Two years ago, Cursor 1.0 launched as a VS Code fork designed to assist developers with AI-powered code autocompletion, likened to an airline co-pilot. Six months ago, Cursor 2.0 introduced an enhanced chat interface capable of controlling the terminal to build entire features, elevating its role to that of an airline captain. Recently, Cursor 3.0 was released with a radical shift in focus: it no longer centers on writing code directly but instead empowers users to act as air traffic controllers managing swarms of AI agents across multiple repositories, machines, and cloud environments simultaneously. Additionally, Cursor unveiled its new Composer 2 model, claimed to be an in-house trained AI more intelligent than Claude Opus 4.6, though these claims rely heavily on internal benchmarks.
The new Composer 2 model initially impressed with its high performance, speed, and cost-efficiency on the “Trust Me Bro” benchmarks, suggesting a significant leap forward in AI coding models. However, controversy arose when it was revealed that Composer 2 is actually based on Moonshot’s Kimmy K2 model, which itself has been accused of training on Claude’s outputs. Cursor apologized for the lack of transparency and released a detailed technical report explaining their reinforcement learning enhancements. Despite the drama, the model’s capabilities remain promising for the emerging zero-code future of programming.
To support this shift from manual coding to agent management, Cursor 3.0 was completely rewritten from scratch using Rust and TypeScript, moving away from its VS Code fork origins. While the traditional VS Code editor remains accessible, the new interface focuses on managing multiple AI agents that can autonomously write and maintain code. This environment integrates professional development tools like language servers, file explorers, remote SSH access, and the ability to run numerous agents across various platforms, allowing developers to largely delegate coding tasks to AI.
A practical demonstration showcased how Cursor 3.0 can accelerate development by managing a swarm of agents to build a complex project rapidly. The user can initiate planning, delegate tasks like creating landing pages or remote server work, and monitor agent progress through intuitive status indicators. The interface also includes features like a built-in browser to test the final product and a design mode where AI can fix UI issues on demand without manual coding. This approach drastically reduces development time, enabling the creation of sophisticated applications in minutes rather than weeks or months.
Despite its innovative features, Cursor 3.0 has faced criticism for resembling OpenAI codecs and for its departure from traditional coding workflows. Nonetheless, the tool represents a significant evolution in AI-assisted software development, emphasizing agent orchestration over direct coding. The video concludes by highlighting Blacksmith, a sponsor offering a faster, cost-effective alternative to GitHub runners, which complements the agent-driven development model by providing efficient CI/CD pipelines essential for managing large-scale AI-generated codebases.
