Cursor Launches Agent-First Coding Tool to Challenge OpenAI and Anthropic
Cursor announced Thursday the launch of Cursor 3, a new interface that lets developers delegate coding tasks to AI agents. The product is the startup's direct response to Anthropic's Claude Code and OpenAI's Codex, which have captured millions of developers in recent months.
The shift matters because the nature of AI-assisted coding has changed fundamentally. Where Cursor built its reputation helping developers write code faster within an IDE, the new standard is agents that handle entire tasks autonomously. Developers increasingly spend their time "conversing with different agents, checking in on them, and seeing the work that they did," according to Jonas Nelle, one of Cursor's heads of engineering.
How Cursor 3 Works
The new interface lives alongside Cursor's existing IDE in the desktop app. A text box lets developers describe a task in plain language. They press enter. The AI agent works without requiring a single line of code to be written.
A sidebar shows all running agents and their status. What distinguishes Cursor 3 from Claude Code and Codex desktop versions is the integration-developers can prompt an agent in the cloud to build a feature, then review the generated code locally on their machine.
The Business Pressure
Cursor faces a capital disadvantage. OpenAI and Anthropic have raised tens of billions more and can afford heavily subsidized subscriptions to acquire customers. Claude Code and Codex users get over $1,000 worth of usage for $200 monthly plans.
Several developers told WIRED they've shifted their AI coding work away from Cursor to Claude Code and Codex. The reason: rate limits and subscription value. Ronald Mannak, founder of Pico AI, says his decision hinges on "whichever tool has the most generous rate limit." Jack Crawford, cofounder of mVara, rarely uses Cursor anymore despite heavy usage last year.
Cursor ended its subsidized subscription plan in June 2025 and moved to usage-based pricing. The company said the shift was necessary to improve margins and build a sustainable business. But it opened a gap competitors exploited.
Cursor's Strategy to Compete
The startup is training in-house AI models it can serve cost-effectively. Cursor recently launched Composer 2, based on an open-source system from Chinese AI lab Moonshot AI, with additional training from Cursor. The company plans to train future Composer models entirely from scratch.
Developers typically choose AI models based on performance, price, and speed. Cursor argues Composer 2 competes on all three fronts.
But training models is expensive. Cursor has historically done more with less, but the coding agent race is intensifying. OpenAI and Anthropic recognize the market size and are investing heavily. Without significantly more capital raised quickly, staying competitive in an agent-first world will be difficult.
Cursor is raising fresh capital at a reported $50 billion valuation-nearly double its valuation from last fall. The company recently expanded into a former movie theater in San Francisco's North Beach, signaling growth from its scrappy startup phase.
For product teams watching this competition, the takeaway is clear: the market has shifted to agents handling developer workflows. Companies that can't afford to subsidize customer acquisition or train proprietary models face mounting pressure to find differentiation elsewhere.
Learn more about AI Coding Courses and Generative Code Courses to stay current on these tools.
Your membership also unlocks:
