Bringing Mobile App Validation into the Agentic Loop
summary
Every step is concrete: an agentic CLI called Kane CLI drives iOS simulators and Android emulators from the terminal, navigates hybrid native/WebView flows without a DOM, and authors deterministic tests. One objective (full checkout) is authored and proven under three minutes, the PRD-to-proof loop generates 18 tests, 46 acceptance criteria and records gaps, and live runs surface real failures (password minimum and invalid credentials). Artifacts include intent-based steps, screenshots and an evidence pack.
AI-written summary · about
chapters (29)
- 0:00The Blind Spot: Every Testing Agent Drives a Browser, Not a Phone
- 1:39Speaker Intro - Shantanu Wali, Director of Product, TestMu AI
- 2:40Coding Agents Already Launch iOS Simulators - Why Is Validation Still Manual?
- 4:42Why Mobile Is Structurally Harder: Slow Authoring, Selectors That Don't Hold
- 5:42Coverage Explosion: Hundreds of OEMs, Models and OS Versions
- 6:43Introducing Kane CLI for Mobile - Driving Simulators From the Terminal
- 7:43Live Demo: One Objective, a Full Checkout Flow on a Simulator
- 9:15Intent-Based Steps in Natural Language, Not Vision-Only Guessing
- 11:55Objective Met in Under Three Minutes, Artifacts Saved
- 12:56Why Mobile Is Different: No DOM, Hybrid Surfaces, Device Matrix
- 14:27One Objective Isn't a Loop - What's Still Missing
- 15:27Introducing Assurance: One Loop From Requirement to Proof
- 16:30Live Demo: Kane CLI as a Skill Inside Claude Code
- 18:02Ingesting a PRD and Deriving Use Cases First
- 19:03Why Feeding the Codebase as Context Biases the Test
- 20:36Reviewing and Approving Use Cases - Human in the Loop
- 21:40Designing Tests From the PRD and Approved Use Cases Only
- 23:1118 Tests, 46 Acceptance Criteria and Recorded Gaps
- 24:41Authoring the Tests Live on the Simulator
- 25:44Real Failures Found: Password Rule and Invalid Credentials
- 26:44The Evidence Pack: Steps, Screenshots and Issues
- 27:44Coverage as Acceptance Criteria, Not Test Count
- 29:47Two Axes of Coverage: Designed vs Proven
- 30:49Takeaways: Mobile Is in the Agentic Loop, Context Is Modelled
- 33:23Q&A: Token Economics of Repeated Agentic Runs
- 35:53How Kane CLI Sees the App Without a DOM; Web View Context Switching
- 37:58Simulators vs Real Devices and CI Best Practices
- 41:00Ride-Hailing Apps, GPS and Sensor Mocking
- 42:02Reusing Existing Appium Test Code as Context
related talks
18mReinventing App Testing with Agentic Apps
19mThe GPT-ification of Mobile App Testing
18mCloud agents that test mobile apps on real devices
11mDragonCrawl: Revolutionizing Mobile Testing with GenAI
47mAI + Dev Environments + CI/CD for Mobile Teams
40mDigitizing Testers
18mThe GPT-ification of Mobile App Testing
35mApp Testing with AI Tools – Make it Easy