Cursor launched Cursor 3 on Thursday, a new agent-first coding interface that puts AI agents at the center of the development workflow. Developers type tasks in natural language, agents execute them autonomously, and a sidebar manages multiple concurrent agents. The product, developed under the code name Glass, is Cursor’s answer to the coding agent tools that Anthropic and OpenAI have shipped over the past 18 months, according to WIRED.
The Product Shift
Cursor’s previous product was an AI-powered IDE where developers wrote code with model assistance. Cursor 3 inverts that relationship. “A lot of the product that got Cursor here is not as important going forward anymore,” Jonas Nelle, one of Cursor’s heads of engineering, told WIRED. The new interface is optimized for developers who spend their time “conversing with different agents, checking in on them, and seeing the work that they did,” Nelle said, rather than writing code themselves.
The differentiator from Claude Code and Codex: Cursor 3 integrates the agent interface alongside the existing IDE in the same desktop app. Alexi Robbins, co-head of engineering for Cursor 3, demonstrated a workflow where a cloud-based agent generates a feature and the developer reviews the code locally. Claude Code and Codex run as standalone products.
The Subsidy Problem
The structural challenge is pricing. Claude Code and Codex offer subscriptions at $200 per month that deliver over $1,000 worth of usage, WIRED has previously reported. Multiple developers told WIRED they’ve shifted away from Cursor because of those subsidized plans. Ronald Mannak, founder of Pico AI, said rate limits drive his tool choices. Jack Crawford, co-founder of mVara, said he rarely uses Cursor anymore.
Cursor moved to usage-based pricing in June 2025 after running its own subsidized plan. OpenAI and Anthropic have raised tens of billions more capital than Cursor, enabling sustained below-cost pricing for customer acquisition.
Cursor’s Counter-Strategy
Cursor is training in-house models to reduce its dependence on the companies it competes against. Composer 2, built on an open-source system from Chinese AI lab Moonshot AI with additional pretraining by Cursor, launched in March. Cursor plans to train future Composer models from scratch. The company is also reportedly raising capital at a $50 billion valuation, nearly double its last round, per Bloomberg.
The coding agent market now has four major entrants shipping competing products in the same month: Cursor 3, Claude Code, Codex, and Claw Code (which crossed 72,000 GitHub stars this week). For developers, more competition means better tools. For Cursor, the question is whether an agent layer built on top of model vendor APIs can survive when those vendors are undercutting on price and shipping their own agents.