AI Agent Use Cases

Added today

11All 202

Trying Codex Automated QA for a Flutter App

The author reports that after integrating Codex automated QA into a Flutter app under development, slow tests and unintended changes to features became problems.

Codex
Dev

Making a 3D Cube Net App for Children's Math Study

Grok Bot reportedly created a 3D app where children can fold all 11 cube nets from their textbooks, which are usually shown only as flat diagrams, using their fingers.

Grok Bot
Learning & Careers

1–20 of 202

Dev

Trying Codex Automated QA for a Flutter App

The author reports that after integrating Codex automated QA into a Flutter app under development, slow tests and unintended changes to features became problems.

How it works

  1. The author applied Codex automated QA to the VibeCheck app being developed with Flutter. Because it was a Flutter app, the tests manipulated the emulator screen using computer control features.
  2. There was a conflict between the computer control feature of the Codex web interface and the app, and even creating a separate control structure did not easily resolve the issue.
  3. When a problem was found during QA, the AI sometimes made its own decisions and changed features that were originally intended to remain unchanged.

Results

The author reports that testing became a bottleneck for the entire development process, and the final product could differ from the original intent. The author considered the third issue to be the most critical.

Notes

There are no specific details or figures about test duration or the exact features that were changed incorrectly.

Source

X Original post ↗

Example prompt · Codex

Attach AI automated QA to a mobile app under development, but create a testing workflow that prevents the AI from changing intended features while fixing bugs.

Input: Repository [repository URL], app framework [e.g., Flutter], list of screens and features to test [list], behaviors that must not change [e.g., payment button position, recommendation algorithm]. First, check how the current AI coding tool supports emulator or screen control, and separate what can be replaced by screen control tests and what by code-level tests.

Order: 1) Document the expected behavior for each feature and approve it. 2) In the QA phase, only report problems found, do not fix code. 3) Make fixes in a separate branch, and if there are changes different from the approved expected behavior document, indicate them at the very top of the PR description. 4) For slow screen control tests, keep only a few core flows and move the rest to code-level tests. 5) To prevent multiple agents from conflicting by manipulating the screen at the same time, run only one at a time.

Deliverables: Expected behavior document, QA report template, test execution time record. First, try with one screen and show the test time and whether any AI-suggested fixes differed from the intended behavior.