Michael Truell, CEO of Cursor:
When we started building Cursor a few years ago, most code was written one keystroke at a time. Tab autocomplete changed that and opened the first era of AI-assisted coding. Then agents arrived, and developers shifted to directing agents through synchronous prompt-and-response loops. That was the second era.
Now a third era is arriving. It is defined by agents that can tackle larger tasks independently, over longer timescales, with less human direction. As a result, Cursor is no longer primarily about writing code. It is about helping developers build the factory that creates their software.
Thirty-five percent of the PRs we merge internally at Cursor are now created by agents operating autonomously in cloud VMs.
Agent Orchestration and Agent Swarms are a couple of ways folks are referring to this idea. Steve Yegge had predicted several weeks ago (a very long time horizon in the world of AI) that this would be the next frontier in agentic engineering.
I remain sceptical though. I'm not saying I don't trust that 35% of the PRs at Cursor are being opened this way; it is very believable, given how good the frontier models are now. But not all PRs are equal, and I wager these 35% are relatively simpler bugs/features. Or that there is a lot more work being done to iterate on the PRs once they are raised.
My argument isn't that this isn't useful work. On the contrary, background agents taking such issues off of engineers' hands is invaluable, as they can now focus on work that provides higher leverage. But I don't believe the logical extension of this is that "the vast majority of development work" will be done this way in a year.