Cursor August 19, 2026 Update Explained

TLDR

Cursor's cloud agents can now work on their own for much longer, automatically responding to updates and fixing code without needing you to babysit them.
You can now give agents long-term goals, and they will keep trying until the job is completely finished.
"Subagents" (helper bots) can now run on their own private virtual computers, letting them test code in isolation without crashing your main setup.
You can now send messages to an agent while it's working without interrupting its current task.
This update is a huge step toward making Cursor a fully autonomous coding assistant. Instead of just writing code while you watch, these cloud-based agents can now monitor your projects, talk to your team in Slack, and fix issues on their own. This matters most to developers and teams who want to automate the boring parts of building and maintaining software.

Subscriptions: Cloud agents can now "subscribe" to events. This means you can have Cursor watch your GitHub Pull Requests (PRs, which are proposed code changes) or specific Slack threads. When something happens there, the agent wakes up and gets to work. For example, if an automated system finds a bug in a PR the agent made, the agent will automatically try to fix it.
Custom Modes: You can now pin a specific "skill" to your chat as a Custom Mode. Think of this as an "always-on" setting that keeps the agent focused on one specific way of doing things. You can turn this on by typing /, picking a skill, and pressing Option+Enter (Mac) or Alt+Enter (Windows).
Isolated Subagents: A subagent is a helper bot that works for the main agent. These helpers can now run on their own virtual computers (virtual machines). This means each helper gets a fresh, isolated copy of your project to test things in. You can even tell Cursor to "swarm" the app with multiple helpers, each looking for bugs in their own private workspace without messing up each other's work.
The /goal Command: You can now type /goal to give the agent a big, long-term objective. The agent will keep working toward that goal until it is totally done. For example, you could tell it to fix all the broken tests, and it will keep plugging away until they all pass.

Better Steering: You can now send a message to an agent while it is working without interrupting what it's doing. Instead of cutting the bot off mid-action, your message will wait until the agent finishes its current step, making it much easier to guide its work without causing errors.

No fixes were mentioned in this update.

No breaking changes in this release.

Cursor August 19, 2026 Changelog