Cursor 3.0: AI Agents Take Over Coding

Cursor 3.0 Ditches Code Writing for AI Agents

The popular AI-powered coding tool Cursor has released its latest version, Cursor 3.0, and it’s a radical departure from its previous iterations. Instead of focusing on helping developers write code, Cursor 3.0 shifts the focus to managing fleets of AI agents. This new direction aims to position developers as air traffic controllers, overseeing multiple AI agents working across different code repositories, machines, and even the cloud simultaneously.

Introducing Composer 2: A Smarter AI Model?

A key component of Cursor 3.0 is its new in-house AI model, Composer 2. When it first arrived, Cursor claimed Composer 2 surpassed top-tier models like Claude Opus 4.6 in performance, citing internal benchmarks. These early claims suggested the model was faster, cheaper, and more intelligent. However, it was later revealed that Composer 2 is actually based on Moonshot’s Kimmy K2 model. This lack of transparency led to an apology from Cursor, but the underlying model’s capabilities remain significant for AI-assisted development.

From VS Code Fork to Standalone Powerhouse

Cursor’s journey began two years ago as a fork of Visual Studio Code (VS Code), designed to add AI code completion. Six months ago, Cursor 2.0 introduced an enhanced chat feature that could manage the terminal to build entire features. Now, with version 3.0, Cursor has been completely rewritten from scratch using Rust. While the familiar VS Code editor is still accessible, the new interface is built around managing AI agents, allowing developers to abstract away much of the direct coding process.

Managing Swarms of AI Agents

Cursor 3.0’s core innovation is its ability to orchestrate multiple AI agents. Developers can assign tasks to these agents and have them work in parallel. For example, one agent could work on a project’s main architecture while another simultaneously develops a separate marketing landing page. Developers can also instruct agents to perform remote work on cloud servers or switch between entirely different projects. This multi-agent approach aims to dramatically speed up development cycles.

Real-World Application: Building a Prototype Quickly

To demonstrate the power of Cursor 3.0, the team tackled the creation of a prototype for a fictional company called “Horse Tinder.” Manually coding this project would take months, and even AI-assisted coding might take weeks. However, using a swarm of AI agents within Cursor 3.0, the team aimed to complete the project by the end of the video. They initiated a plan mode to outline the architecture, while other agents simultaneously worked on different aspects like a landing page and remote server tasks.

Seamless Workflow and Agent Monitoring

The new interface allows developers to monitor the status of each AI agent. A yellow dot indicates that the agent requires human input, often to approve potentially unsafe commands. A blue dot signifies that the agent has completed its task and is ready for review. Within minutes, the project generated thousands of lines of code. The tool provides access to git history, a terminal, and a file explorer, all within the Cursor environment.

Interactive Product Testing and Design Iteration

A standout feature is the built-in browser within Cursor 3.0. This allows developers to navigate directly to the running application and test the product in real-time. After reviewing the initial design, which was deemed “horshit” by the presenter, they entered design mode. Developers can then highlight specific elements, like buttons with unreadable text, and instruct the AI to fix the CSS. This allows for rapid iteration and refinement without needing to manually adjust code.

Why This Matters

Cursor 3.0 represents a significant shift in how software development might be approached. By empowering developers to manage AI agents rather than write every line of code, the tool promises to accelerate project timelines dramatically. This could lower the barrier to entry for creating complex applications and enable faster innovation across industries. The focus on agent orchestration hints at a future where AI takes on more of the heavy lifting in software creation, freeing up human developers for higher-level design and strategic tasks.

Sponsor Spotlight: Blacksmith

The article is sponsored by Blacksmith, a service designed to speed up GitHub Actions. Blacksmith claims to run actions twice as fast for 75% less cost by using high-performance gaming CPUs. It also offers full observability for CI pipelines. Developers can try Blacksmith for free, receiving 3,000 free minutes per month.

Source: Cursor ditches VS Code, but not everyone is happy… (YouTube)