Expo Sells Cloud Phone Simulators to AI Coding Agents at $0.10 a Minute

Expo opened its cloud simulators to every paid customer on November 12, 2026, charging $0.10 per simulator minute for AI coding agents to run and verify app changes on remote iOS simulators and Android emulators. The launch turns a limited-access preview into a broadly available product, and it targets the problem Expo sees as the new bottleneck in software teams: verification, not code generation.

The move matters because it extends a real business line into the agent economy. Expo already sells build, CI/CD, app-store submission, over-the-air update and monitoring services. Simulators adds a testing stage to that stack, and it arrives as agents produce changes faster than engineering teams can review them.

What Changed on November 12?

The paid-plan launch expands the EAS Simulator preview that Expo introduced on September 14, 2026. Access before now ran through a waitlist. An Expo blog post dated November 10, 2026 was still directing developers to that waitlist, which makes broader paid-plan availability the new step rather than the first public demonstration of cloud devices.

Pricing is usage-based. According to Expo's pricing page, the Starter, Production and Enterprise plans each include 60 simulator minutes per month until 2027, with usage-based pricing beyond that allowance. Paid sessions can run for up to 115 minutes.

Why Do Agents Need Their Own Phones?

Mobile development has a hardware constraint that web development does not. iOS simulators run only on Apple hardware, while cloud agents typically run on Linux. An agent that cannot run the app has to assess code without seeing the resulting behavior.

The local alternative eats developer-machine capacity. In Expo's example, seven agents working in parallel can require seven simulators or emulators, and Android emulators can each use 4 GB or more of RAM. Remote sessions shift that compute onto Expo's infrastructure and let teams run an iOS simulator from Linux, from Windows through the Windows Subsystem for Linux, or from a cloud sandbox without a local Mac.

The intended loop starts with a bug report in a tool such as Linear. An agent reproduces the issue, writes a fix and tests it on cloud devices in parallel, away from the developer's machine. It can attach a screen recording, screenshots and logs to the pull request, while teammates open the running build from a link without a TestFlight build.

Expo co-founder and CTO James Ide described the review model the company is after: "The agent was the fast part. Expo Simulators gives every agent a simulator, so a fix arrives for review already tested on several simulators, and the only decision left is whether it's the right code. Watch it run, then merge."

The service works with Cursor, Claude Code and Codex. It adds no AI model of its own. Expo is supplying the device runtime and evidence rather than another coding model.

What Can the Simulators Actually Run?

A session can be started through EAS CLI, the REST API or an authenticated creation link, according to the Expo Simulators documentation. The session can install:

• an Expo Build
• a build whose native code matches the project's fingerprint
• an application archive
• Expo Go

Without an application source, the session starts with a blank device. Agents and test jobs can control apps through agent-device, Argent or Appium, and a browser preview lets people watch the run. The official Expo Simulators skill packages access checks, build preparation, session limits, device control, evidence collection and cleanup for agents.

The product page lists support for React Native, Swift, Kotlin, Flutter, Unity and other frameworks. Apps do not have to be built with Expo or React Native, but the build must run on an iOS Simulator or Android Emulator, and the project must be linked to Expo Services. Each session runs in an isolated environment with session-scoped credentials and access-controlled previews. Stopping a session disables the simulator and controller endpoints.

What Limits Come With Paid Access?

Parallelism does not create unlimited account capacity. Expo's usage and billing guide says simulator sessions draw on the account's job concurrency and queue when no capacity is free.

Session lifecycle has a cost trap for automated workflows. Ending a CI job or returning from a non-interactive command does not stop the remote device. Teams must stop it explicitly or rely on a duration or idle limit; otherwise an unattended session continues consuming usage. That makes cleanup and session limits part of cost control for agent-driven pipelines.

The limitations guide adds three material qualifications:

• the simulator CLI commands remain experimental
• the service does not provide physical devices
• recordings may miss frames and therefore cannot prove or disprove a brief frame drop

A cloud runtime can verify app behavior, but it does not eliminate the need for physical-device testing or profiling where those tools are required.

What Comes Next: Agent Sessions

Expo's next disclosed step is Agent Sessions, a planned enterprise layer for triggering agents from sources such as customer feedback and for protecting against prompt injection, where malicious input could direct an agent to expose data. Expo expects to share details by the end of November 2026. This is a roadmap item, not a capability included in the current launch.

The overall shape of the product is a division of labor rather than an autonomous release process. Agents supply tested changes and runtime evidence, account limits govern how much runs at once, and people retain the merge decision — a structure that keeps the human review step intact even as the testing work moves off the developer's machine.