Inspect the application, not just the source
Software Mansion describes Argent as a toolkit of skills and an MCP server that connects coding assistants to mobile simulators and emulators. Its capabilities include application interaction and performance investigation. The project also describes recording React and native iOS profiles together to investigate slow commits.
That makes it worth evaluating when a bug is easy to see but difficult to explain from code alone. A delayed tap, an obstructed control or a stuttering list needs evidence from the running application.
Begin with one reproducible interaction
Choose a short scenario: open a product list, scroll to a known item, open its detail screen and return. Write down the build configuration, platform and test data. Run it without changing the code first.
Keep the investigation focused on the observed symptom. If the issue is a delayed screen transition, start with that transition. Changing every animation and list setting at once makes it difficult to attribute any improvement.
Use a trace to choose the next step
A profile should help narrow the question. Does the slow interaction coincide with repeated React work? Is the UI waiting for data? Does an expensive native operation happen at the same time? Treat these as hypotheses to test, rather than conclusions drawn from a single frame.
After a change, repeat the same interaction with the same data. Compare the trace and the visible result. If the improvement disappears when you repeat the run, the fix is not yet established.
Know the limits of the environment
An emulator is useful for reproducing behavior, but it does not establish how a lower-powered phone will perform. Before release, repeat the critical path on a physical device and use a release build when assessing user-facing performance.
Keep test accounts and synthetic records available for tool-assisted investigations. Give the assistant a bounded task and review changes before they reach a shared environment. Check the project's current documentation for supported platforms and setup requirements.
The useful outcome is a small, explained correction with evidence. That fits the same approach we use for quality and security reviews: reproduce the problem, change the relevant behavior and verify the result.