Cursor Gives AI Agents Their Own Computers, Signaling a Shift in How Developers Work

Cursor, the AI coding startup valued at $29.3 billion, announced a significant update to its cloud agents that gives each AI agent its own virtual machine with a full development environment. The agents can now build software, test it, interact with it through a browser and desktop applications, and produce video recordings, screenshots, and logs to prove their work — all without touching a developer's local machine.

The update represents what the company calls "the biggest shift in how we build software since the move from Tab autocomplete to working synchronously with agents." Cursor says more than 30% of the pull requests it merges internally are now created by agents operating autonomously in cloud sandboxes.

What the agents can do

Unlike local coding assistants that generate code and hand it off, Cursor's updated cloud agents iterate on their own output. They can spin up servers, navigate web pages, manipulate spreadsheets, resolve merge conflicts, and verify UI changes — all inside an isolated VM. Developers can trigger agents from the web, mobile, Slack, GitHub, or Cursor's desktop app.

In one internal example, an agent was tasked with reproducing a clipboard exfiltration vulnerability. It built an exploit page, launched a local server, loaded the page in Cursor's browser, and recorded itself executing the full attack flow — all without human intervention. In another case, an agent spent 45 minutes autonomously testing Cursor's documentation site, checking sidebar navigation, search functionality, theme switching, and other UI elements.

The key shift is parallelism. Local agents compete for a single machine's resources, limiting developers to a few concurrent tasks. Cloud agents remove that bottleneck entirely.

"Instead of having one to three things that you're doing at once that are running at the same time, you can have 10 or 20 of these things running," Alexi Robbins, co-head of engineering for asynchronous agents at Cursor, told CNBC. "You can have really high throughput with this."

What this means for developers

The implications are significant for developer workflows. Rather than breaking tasks into small, manageable chunks and closely supervising AI output, developers can now delegate larger, more ambitious tasks and review the results after the fact. Cursor frames the developer's evolving role as one focused on "setting direction and deciding what ships" rather than line-by-line implementation.

Jonas Nelle, Cursor's other co-head of engineering for asynchronous agents, described the shift in blunt terms: "They're not just writing software, writing code, they're sort of becoming full software developers."

That framing will likely provoke debate. While autonomous agents handling routine fixes and feature implementation could free up senior developers for higher-level architecture and design decisions, it also raises questions about skill atrophy, code quality oversight, and the changing economics of software teams.

A crowded and intensifying market

Cursor isn't operating in a vacuum. Anthropic's Claude Code has grown to over $2.5 billion in run-rate revenue, OpenAI's Codex has surpassed 1.5 million weekly active users, and Microsoft's GitHub Copilot counts more than 26 million users. Cursor, which crossed $1 billion in annualized revenue last November, needs features like autonomous cloud agents to differentiate.

The company says its near-term roadmap includes coordinating work across many agents simultaneously and building models that learn from past runs. The longer-term vision is what Cursor calls "self-driving codebases" — where agents not only write code but merge PRs, manage rollouts, and monitor production systems.

For now, developers can try the updated agents at cursor.com/onboard, where the agent will configure itself on a codebase and record a demo of its work.