Cursor shipped two updates in quick succession that change what “using an AI coding tool” means. Version 3.7 (17 June) added cloud environment setup and cloud subagents — type /in-cloud and Cursor spins up an isolated VM on its own branch to fix CI, investigate a bug, or explore a codebase, while your laptop stays free. Version 3.9 (29 June) took the local machine out of the loop entirely: an iOS app that lets developers launch, check on, and steer those same cloud agents from a phone, each running in its own fully-provisioned environment for testing and verification.
Why this matters more than another feature list
The direction is the interesting part, not any single feature. A year ago, AI coding tools meant faster autocomplete inside an editor a developer had open. Now the unit of work is an agent running independently in the cloud, and the developer’s job is closer to reviewing and directing several of them at once — from a laptop, or increasingly, from a phone between meetings. Cursor isn’t alone in this: Anthropic, Google, and Cognition have all shipped some version of parallel or backgrounded agent execution in the last quarter. The tools are converging on the same model faster than any of them are differentiating from it.
So what
For anyone commissioning software, this changes what a competent development partner’s workflow looks like. A team running one agent per developer, watched closely, is behind the curve compared to one running several agents in parallel per developer with proper review gates — the difference shows up in delivery speed and, if review discipline slips, in code quality risk. It’s worth asking a prospective partner directly how they supervise parallel AI agents, not just which tool they use. We cover how we structure that supervision on our AI-assisted development page, or get in touch to talk through a project.