Cursor now lets you run coding agents from your phone

For years, writing code meant being tethered to a desk, watching terminal logs, and waiting on local builds. Cursor wants you to close the laptop and manage your codebase from your phone while waiting in line for coffee.

Cursor introduced Cursor 3, a new agentic interface developed under the code name Glass. The update lets developers launch autonomous AI coding agents on remote machines and control them asynchronously from mobile devices, turning software development into a location-independent supervisory job.

An agent cockpit inside the IDE

The new interface lives directly inside the existing Cursor desktop app, alongside the traditional IDE. Right in the center sits a natural-language input field that looks more like a chatbot than an editor. You describe what you need, hit enter, and the agent gets to work without requiring you to touch a single line of code. A left sidebar displays and manages all active agents running across your environments.

Alexi Robbins, co-head of engineering for Cursor 3, demonstrated how a user can spin up an agent in the cloud to build out a feature, then pull the generated code down to review locally. Jonas Nelle, another engineering head at Cursor, told WIRED that software engineering has completely shifted in recent months, noting that much of what brought Cursor to its current scale will not matter as much going forward.

The startup is now reportedly raising fresh capital at a $50 billion valuation, nearly doubling its valuation from last fall. It has also expanded into a converted movie theater in San Francisco’s North Beach, swapping a messy pile of shoes at the entrance for orderly shoe racks as it scales up.

Feeling the squeeze from subsidized giants

The launch comes as model makers Anthropic and OpenAI aggressively muscle in on Cursor's territory. Tools like Claude Code and OpenAI's Codex have siphoned away developers with heavily subsidized subscriptions. Developers on $200-a-month tiers for those services reportedly receive well over $1,000 worth of compute.

That disparity is already changing developer behavior. Ronald Mannak, founder of Pico AI, said he largely shifted from Cursor and Windsurf to Claude Code and Codex because he simply follows the most generous rate limits. Jack Crawford, co-founder of mVara, echoed the sentiment, noting he rarely touches Cursor anymore after relying heavily on it last year.

Cursor, by contrast, ended its own subsidized plans back in June 2025 to introduce usage-based billing, a move aimed at protecting unit economics that frustrated parts of its community. With model providers holding billions more in venture backing, Cursor cannot afford to win a pure pricing war.

Why it matters

For developers, Cursor 3 brings real workflow freedom. Instead of babysitting an LLM prompt by prompt, you can assign high-friction tasks to remote instances and monitor pull requests from your phone. By keeping the IDE integrated with the agent dashboard, Cursor spares you from having to choose between a command-line agent and a full-featured editor for final code review.

For the broader industry, this release signals that the code-completion era is officially over. The battleground has shifted to multi-agent orchestration, where foundation model companies are attempting to vertically integrate the entire developer stack and subsidize it into oblivion.

We think Cursor’s decision to blend the agent console directly into its desktop IDE is clever because it doesn't force engineers to abandon local tooling. However, relying on product polish and remote control features looks precarious when competitors are essentially giving away state-of-the-art inference at an 80 percent loss. If developers keep chasing raw compute limits, Cursor may find that superior UX is not enough to offset subsidized tokens.