Expo Simulators ships cloud phones for agents at $6 an hour

Expo shipped cloud simulators for coding agents and the Hacker News thread got 24 points.

Twelve comments. One of them, word for word: "wow this is awesome !" with a space before the exclamation mark. Another says "my agents are going to love this."

The whole thread reads like people being polite to whoever posted the link.

That got under my skin. Not because the product is weak. Because this is the missing piece for agentic mobile work, and the room yawned.

On November 12 Expo made Expo Simulators generally available. It hands your agent a cloud iOS Simulator or Android Emulator on demand. No Mac, no nested virtualization, no seven simulators chewing through your laptop's RAM.

The agent installs a build, taps through the app, and comes back with a recording.

And it bills at $0.10 a minute. That number is most of the story, and i'll get to it.

What you actually type

npx eas-cli@latest simulator --platform ios

That one line starts a session and prints a link to a live view. You can also hand your agent a skill and let it work out the flags:

npx skills@latest add expo/skills --skill eas-simulator

The agent drives the remote device with the same commands it uses locally. agent-device is the default controller. argent and appium are options too. If you only want to look at a build yourself, --type web-preview-only gives you a stream with no controller attached.

i expected this to be the hard part. It isn't. The hard part is everything after.

Why local simulators fall apart

i used to think running agents locally was fine. It is, until you run more than one.

Two agents on the same machine fight over the same simulator. One reinstalls the app while the other is mid-test. You get a failure that has nothing to do with the code.

Expo names this directly. Each session gets its own VM, so parallel agents never share a device.

The comment i keep coming back to is from colinhunt on that thread: "Agents on my computer were fine to run in a normal dev loop but I really needed this for the background agents."

That's the real use case. Not your laptop. The agent in a cloud sandbox at 2am that has no device at all.

What the first session looks like

You run the command and it prints a URL. You open the URL and there is an iPhone, already booted, inside a browser tab.

No Xcode download. No 40-minute first launch. No spinning wheel while your fans ramp up.

I haven't run a session myself yet, so take the rest of this as what the docs and the demo show. The session page keeps a timeline of the agent's actions. Tap here, type there, wait three seconds.

Here's the question people always ask: does it replace a real device? No. Simulators never do. They miss push notifications on a locked screen, camera quirks, and anything that depends on a real network handoff.

But it replaces the part where you build the app yourself just to look at it. That's most of the time.

The Mac claim is half true

Someone asked if they could now build iOS apps from Linux. The reply came back cheerful: yes, your Linux agents can verify against a remote iOS simulator.

Then lukisk asked the obvious question. How do you build an iOS app without macOS and Xcode? Vendor lock, right?

He's right, and the thread moved on without answering him.

you still need a Mac somewhere, or EAS Build, to produce the .app in the first place.

Expo Simulators removes the simulator, not the compiler. That's still a big deal. It's just not the thing the headline implies.

What a session actually keeps

This is the part that sold me, and it's buried under the infrastructure talk.

Each session records:

• video of the whole run
• screenshots at any point the agent takes one
• app and native logs
• network traffic
• CPU and memory usage

Artifacts stay for 30 days. Your agent can read its own logs and attach screenshots straight to the pull request.

That changes review. Instead of 500 lines that look about right, the reviewer watches a 40-second clip of the thing working. Nobody pulls the branch. Nobody builds it locally.

Expo's own demo is an agent that ran the app at the largest text size, saw labels overlapping, fixed the layout, and attached before and after shots. A real accessibility check, done without a human opening Xcode.

The pricing math

Here's where i stopped nodding.

$0.10 a minute is $6 an hour. For one verification run, that's nothing. Cheaper than your time.

But the entire promise is that agents check their own work. Constantly. A loop that verifies 40 times in an afternoon is $24. Spread that across a team and a CI pipeline, every day, and the meter adds up.

Paid plans include 60 simulator minutes a month until 2027. Sixty. That's ten hours of agent time per month, which is roughly two afternoons of a busy agent.

There's also a cap. The blog says no session runs longer than two hours. The pricing page says 115 minutes. Small gap, but you notice it when you're building a budget.

A short detour about the phone drawer

Every mobile team has a drawer. Old iPhones, a Pixel with a cracked screen, a tablet nobody remembers buying. It's the QA department, held together with tape and hope.

You know the one. Somebody's old phone, still charged, still on the wifi, labeled with a sticky note that says "do not reset."

The drawer exists because simulators lie. They're fast and clean and they never have a low battery or a notification covering your button.

Cloud simulators don't fix that. They just move the lie to a machine you're renting.

Who should skip this

If you have one agent, on your own Mac, in a normal dev loop, you don't need this.

Local simulators are free and already on your machine. Expo says so itself, near the bottom of the announcement.

The second group that should wait: anyone on the Free plan. Simulator minutes there still say "coming soon." Access starts at Starter, $19 a month. There's no free tier to try it, so you pay to find out if it fits.

And if your app is a simple form over a REST API, the agent can verify most of that without a device at all. You're buying a screen recorder for bugs you don't have.

The thing i keep thinking about

Twenty four points.

Expo built infrastructure that a dozen agent startups will end up depending on, and the internet replied with a shrug and a typo.

Maybe that's just Hacker News now. Launches don't get a reception anymore, they get a log line.

But i keep thinking about that Linux question nobody answered. The person asking wanted to build iOS apps without a Mac. What they got was a way to test them.

Half a product is still a product. It's just not the one the headline sold.