Cloud Agents with Computer Use
read at source ↗ cursor.com
Cloud Agents with Computer Use
Source: Cursor Date: 2026-02-24 URL: https://cursor.com/changelog/02-24-26
Summary
Cursor ships computer use for cloud agents: agents run in isolated VMs with full dev environments, can access and test the software they build, record demo videos and screenshots, and generate merge-ready PRs with artifacts (videos, screenshots, logs). Available across web, desktop, mobile, Slack, and GitHub integration.
Implications
Agent-IDE feature race. Computer use for cloud agents is Cursor’s most direct response to Anthropic’s Claude computer use capability and Devin’s autonomous agent model. The ability to test software the agent just wrote — including UI testing with screenshots — closes the write-verify loop that previously required a human. Generating demo videos as PR artifacts is a product layer on top of code: the agent is now doing QA and documentation, not just coding.
Pricing model evolution. Isolated VM per agent with full dev environment is compute-intensive. This release almost certainly has per-usage pricing implications that aren’t announced here — watch for a follow-up pricing update or a “compute units” tier. The Devin comparison is apt: Devin charged ~$500/month, positioning autonomous agents as high-value/high-cost.
Watch: Whether computer use agents can be configured to run on self-hosted infrastructure (extending the self-hosted cloud agents capability), and whether the video/screenshot artifacts land in a reviewable dashboard.