AI NewsProductivityAnnouncement
Shopify built Remote Dev so AI agents stopped killing engineers' laptops
Shopify's AI mobile agents used to lock up developer laptops after three or four parallel runs. A fleet of remote Macs for iOS and Linux VMs for Android unlocked the volume. River's share of merged mobile pull requests went from 6 percent to 61 percent.

Why it mattersA laptop was the real limit on parallel AI work for many mobile teams. Moving the simulator off the laptop is what let the agents actually land work.
A mobile team at Shopify has been running an AI coding agent called River since earlier this year. The team writes that "just three or four workflows were enough to max out a laptop". Compilers, iOS simulators and Android emulators all wanted the same resources. That is from the Shopify Engineering post of 22 October. The team's answer is Remote Dev, a shared fleet that moves the heaviest parts off the laptop.
The numbers are the reason to read it. River's share of merged mobile pull requests started at 6 percent. After Remote Dev opened up, that share grew to 61 percent of merged mobile pull requests. The authors are Michael Garfinkle, Mauricio de Meirelles and David Cortes.
Two platforms, one SSH session
The iOS side uses a fleet of remote Macs. Each Mac arrives pre-warmed. The code is checked out. Tools are installed. The app is already built. Caches are warm. The team writes that this cuts setup "from over 30 minutes to just a few minutes". Each agent gets an isolated environment on one of these Macs.
The Android side is harder, because macOS does not support nested virtualisation natively. The team's answer is to connect each session to a separate Linux VM with nested virtualisation enabled. A developer or an agent reaches both platforms through the same SSH session. An agent can run the build, test it, screenshot it and open a pull request. The developer's laptop never renders the simulator.
Live preview over WebSockets
Remote Dev also streams the iOS Simulator and the Android Emulator over WebSockets. The team writes: "anyone open a link and use the actual running app in real time". So a reviewer, a product manager or an agent gets a browser-based view of the running build. Nobody has to install the app locally to see what changed. That answers the obvious objection to moving the simulator off the laptop, which is that the developer loses the view.
What a reader can take from this
The pattern is portable. A mobile team running one coding agent with no concurrency has no need for this. A team running three or four parallel agents hits the same wall Shopify did. The fix is on the Shopify page in enough detail to copy. The two parts that matter are the pre-warmed Macs with warm caches. The other part is Linux VMs with nested virtualisation that handle Android emulation separately from iOS.
The reproducible result bar for a practice story is where this one lands. The 6 percent to 61 percent move is Shopify's own measurement of its own merged pull request share. It is not a benchmark, and another team will see a different number. The method is explicit enough that a mobile team can repeat it and measure its own change. A team that was told AI agents do not work for mobile has a counter-example here. The simulator is too heavy on a laptop. It is not too heavy on a shared fleet.
The Shopify page also lists what the team plans to add. The list is continuous health monitoring, automated crash detection with fixes, and automated pull requests for crashes an agent finds. Those are not live yet. The core Remote Dev platform is.
Source
Shopify Engineering, with a write-up on InfoQ.
This item was written by an AI system from the linked source. Reveneau is responsible for what it publishes.
Get AI News in your inbox
New developer tools, model and agent releases, and how teams are actually using them to release software. Short, and only when there is something worth reading.

