Cursor has announced a new version of its development environment in which AI plays a more central role. With Cursor 3, the company is introducing a workspace designed for collaboration with so-called agents—systems capable of independently writing, modifying, and testing code.
The update marks a further shift in software development, where developers do less manual programming and increasingly rely on the work of AI systems. According to Cursor’s creators, the industry is in a phase where multiple agents work in parallel on improvements to codebases. At the same time, developers still spend a lot of time coordinating those systems and managing various tools and interfaces.
Cursor, developed by Anysphere, has reportedly raised over $3 billion in funding from companies including Nvidia and Google. The platform combines its own AI models with external models such as Claude to automate programming tasks, from building functionality to detecting errors.
Cursor 3 aims to reduce fragmented workflows by providing a unified working environment. The interface has been redesigned and centers around agents. Multiple repositories and workspaces are consolidated into a single view, allowing developers and AI systems to work on different projects simultaneously.
Natural language becomes the input for code generation
A key feature is the introduction of a chatbot-like interface. Developers describe in natural language what functionality they want to build and choose which model will perform the task. The system then generates code and visual output such as screenshots and demonstrations. According to SiliconANGLE, Cursor reports that this allows for faster verification of whether the code works as expected.
The architecture uses multiple agents that perform tasks simultaneously. Some run in the cloud and can process tasks in parallel with additional computing power. Other agents run locally, allowing developers to inspect, modify, and test code immediately.
Cursor 3 makes it possible to switch between these environments. Code can be generated in the cloud and modified locally, or conversely, continue running in the cloud when a user goes offline. According to the company, its proprietary Composer 2 model is optimized for these types of workflows.
Another new feature allows user interfaces to be customized using natural language. In Design Mode, developers select elements and describe changes, after which agents automatically implement them.
In addition, Cursor displays an overview of steps for each task, including explanations, error messages, and visual feedback. Developers can use this to make adjustments with additional instructions.
Users can send commands to multiple AI models simultaneously and choose the best output. The code management workflow has also been updated, with a revamped view of differences and faster review of changes.
Also read: Cursor’s advanced AI coding model has Chinese origins