Cursor Unveils an Innovative AI Agent Experience to Compete with Claude Code and Codex

Cursor announced on Thursday the introduction of Cursor 3, a fresh product interface that empowers users to generate AI coding agents to handle tasks for them. This offering, developed under the code name Glass, is Cursor’s answer to agentic coding tools such as Anthropic’s Claude Code and OpenAI’s Codex, which have gained traction among millions of developers in recent months.

“Our profession has transformed dramatically in the past few months,” stated Jonas Nelle, one of Cursor’s engineering leaders, during an interview with WIRED. “Much of what brought Cursor to this point is becoming less critical moving forward.”

Cursor increasingly competes with top AI labs for developers and enterprise clientele. The company has been a pioneer in integrating AI models from OpenAI, Anthropic, and Google into developer tools, making it one of the largest customers of these AI firms. However, in the past 18 months, OpenAI and Anthropic have rolled out their own agentic coding products, offering them through heavily subsidized subscriptions that exert pressure on Cursor’s business.

While Cursor’s primary product allows developers to code in an integrated development environment (IDE) and utilize an AI model for assistance, newer products like Claude Code and Codex focus on enabling developers to delegate entire tasks to an AI agent—sometimes allowing for the launching of multiple agents simultaneously. Cursor 3 represents the startup’s take on an “agent-first” coding solution. As noted by Nelle, the product is tailored for a scenario in which developers spend their time “engaging with various agents, monitoring their progress, and reviewing their outputs,” rather than writing code on their own.

Cursor is introducing its new agentic coding interface within its existing desktop application, where it will coexist with the IDE. At the center of a new window in Cursor, there’s a text box where users can input, in natural language, a task they want an AI agent to perform—it resembles a chatbot more than a traditional coding interface. After hitting enter, the AI agent immediately gets to work without necessitating the developer to write any code. On the left sidebar, developers can oversee and manage all the active AI agents they have in Cursor.

What sets Cursor 3 apart from desktop apps like Claude Code and Codex is its combination of an agent-first approach with Cursor’s AI-enhanced development environment. During a demonstration, Cursor’s co-head of engineering for Cursor 3, Alexi Robbins, showcased to WIRED how users can instruct a cloud-based agent to develop a feature, and then review the generated code locally on their device.

Nelle and Robbins emphasize that it’s irrelevant which interface developers utilize; their primary goal is to have users engaging with Cursor.

Competing With the AI Labs

I visited Cursor’s office in North Beach, San Francisco, last week. The startup is reportedly securing new funding at a valuation of $50 billion—almost double its worth from a funding round last fall—and has moved into a former movie theater. Previously, employees would toss their shoes in a pile by the entrance, but now a series of large shoe racks indicates one way the company is maturing.

Still, Cursor maintains the atmosphere of a startup. Employees convey that this is part of the allure of working there; the company can move rapidly and doesn’t feel overly corporate. However, as it strives to keep pace with Anthropic and OpenAI in the agentic coding competition, that scrappiness might not suffice. This struggle to develop the finest AI coding agent could mark Cursor’s most capital-intensive phase to date.