Cursor 2.0 Introduces Its Own Coding Model and Multi-Agent Interface
Cursor 2.0 introduces its own AI coding model, Composer, and a new interface for parallel work with multiple agents.
(Image: Andrey_Popov/Shutterstock.com)
With version 2.0, the AI-powered development environment Cursor significantly expands its range of functions. The IDE relies on so-called agentic development, where multiple specialized AI agents collaboratively write, review, and test code. In addition to a new interface for parallel work with such agents, the company is also introducing Composer, its first proprietary coding model.
Proprietary Model for Agent-Based Programming
According to Cursor, Composer is approximately four times faster than models with comparable artificial intelligence and optimized for low latency in coding agents. The model apparently answers most requests in under 30 seconds. Test users have particularly highlighted the fast interaction and reliability in multi-stage tasks.
The model uses tools such as semantic codebase search to better understand larger codebases and work within them in a targeted manner—a strength that is said to be particularly beneficial when integrating different agents.
Focus on agents instead of Files
The new interface places agents at the center of the workflow. Developers are intended to work more result-oriented, while the agents handle the implementation. Those who still want to work directly in the code can switch to a more classic IDE view at any time.
Cursor 2.0 also allows multiple agents to run in parallel on separate Git worktrees or remote machines. This enables multiple models to work independently on the same task; the best result is then selected. According to Cursor, this process leads to a noticeable increase in quality for more complex tasks.
(Image:Â Cursor)
Adjustments to Review and Testing
With the update, the Cursor development team also addresses two bottlenecks from previous use of AI agents: code reviews and tests. Version 2.0 is intended to optimize the review of changes through new comparison views and an integrated browser tool for testing the generated code. This should enable the AI to automatically validate and improve its results.
Videos by heise
Cursor 2.0 is available for download immediately via the download page. A complete list of all new features can be found in the changelog.
(mdo)