When developers hand off a task to Cursor's new cloud agents, something unusual happens. The agent doesn't just start coding in the same editor the developer is using. It fires up its own virtual machine. Its own development environment. Its own compute resources.

It's like hiring a developer — but the developer doesn't need your laptop.

"We're essentially giving every agent its own laptop," said Alexi Robbins, co-head of engineering for asynchronous agents at Cursor. "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. You can really have high throughput with this."

This is the next evolution of AI coding tools. And it might change what it means to be a developer.

What Changed

Cursor announced a major update to its AI coding agents on February 24, 2026. The key difference: agents now run on their own cloud-based virtual machines, completely separate from the developer's local environment.

Previously, even the most advanced AI coding agents shared resources with the human developer. They ran in the same editor, on the same machine, competing for compute power. If you wanted three agents working on different problems, your laptop would creak under the load.

Cursor's cloud agents solve this. Each agent gets its own virtual machine — essentially a cloud computer that behaves exactly like a physical one. The agent can install dependencies, run tests, spin up servers, and build entire projects without touching the developer's machine.

The Numbers Are Staggering

Cursor shared some remarkable statistics about internal adoption:

35% of all pull requests at Cursor are now generated by agents running on their own virtual machines.

That means more than one in three code changes proposed to Cursor's codebase come from AI — not externals, not contractors, but the company's own agents.

And it's not just code generation. The updated agents can:

• Test their own changes automatically
• Record their work through videos, logs, and screenshots
• Iterate on features until they're complete
• Handle complex, multi-step tasks that used to require human oversight

How It Works

Developers can trigger these agents from multiple surfaces:

• Cursor's desktop app
• The web interface
• Mobile devices
• Slack
• Microsoft's GitHub

The agent receives the task, spins up its virtual machine, gets to work, and reports back when it's done. During the process, it can run tests, fix bugs, and refine its approach — just like a human developer would.

"We're thinking of this less like a new feature and more like, 'This is what it's going to look like working with agents,'" said Jonas Nelle, Cursor's co-head of engineering for asynchronous agents. "They're not just writing software, writing code — they're becoming full software developers."

The Competition Is Intense

Cursor may have pioneered AI coding, but the market is heating up fast.

The battle for developer mindshare has become one of the most competitive spaces in AI.

What Developers Actually Do Now

The shift is changing the developer workflow in subtle ways.

Instead of writing every line of code, developers are becoming reviewers and architects. They define what they want built. They check the agent's work. They focus on "questions of taste and judgement" — the creative decisions that require human context.

Cursor's internal experience suggests this works. Their agents aren't replacing developers — they're multiplying them. One developer can now oversee what would have previously required a small team.

"We think of this as another step change," Nelle said. "You as an individual can do so much more by working with these agents."

The Bigger Picture

What Cursor is doing with cloud agents reflects a broader shift in AI: from tools that assist to agents that execute.

The original promise of AI coding was autocomplete — finishing your sentences. Then it was chat — answering questions. Now it's agency — completing tasks.

The cloud agent model pushes this further. By giving each agent its own compute resources, Cursor is essentially creating a workforce of virtual developers. They can work in parallel. They can work continuously. They can handle volume that no human team can match.

The implication isn't just faster coding. It's a different model for what software development looks like.

When 35% of your pull requests come from agents, you're not just using AI to help. You're running a hybrid team.

Sources:

• CNBC - Cursor announces major update to AI agents
• Cursor Features
• Cursor Blog
• Faros AI - Best AI Coding Agents for 2026