Cursor 3 Launches Agent-First Coding Interface to Challenge OpenAI Codex and Anthropic Claude Code

The new interface, developed under the code name Glass, centers around a natural language prompt box where developers can describe tasks for AI agents to complete autonomously. Unlike traditional integrated development environments, Cursor 3 positions itself as a conversation platform where developers manage multiple AI agents rather than writing code directly.

Agent-First Architecture and Integration

Cursor 3 operates within the company's existing desktop application, integrating the agent interface alongside its traditional IDE. Developers can prompt cloud-based agents to generate features and review the resulting code locally on their machines. A sidebar allows users to monitor and manage multiple concurrent AI agents.

Jonas Nelle, one of Cursor's heads of engineering, described the shift toward agent-driven development workflows. According to Nelle, the product anticipates a future where developers spend their time "conversing with different agents, checking in on them, and seeing the work that they did" rather than writing code themselves.

The integration approach differentiates Cursor 3 from standalone desktop applications for Claude Code and Codex, though the core functionality remains similar across platforms.

Competitive Pressure from AI Labs

Cursor faces mounting competition from OpenAI and Anthropic, which offer coding tools through subsidized subscriptions that provide over $1000 worth of usage for $200 monthly plans. Several developers interviewed by Wired reported shifting from Cursor to Claude Code and Codex primarily due to these generous rate limits.

The startup previously offered heavily subsidized pricing until June 2025, when it transitioned to usage-based pricing to improve margins. This shift upset some users but reflected Cursor's need to build a sustainable business model while competing against AI labs with significantly larger funding rounds.

Ronald Mannak, founder of Pico AI, and Jack Crawford, cofounder of mVara, both indicated they have largely moved away from Cursor in favor of Claude Code due to subscription value. Rate limits and pricing appear to drive tool selection more than feature differentiation for many developers.

In-House Model Development Strategy

Cursor has begun training proprietary AI models to reduce dependency on third-party providers and improve cost efficiency. The company recently launched Composer 2, an AI model based on an open-source system from Chinese AI lab Moonshot AI, with additional pretraining and post-training by Cursor's team.

Nelle indicated that developers typically select AI models based on performance, price, and speed, positioning Composer 2 as competitive across these dimensions. Cursor plans to train future Composer models from scratch rather than building on existing open-source foundations.

However, model training requires substantial capital investment. As OpenAI and Anthropic increase their focus on coding tools, Cursor's ability to compete may depend on raising additional funding to support model development and infrastructure costs.

Implications for European Development Teams

European development teams evaluating AI coding tools should consider several factors beyond feature parity. Usage-based pricing models may provide more predictable costs for enterprise budgets compared to subscription plans with unclear long-term sustainability. Additionally, in-house model development by vendors like Cursor could offer more control over data handling and compliance with European regulations.

Multilingual development teams may benefit from evaluating how different coding agents handle non-English codebases and documentation. The agent-first approach also raises questions about code review processes and quality assurance workflows that European enterprises typically require.

Cursor 3 represents the startup's attempt to maintain competitiveness in an increasingly crowded market dominated by well-funded AI labs, though pricing pressures and capital requirements present ongoing challenges. According to reporting by Wired, Cursor is pursuing additional funding at a $50 billion valuation to support its expansion in the agent-driven coding market.