What's this about?
Cursor updated its cloud agents so they pick up work from events, hold a goal until it's met, and stay on course through long sessions.
What exactly changes when an agent is "event-triggered" rather than invoked?
An invoked agent waits for you to ask. An event-triggered agent watches for something happening — a failing test, a new issue, a deploy — and starts working on its own. Combined with "hold a goal until it's met," that's a shift from a tool you use to a process that runs. The design difficulty is that nobody is watching when it starts, so the agent has to decide on its own when it's done, when it's stuck, and when to ask for help — which is exactly the long-horizon reliability problem the harness research has been circling all month.
Why is this worth noticing?
2,255 likes
- It's the same direction as Anthropic's daily maintenance routines from last week: scheduled or triggered agent work with human review at the end rather than human supervision throughout.
- "Stay on course through long sessions" is the honest description of what remains hard, and matches what StateM and Demystifying Agent Skills both identify — execution drift, not capability, is what breaks long tasks.
- It arrives days after Origin, and the two fit together: agents that work continuously in a repository need hosting built for continuous agent writes.