TL;DR
Expo Simulators provides cloud iOS simulators and Android emulators that coding agents can drive and record. Expo advertises $0.10 per simulator minute and 60 monthly minutes included in paid plans until 2027. Recorded evidence makes review easier, but it does not replace a well-designed test or prove an app is correct.
Introduction
Coding agents can generate a pull request faster than a reviewer can reproduce its behavior. Mobile apps make the gap particularly visible: a reviewer may need the right build, platform tooling and device configuration just to see whether a change works. Running several agents at once multiplies that setup burden.
Expo’s answer is to move the test device into a cloud session. Its launch announcement describes on-demand iOS simulators and Android emulators for agents, continuous integration and human reviewers. The service is available on paid plans, according to Expo. (Source : Expo — Introducing Expo Simulators)
An execution environment, not a test strategy
A session gives an agent a device it can operate remotely. Expo says sessions run on separate virtual machines and support familiar controllers including agent-device, Argent and Appium. An agent can install a build, interact with it and leave a screen recording, screenshots and logs for review. (Source : Expo — Introducing Expo Simulators)
That division matters. The simulator supplies an environment and an evidence trail; the team still defines what should be tested. A video of a login screen successfully opening does not establish that expired credentials, network loss or accessibility settings are handled correctly. Evidence is valuable only when the scenario has a meaningful expected outcome.
A practical pull-request workflow would identify the changed user journey, prepare test data, start an isolated session, execute the scenario and attach the recording and logs. The reviewer can then compare the evidence with the change rather than treating an agent’s assertion of success as proof. Independent automated tests and code review still matter.
What parallelism costs
Expo’s announcement quotes $0.10 per minute for iOS and Android and 60 minutes per month included in Starter, Production and Enterprise plans until 2027. These are included minutes attached to paid plans, not a promise that anyone on a free plan can run a free hour. The billing documentation says access is on Starter, Production and Enterprise plans. (Source : Expo — Pricing; Expo documentation — Usage and billing)
As a simple illustration, seven parallel sessions running for ten minutes each consume 70 simulator-minutes. At the advertised rate, that is $7 of gross usage before any included allowance or plan-specific terms. Parallelism reduces elapsed waiting time; it does not eliminate aggregate device minutes. This example excludes build time, model inference, plan subscription fees and other possible costs.
The distinction helps teams compare architectures. A single developer with one short test may prefer a local simulator. A CI system running many independent checks may value isolation and evidence enough to pay for cloud sessions. Neither choice wins on device-minute pricing alone: waiting time, machine management and review effort are part of the total cost.
Session limits are a safety feature, not a cleanup plan
Expo’s billing documentation sets a maximum session duration of 115 minutes and allows shorter limits and idle timeouts. It warns that a non-interactive session can continue after the command or CI job ends until explicitly stopped or until a configured limit is reached. Sessions share account job concurrency and may queue when no capacity is available. (Source : Expo documentation — Usage and billing)
That creates two operational requirements. First, stop sessions in a cleanup step even when a test fails. Second, set a conservative maximum and idle timeout as a backstop. A queue changes latency and throughput even if queued time does not correspond to active device usage. Measure the whole pipeline, not just the time spent tapping through the app.
Expo’s launch page says the service is available on paid plans. Prices and entitlements can change, so check the current terms before building a financial forecast around them.
Evidence has its own security boundary
An isolated VM can prevent concurrent agents from stepping on one another, but the resulting recording and logs may contain sensitive information. A realistic test account, personal data on screen or a token in a network trace can turn a helpful review artifact into a data-retention problem.
Use synthetic accounts and restrict access to session artifacts. Check retention, sharing links and deletion policies against the team’s requirements. Isolation is not a substitute for deciding who can view captured evidence or what may be recorded. Expo describes sessions as isolated and says recordings and logs remain available for review. (Source : Expo — Stability and security)
The product’s strongest proposition is therefore not that agents no longer need humans. It is that the human can inspect a concrete trace rather than reproduce every change from scratch. That only scales when the test scenario, access control and stopping behavior are as deliberate as the generated code.
FAQ
Is the included hour available on the free plan?
Expo’s announcement ties the 60 monthly minutes to paid Starter, Production and Enterprise plans. Check current plan terms before relying on the allowance.
Does a recording prove the change is correct?
No. It shows what happened during a particular run; it does not establish coverage of untested cases or correctness of the test itself.
Why give each agent a separate session?
Expo’s isolation model is designed around separate sessions. Separate devices reduce interference when agents install builds or change app state concurrently.
Does closing CI automatically stop the device?
Not necessarily. Expo’s documentation says non-interactive sessions can continue until stopped or a duration or idle limit is reached.
Further Reading
— The Agent Report