Expo launches cloud simulators for AI agent verification | Magica
Expo opens paid cloud simulators for AI agents to check mobile apps
Editorial Team
••📖8 min read
Expo Simulators gives coding agents remote iOS and Android environments with recordings for reviewers. The launch extends Expo's cloud development services into verification, with per-minute charges and shared capacity limits.
Expo has opened cloud iOS simulators and Android emulators to paid accounts, with an access exception for annual Enterprise plans.
Agents and reviewers get session recordings, screenshots and logs; the service also supports existing test suites.
Sessions cost $0.10 per simulator minute and share account concurrency, so parallel testing still faces a meter and a queue.
Expo, which supplies open-source app development tools and hosted cloud services, launched Expo Simulators on November 12, 2026. The service gives AI coding agents remote devices on which to run mobile apps and gather evidence for reviewers, extending Expo's business into the execution of agent-written changes.
Expo says in its launch announcement that verification is becoming a bottleneck as coding moves to background agents and parallel workflows. Each session runs in its own virtual machine, separating agents that could otherwise overwrite one another's app installations. Teams can watch the app live in a browser and inspect recordings, captured screenshots and logs afterward. Expo says it uses the same infrastructure and isolation layer as Build, Submit and Workflows, which already process millions of jobs; that is a company claim about its existing services, not a measure of Simulators adoption.
The company's earlier cloud offering, Expo Application Services, or EAS, began with Build and Submit: hosted services for compiling Expo and React Native apps and submitting them to app stores. Simulators adds a metered environment for running those apps during development and review. The practical question is whether easier access to devices produces enough useful verification to justify the additional spending.
Expo’s product illustration shows a browser-based simulator session with device controls and performance panels. Image: Expo. Source: Expo.
Remote execution, with familiar testing tools
An agent starts a session, installs a compatible build, interacts with the app through a controller and collects evidence. Expo supports agent-device, Argent and Appium, tools that control the remote device; any coding agent able to run shell commands can use the service. Expo says sessions also collect network traffic and CPU and memory usage, giving reviewers more than a video of the interface.
Google launched Gemini 4 Hydrogen and automatic model switching in Antigravity, claiming near-Carbon task success with less quota and faster completion. Lower token rates support the efficiency pitch, but reasoning settings, mixed competitive results and inconsistent benchmark labels leave the cost of completed work to be tested.
Editorial Team
The infrastructure removes a constraint for agents working away from a developer's laptop. The iOS Simulator requires a Mac, while Android emulators can encounter virtualization limits in cloud environments. Expo hosts the virtual devices, allowing the agent or continuous integration (CI) job to run elsewhere. Its announcement nevertheless says a local simulator remains the right tool for one agent on a developer's own machine.
The service accepts apps beyond Expo and React Native. Its product page lists Swift, Kotlin, Flutter and Unity among supported frameworks, provided the build runs on an iOS Simulator or Android Emulator. A project linked to Expo Services is required, but the build can come from an external URL. Existing Appium suites can connect, and agent-device can run supported Maestro YAML test flows, with only a subset of Maestro commands available.
Cloud mobile testing already has precedents. Amazon's cloud division described uploaded scripts, reports and screenshots in its AWS Device Farm launch announcement, which covered testing Android and Fire OS apps on real devices. That differs from Expo's virtual devices, but it narrows the novelty: Expo is integrating hosted execution and review evidence into its agent and app development workflows.
The open-source testing framework e2e provides one integration. Its integration walkthrough describes allocating one Expo session per test worker and running iOS tests from a Linux CI runner. Expo says the integration stops sessions when the run ends.
The e2e example mixes agent-directed actions, exact assertions that need no model call, and a separate model judging the screen. It can also run fully scripted tests. This gives teams a choice about where to use AI rather than requiring an agent for every action.
Expo lists $0.10 per minute for either platform, 60 included monthly minutes on Starter, Production and Enterprise plans until 2027, and a 115-minute paid-session limit. Graphic: Magica; sources: Expo. Source: Expo.
The bill includes more than device minutes
Expo lists the same simulator price for iOS and Android: $0.10 per minute. Its launch announcement says Starter, Production and Enterprise plans include 60 minutes per month until 2027. Access also requires a paid plan; the per-minute rate is not a standalone subscription price.
For illustration, ten sessions running for 20 minutes each total 200 simulator minutes, or $20 at the listed rate before included minutes and other charges. Parallel execution changes elapsed time, not the sum of those device minutes.
The usage documentation says sessions consume the account's Expo Services job concurrency and queue when capacity is exhausted. Paid sessions can run for up to 115 minutes. The session timer begins when the device is ready; the underlying job has a hard limit five minutes longer. Queue time does not count toward the separate 15-minute startup timeout.
Startup and model use also affect the economics. The e2e walkthrough estimates about three minutes to queue, boot and install a build for a session. That is an integration description, not a general startup guarantee. Its agent tests require an AI-provider credential; without one, the example skips those tests while retaining deterministic tests.
Caching can reduce repeated model work. In e2e's copy of a simple greeting-app example, it reports that a first run took 31.6 seconds and 11 model calls, compared with 9.7 seconds and two calls on the next run. The framework saves successful actions for replay and returns control to the agent when replay breaks. Those figures describe that example's test runs, not full build-to-review latency or a measured saving across production apps.
Session cleanup remains a spending control. Closing a terminal, returning from a non-interactive command or ending a CI job does not stop the remote device. An unstopped session runs until its maximum duration or a configured idle limit. Teams can set shorter limits; merely watching the preview does not reset the idle timer.
Paid access has an exception the headline plan list does not convey. The general-availability implementation describes access for paid accounts in good standing but excludes annual Enterprise prices at launch. Those customers are directed to their account executive.
Expo’s company-reported accessibility example pairs before-and-after screenshots at the largest text size with simulator session links. Image: Expo. Source: Expo.
Review evidence must match the code under review
A recording shows the path an agent exercised. It does not establish that the agent chose sufficient tests or that unvisited paths work. Expo's own workflow instructions expose another risk: recording the wrong build or device.
Its agent workflow reference warns against using a stale release build as evidence of live edits. It also warns that agent-device can silently fall back to a local simulator when remote configuration is missing. The checks therefore confirm the session ID and remote connection, rather than treating a plausible screenshot as proof of remote execution.
For live iteration, a development build connects to Metro, the server supplying JavaScript and assets, so edits appear through Fast Refresh. For a fixed screenshot, Expo's verification guide calls for a build matching the current source. The product page also describes reusing a build with matching native code and repacking it with current JavaScript, avoiding a full native rebuild. These are distinct ways of keeping the running app aligned with the change being checked.
Access and sharing need explicit choices. The setup guide says preview links require an Expo login with project access by default, with a public-link option available. Starting a new session requires the Developer role or higher, according to the product page. Controller credentials go into a local session configuration file that must not be committed or shared.
Live development introduces a separate exposure: the workflow reference says the tunnel carrying Metro's manifest, JavaScript and assets is publicly reachable by URL while running. Account signing authenticates tunnel creation; it does not establish that requests to that development-server URL require authentication.
What an adoption trial still needs to establish
Expo says design partners used the service for several months before launch. The November 12 v27.8.0 release replaced waitlist messaging with paid-plan access and removed experimental labels from simulator commands except egress-related functionality. Those dated changes clarify the release stage despite preview or experimental wording remaining in some guides.
The same release includes fixes for black iOS previews on multi-display simulators and an Android download dependency that could fail a session. The release history also records a subsequent November 12 change to warn about simulator and workflow outages before starting a session. These changes matter to a product whose value depends on successful execution and usable evidence; they do not establish an overall failure rate.
For a team deciding whether to adopt it, the next comparison should use its own app: time from build to review, queue and startup delays, simulator and model spending, and failures caught before a merge. The launch materials establish the availability of remote execution and retained artifacts. Whether those artifacts reduce review time or escaped bugs remains a question for measured use.
Japan is considering safety and reporting duties backed by enforceable orders for advanced AI. The proposal would strengthen its nonbinding framework, but capability thresholds, covered businesses and penalties remain unresolved.
Google says its new fast AI model is reaching Search users worldwide, improving complex answers and source discovery. The rollout expands access to Gemini 4, but its effect on answer quality and website traffic remains unmeasured.
Four nonprofits are asking a California court to stop five AI developers from building or deploying more capable systems. The case seeks enforceable limits even as OpenAI has previously reported a training pause and is launching defensive AI tools.
Kleiner Perkins and Clearlake Capital led Mirendil’s $1 billion Series A. Its agent-led model-building experiment points to smaller research teams, but company benchmarks, substantial compute costs and a private beta qualify the promise of wider access.
Mount Thor raised $45 million in equity, alongside separate hardware financing, to expand Apple infrastructure for AI agents and model inference. Existing Mac clouds and workload-specific performance limits put its demand and cost claims to the test.
Salesforce Ventures led a Series C that values Q-CTRL above $1 billion. The funding backs GPS-free navigation and quantum computing software, while undisclosed revenue and contested performance benchmarks leave the commercial case open.
Adobe demonstrated 10 experimental creative tools and opened Deep Brush and Hands On for testing. The lineup explores more direct control over AI edits, but existing alternatives and unproven production performance limit claims of novelty.
House Foreign Affairs Committee Chairman Brian Mast says outgoing Nvidia chief Jensen Huang has agreed to questioning, with a date still under negotiation. Speaker Mike Johnson says support for limiting chip exports to China is growing as lawmakers examine allegations involving Huang and a cloud partner.
Leaked correspondence and a reconstructed publishing program expose John Mark Dougan’s use of AI to stock imitation news sites with pro-Russian material. The records illuminate funding and working methods, while his claims of influencing other AI systems remain unproven.
William Blair raised AMD to Outperform as server-chip prospects and valuation improved. AMD already expects revenue growth above 80% in 2027, but customer incentives, custom accelerators and weaker quarterly cash generation complicate the investment case.
Apple’s $299 first smart display earns praise for fast home controls and useful kitchen features, but reviewers found gaps in how it keeps family members’ messages separate. Its strongest advantages also depend on owning compatible Apple devices and accessories.
Donald Trump proposed Apple executive chairman Tim Cook to succeed Jensen Huang at Nvidia. Cook’s interest and the seriousness of the suggestion remain unclear, while finance chief Colette Kress has taken interim charge.
Norwest’s largest fund is a third bigger than its predecessor, giving the Wells Fargo-backed investor more capital for AI alongside healthcare, consumer and industrial businesses. The firm has not disclosed how much will go to AI.
Nebius raised its 2026 capital spending forecast to about $27 billion while steering revenue to the low end of guidance. Better customer prepayments support expansion, but delayed capacity and a $325 million non-recurring contribution complicate its quarterly revenue beat.
Megaspeed founder Alice Huang is reportedly living in Beijing, but her activities and the espionage allegations remain unresolved. Jensen Huang’s resignation changes Nvidia’s leadership without settling questions about her business connections and access to its chips.
Singapore says it barred former Megaspeed chief Alice Huang after she gave inaccurate information about her role, while finding insufficient evidence against her over server declarations. The wider investigation still depends on tracing shipments beyond Malaysia.
Nvidia’s longtime finance chief takes over from its co-founder as the board promises continuity, searches for a permanent CEO and faces unresolved questions over security and executive conduct.
Chicago Public Schools is seeking a way to preserve Google Search without its generative AI features as parents, educators and students press for stronger district rules. Google’s planned access changes turn an AI safety dispute into a test of who controls classroom technology.
Warby Parker scheduled November 12 preorders and holiday deliveries for its first Android XR glasses. Google’s backing combines development funding with a conditional equity investment, while the retailer expects only a modest initial sales contribution.
