~/talk

Workshop: The Full Agentic QA Loop for Mobile Applications

on YouTube 57m
Talkat TestMu Conference 2026 · 2026-08-19..08-21· [ai-assisted-development, testing]

speaker

name:Shravan Mahajan
links:none on file

resources

video:https://youtu.be/SMPZuV2Eguk
slides:none on file

summary

Workshop demos an end-to-end mobile QA loop using Kane CLI driving an iOS simulator to ingest a PRD, generate a context graph, and auto-produce tests: it extracts four use cases, creates seven acceptance criteria, two scenarios and two runnable tests, runs a first LLM authoring pass then replays with zero credits, auto-corrects input (email), captures a portable .evidence pack with actions/network/assertions, and reports 71% coverage of designed criteria for the single test run.

AI-written summary · about

chapters (66)

  1. 0:00Start
  2. 0:20Why Most AI Testing Demos Stop at Generating a Test
  3. 0:36The Loop: Author, Evidence, Assurance, Maintenance
  4. 0:51Speaker Intro - Shravan Mahajan, Software Engineer at TestMu AI
  5. 1:06Housekeeping: Hashtag, Recordings and the YouTube Channel
  6. 2:22This Session Moves the Same Loop to Mobile
  7. 2:38What Kane CLI Is: Context and Test Generation in the Terminal
  8. 3:09AI Writes Almost All the Code - So Who Validates It?
  9. 3:24The Bias Problem: A Harness That Bends to the Implementation
  10. 3:55An Agent That Rewrites the Test When the Build Changes
  11. 4:58Why the Implementation Can Differ From the Spec
  12. 5:13Design Tests, Run Tests, Then Check for Drift
  13. 5:29The Verbs: Ingest a Context, Design From It
  14. 5:59Scenarios and Acceptance Criteria, Not Just Test Cases
  15. 6:30Why Mobile Is Harder Than Desktop
  16. 7:32The Demo App: A Static V16 Food Ordering App
  17. 8:33Prerequisites: Mac Silicon, Install and an OAuth Login
  18. 9:34A Pause While the Audience Sets Up
  19. 11:14Cloning the Public Repo With the App Build and Sample PRD
  20. 12:24Touring the Repo: The app Folder and the docs Folder
  21. 12:55The Sample PRD That Will Drive Everything
  22. 13:45Ingesting One PRD Into a Context Graph
  23. 14:17Why the Graph Exists - Don't Re-Derive What You Already Know
  24. 14:47Handling a New Version of the PRD Without Redoing Everything
  25. 15:03Running the context ingest Command Live
  26. 15:33Where the Context Directory Lives in Your Own Repo
  27. 16:19Watching the Agent Think as It Generates Use Cases
  28. 16:34What a Use Case Is - an Independently Testable Entity
  29. 17:38Four Use Cases Extracted From the PRD
  30. 18:09Each Use Case With Its Description and Criteria
  31. 19:14Designing Tests From a Single Use Case
  32. 20:00Re-Reading the Source in Case the PRD Has Drifted
  33. 20:30The Design Phase Produces Criteria and Scenarios First
  34. 20:45What an Acceptance Criterion Is - an Individually Verifiable Step
  35. 21:16A Negative Criterion: Wrong Password Raises an Error Prompt
  36. 22:01Why a Short PRD Yields Vague Criteria - and a Verbose One Doesn't
  37. 23:19The Agent Asks About Test Credentials From the PRD
  38. 24:06Reading the PRD's Screen-by-Screen Detail
  39. 24:56Two Scenarios: Blocked Sign-In and Successful Sign-In
  40. 25:11A Second Pass to Check Each Scenario Stands Alone
  41. 26:16Three More Acceptance Criteria at the Scenario Level
  42. 26:46Authoring the Runnable Tests Themselves
  43. 27:02The Agent Asks What Starting State to Design For
  44. 28:00Two Tests Authored - One Negative, One Positive
  45. 28:34Session Totals: Seven Criteria, Two Scenarios, Two Tests
  46. 29:34Reading the Generated Test File and Its Presence Checks
  47. 31:52The TUI: Selecting a Device and App From the Terminal
  48. 36:50Installing the App and Killing a Stray Background Session
  49. 37:21The First Run Is One LLM Pass That Authors a Replayable Test
  50. 38:17The Agent Auto-Corrects a Stray Full Stop in the Email
  51. 39:30All Four Steps Pass - On to the Evidence
  52. 40:38Inside the Evidence: Autoplay of Every Action Taken
  53. 41:08LLM-Powered RCA, Shown Only When Something Fails
  54. 41:40Coverage Mapped Back to the Designed Acceptance Criteria
  55. 43:41Installing Kane CLI as a Skill Inside Claude Code or Cursor
  56. 44:11Same Commands for iOS and Desktop - Nothing Platform-Specific
  57. 45:27Q&A: What an Annotated Screenshot Can and Cannot Tell You
  58. 45:57Q&A: How Autonomous Agents Should Handle Flaky Mobile Tests
  59. 46:44Q&A: Desktop Apps and System-Level Interactions
  60. 47:44Q&A: Feeding Visual and Console Errors Back to a Terminal Agent
  61. 48:46Q&A: Describing Custom Gestures Without Introducing Ambiguity
  62. 49:33Q&A: How Kane Differs From Other AI Testing Platforms
  63. 50:36Q&A: Measurable Advantages in Accuracy, Reliability and Coverage
  64. 51:52Q&A: What Happens If Kane CLI Crashes Mid-Run
  65. 52:38Q&A: Looping Individual Tests vs Running the Whole Suite
  66. 54:14Q&A: Which Evidence Sections Appear Only on Failure

related talks

~/search
type to search 6,249 conferences, 138,987 speakers and 185,172 talks
~/help
cmdctrl k
search every conference, speaker and series
/
focus the filter search
h
home
c
conferences
s
speakers
e
series
t
light / dark — the status bar button does the same
?
this help
esc
close this, then clear every filter — or just the search, while you are typing in it

Ignored while you are typing. Pages without a filter rail have no search to focus.