in which code is written, tested, reviewed, and deployed — gains three assets of compounding strategic value. The first is distribution: a code editor that becomes the default environment for millions of developers is a platform through which adjacent products and services can be delivered with negligible customer acquisition cost. The second is data: every keystroke, every accepted suggestion, every rejected recommendation generates training signal that improves the underlying models, creating a flywheel effect that advantages early leaders. The third is recurring revenue of exceptional durability, because developers do not change their primary tools casually, and an AI coding assistant that has learned the patterns and preferences of a specific team becomes more valuable with each passing month.

Stripe’s adoption of Cursor across its engineering organization illustrates the dynamics at work. The payments company, which employs approximately three thousand engineers and processes hundreds of billions of dollars in annual transaction volume, began with a small pilot in its infrastructure team and expanded to organization-wide deployment within four months — a pace of enterprise adoption that Stripe’s own internal technology procurement team described as unprecedented. The compelling factor, according to engineering leadership, was not the raw speed improvement (though Stripe estimates a twenty to thirty percent increase in code output per engineer) but the reduction in the cognitive overhead of navigating a codebase that has grown, over fourteen years of development, to a scale that exceeds any single engineer’s capacity to hold in memory. Cursor’s ability to maintain project-wide context and surface relevant code patterns transformed, in Stripe’s assessment, not the speed of coding but the accessibility of the codebase itself.

The business model that Anysphere has constructed around Cursor is notable for its simplicity in an industry that often confuses complexity with sophistication. The product is offered at twenty dollars per month for individual developers and forty dollars per month for business accounts, with no free tier — a pricing decision that the company’s CEO, Michael Truell, has defended on the grounds that a product worth using is worth paying for, and that a free tier attracts users whose usage patterns generate cost without generating insight into the needs of serious developers. This approach has produced annual recurring revenue that industry sources estimate at approximately three hundred million dollars, a figure that, if accurate, would make Anysphere one of the fastest companies in the history of enterprise software to reach that milestone.

The technical architecture underlying Cursor’s capabilities is itself a subject of intense industry interest and speculation. The company employs a multi-model approach, routing different types of coding tasks to different AI models based on the complexity and context requirements of each task — a strategy that allows Cursor to balance capability against latency and cost in ways that competitors using a single model cannot easily replicate. Anysphere has trained proprietary models for specific coding tasks, including code completion and codebase search, while relying on frontier models from Anthropic and OpenAI for the more complex reasoning tasks that Composer handles. This architectural flexibility has proven to be a significant competitive advantage, allowing Cursor to adopt new model capabilities within days of their release rather than the weeks or months required by competitors whose products are more tightly coupled to a single model provider.

The broader implications of the AI coding revolution — and it is, by any reasonable assessment, a revolution — extend to questions that the technology industry has not yet fully confronted. If AI tools can increase the output of each developer by a factor of two to five, which is the range that current evidence suggests, the economics of software development shift in ways that favor smaller teams with greater ambition. The ten-person startup that could previously build a product of moderate complexity can now build one of considerable complexity. The thousand-person engineering organization may discover that its optimal size is five hundred, or three hundred, or some number that its human resources department has not yet been asked to model. The democratization of coding capability is real and accelerating, and its consequences for employment, for the structure of technology companies, and for the distribution of the economic value created by software have only begun to be understood.

What is clear, as Anysphere prepares to release Composer 2 and its competitors sharpen their own offerings, is that the question of who writes the world’s software — and with whose assistance — is being answered in real time, in a competitive arena where the pace of capability improvement is measured in weeks rather than years, and where the difference between the leading product and its nearest competitor is measured not in features but in the depth of contextual understanding that separates a tool from a collaborator.