Pinned Tweet
Stop manually testing your app before every release
We built agentic QA that just works:
- No code, no locators, no SDK.
- Supports iOS and Android, any framework (Native, React Native, Flutter).
- Works with local and cloud real devices.
Install the Mac app, plug in your phone, watch agents do the work
Wanna try? Comment "INTERESTED" and I'll send you free early access
Locators and AI vision are not fighting for the same job.
Teams debate which one is better as if one tool must win. But this is not a battle of performance. It is a boundary of coverage.
Locators lookup elements in milliseconds on stable trees. Where your UI structure is predictable, locators work fine. Keep them.
Vision takes over where trees do not exist: Unity games, canvas UIs, Apple Pay popups, and OS system alerts. A locator simply cannot query what is not in the DOM.
Strong teams do not pick a camp. They keep locators for standard views and use vision to cover the screens that used to be strictly manual.
How much of your mobile app is still tested manually simply because traditional locators cannot read those screens?
roke down the full comparison matrix and technical trade offs here:
noqa.ai/blog/locator-based-v…
On backend & web, AI agents finish the job. On mobile, they were stuck behind the touch screen.
The noqa CLI gives them hands.
Now Cursor and Claude Code can connect a real device, read the screen, and drive it by plain-English descriptions instead of flaky selectors.
The agent that built your screen can finally verify if it actually works.
Grab it in the noqa macOS app: Settings → Tools
The full story of why we built it, and how the agent reads the screen: noqa.ai/blog/noqa-cli/
If anyone needs real iOS/Android devices or Android emulators in your Claude code – use github.com/noqa-ai/noqa
Point your coding agent at it: github.com/noqa-ai/noqa
Simulators lie. In-app purchases behave differently, real hardware performs differently.
Your coding agent should be testing on the real thing, from the terminal.
Your coding agent can build the app. It still can't test it on a real device.
So we built a CLI for noqa. Your agent drives a real device straight from the terminal.
• Plain commands in, screenshots out
• Real devices, not just simulators
• Point at an element, noqa taps it
Try it now.
More on GitHub: github.com/noqa-ai/noqa
I keep seeing the same thing in Flutter teams: pre-release testing becomes the bottleneck.
Flutter apps are testable, but the usual automation path is still harder than it should be because the UI is rendered differently from native apps. That's what makes pre-release testing slower and more manual than most teams want.
That's the problem we wanted to solve with our visual agent. It doesn't care about the element tree - it looks at the screenshot and acts on what it sees. That means there's no need to add accessibility IDs or other test-specific setup. You just describe what you want in plain English, and it does the rest.
Noticed a pattern: new agentic mobile QA tools are launching simulator-only.
Surprised me - every team we've talked to building noqa says they need real devices. Not because simulators are useless, but the scary pre-release stuff doesn't show up there. IAPs behave differently. Real hardware performance differs.
Building noqa, we had to choose: simulators only, or real devices too?
We went with both. Simulators for fast dev feedback. Real devices before release, when you need to know things actually work.
Shipping a broken paywall is the most expensive bug in mobile
The scariest bug you can ship isn't a crash. It's a broken purchase flow that nobody notices for days - while revenue bleeds out quietly, ad campaigns keep spending, and algorithms train themselves on data from users who were trying to buy but couldn't.
Most teams handle it the same way - manual testing before every release. One paywall is manageable, but they multiply - web funnels, subscription tiers, separate flows for different segments. It quietly becomes one of the bigger parts of the regression cycle.
We built noqa to automate exactly this. An agent that runs through purchase flows on its own - paywalls, system dialogs, subscription confirmations. You describe the goal in plain English, it does the rest.
The kind of thing that lets you ship on Friday without checking Slack all weekend.
Here's what it looks like in practice ↓
We’re building noqa.ai specifically to address the last bottleneck in mobile app development: testing. Now you can run agents to test your app on real devices.
xitter.cf/romainhuet/status/2053…
