QApilot

QApilot is an AI-native, autonomous testing platform built for mobile apps, helping teams create, execute, and maintain tests with far less manual effort.

[QApilot @QApilot]

Agent translates your whole app into German in 8 minutes. Every screen. Looks done.

Then someone opens it on a small phone.

Checkout button says "Zur Kas". Settings cut off mid-word. Error message runs off the screen.

Nobody looked at the German version before it shipped.

[QApilot @QApilot]

@svpino @vercel Most of that trust gap shows up after the code leaves your laptop. It passed everything on a fresh simulator, then someone opens it on a 4 year old iPhone on train wifi and the login just spins. Nobody wrote a test for that.

[QApilot @QApilot]

https://t.co/zhdo0YeGco

[QApilot @QApilot]

Saw thousands of people this week saying their phone dies by 3pm.

That's who opens your app. Low Power Mode on, 12% battery.

You tested on a new phone, fully charged, office wifi.

On theirs the background sync just doesn't happen.

Maybe stop charging the test phone.

[QApilot @QApilot]

@EXM7777 Add one more question: where did it say done when it wasn't. Those spots are the checks worth writing first. Usually it's the same few every day. Never actually ran it. Only tried the happy path. Looked fine on a big screen, broke on a small phone.

[QApilot @QApilot]

@eptwts The testing part is where it shows.
An internal tool only has to work for you, on your laptop, on good wifi.
A real product gets someone on a 4 year old Android with the font cranked up, tapping pay twice on a bad signal.
None of that comes up while you're prompting.

[QApilot @QApilot]

We run that check from the IDE now: https://t.co/pZVNpONqSu

[QApilot @QApilot]

Coding agent marks the mobile change done.
PR is green. Merge is one click away.

Nobody opened the app on a real phone.

On a cheap small phone the pay button hides under the keyboard. You only catch it at checkout.

Still wondering who verifies what it shipped.

[QApilot @QApilot]

@ChShersh Honestly that's most of testing. I don't trust a test until I've watched it fail once. And I don't trust "it works" until it's run on the oldest, cheapest phone we have.

[QApilot @QApilot]

Foldable phones mean new challenges for mobile app testing.

From split-screen layouts to fold-and-unfold transitions, can your app handle it all without breaking critical user journeys?

Here's what QA teams need to know about testing for iPhone Duo https://t.co/FHb25mAH2Q

[QApilot @QApilot]

20x faster. 153x cheaper. But less accurate.

We tested Typesafe AI’s JEV for on-screen classification to see what happens when you optimize for speed and cost.

The results? An interesting trade-off worth exploring.

Read our findings :
https://t.co/0pc0xuN8BC

#AI #MobileTesting #QualityEngineering

[QApilot @QApilot]

@Yumzlef Those 2 rewrites nobody asked for are the sneaky ones. They still pass, so green tests never flag them. Simple check: list the files each task should touch, then compare with the actual diff. Anything outside that gets a human look. That email edit would have popped right away.

[QApilot @QApilot]

Android 17 stops letting apps force portrait on tablets and foldables.

So the phone app you only checked upright now opens wide.

Buttons stretched across the screen. Lists with big empty gaps. Forms floating in the middle.

When did your team last open the app on a tablet?

[QApilot @QApilot]

@omarsar0 Same input, same verdict is most of what you want from a judge. I'd also hand it a short list of things it can check on the actual screen, like did the button state change, instead of asking if the run looks good overall.

[QApilot @QApilot]

@EXM7777 Faster browser control helps. I'd still open the page after the agent says it finished the flow. The step log can look clean while the UI never moved.

[QApilot @QApilot]

@jamonholmgren A linter that fails without saying why just burns tokens and looks like progress. I'd fix those harness error paths before adding another skill.

[QApilot @QApilot]

@PaulSolt @twostraws Agents keep reaching for deprecated SwiftUI APIs like nothing changed. This skill is a good stop. I'd still tap the screen once on a real phone after, since skill checks miss the layout that only breaks when you rotate.

[QApilot @QApilot]

@omarsar0 Yep. Green on PR A, B, and C means nothing if checkout still fails once all three land. The useful review is the joined path, not each diff alone.

[QApilot @QApilot]

We run that proof on your phone and hand the agent a report. https://t.co/pZVNpOMT2W

[QApilot @QApilot]

Claude Code can drive a real Android phone over a cable.
Same job on a real iPhone is a signed runner, a tunnel, and a setup that dies when the phone locks.

Your agent shipped the change. Who proved it on the device?