~/talk

Autonomous Quality Engineering: AI Systems That Generate Execute Heal Learn and Govern Quality

on YouTube 41m
Talkat TestMu Conference 2026 · 2026-08-19..08-21· [ai-assisted-development, machine-learning]

speaker

name:Jyotheeswara Reddy Gottam
links:none on file

resources

video:https://youtu.be/JxjC52CmOdk
slides:none on file

summary

A framework that actually runs: generative LLMs plus a multi‑agent coordinator use RAG and vector stores to draft tests, mutation testing to prove they would catch bugs, and cross‑model validation (generate with one model, verify with another). A prototype shifts a 600‑test suite to targeted regression of ~200 tests based on commit risk, layers self‑healing into L1–L4 fixes, logs every decision into a quality memory, and adds governance (identity, policy, audit). Results from two years of builds show RL improves learning but trust still depends on the surrounding systems, not a single chatbot.

AI-written summary · about

chapters (49)

  1. 0:00Start
  2. 0:05Session Opens - Autonomous Quality Engineering
  3. 0:35The Framework: Generate, Execute, Heal, Learn, Govern
  4. 1:06Speaker Intro - Jyotheeswara Reddy Gottam, Senior Software Engineer at Walmart Global Tech
  5. 3:08What This Talk Is: The Good, the Bad and the Ugly of Two Years
  6. 4:09"Can We Run More Tests?" Was Never the Right Question
  7. 5:10We Can Ship Faster - The Problem Is Building the Confidence
  8. 5:41Each One Is Useful; Together They Become a Loop That Improves
  9. 6:13The Three Daily Workflows in Quality Engineering
  10. 7:44Feeding All Three Into the Coordinator
  11. 8:45Generate: Where We Started Was "Prompt In, Test Out"
  12. 9:46Good Test Generation Is a Pipeline, Not a Prompt
  13. 10:17Draft From Everything, Then Ground It With RAG and Vector Stores
  14. 11:18Treat AI Output Like a Junior Engineer: Trust but Verify
  15. 12:18Before a Generated Test Joins the Suite, Make It Earn Trust
  16. 13:18Match the Intent - Is It Actually Asserting Anything?
  17. 13:51Use Mutation Testing to Prove It Would Catch a Real Bug
  18. 14:21Execute: The Suite Is a Portfolio, Not a Checklist
  19. 15:23Why You No Longer Run All 600 Tests Every Time
  20. 15:53Targeted and Risk-Based Regression in Practice
  21. 16:53Keep a Golden Set That Runs Regardless
  22. 17:23Classify the Failure Instead of Burning Compute on Reruns
  23. 17:53Heal: The Goal Is Not a Green Build - It's a Trustworthy Signal
  24. 18:25Why Trust Dropped When the Signal Moved From Humans to AI
  25. 18:58Level One and Two: Locators, Waits, Test Data and Environment
  26. 19:28Level Three: Test Code Itself - Don't Heal Away a Real Failure
  27. 19:59Before AI, Tests Failed Loudly - Now They Fail Quietly
  28. 20:29Level Four: Product Code - Requires a Human Review
  29. 21:01Treat Every Self-Healing Change as an Experiment
  30. 22:03Learn: Every Stage Produces Data With Real Signal in It
  31. 22:34Quality Memory - Not a Model Checkpoint, an Organisational Memory
  32. 23:35Production Should Improve the Next Round of Testing Every Time
  33. 24:05Govern: Not Rules to Make Your Life Harder
  34. 24:36Identity, Policy, Evaluation, Audit and Rollback Around Every Action
  35. 25:06The Thing Nobody Asked For: You're Now Testing AI Too
  36. 26:08The Quality Dimensions: Groundedness, Robustness, Non-Determinism
  37. 26:38The Maturity Journey - and Being Honest About Where They Are
  38. 27:40Autonomous QE Doesn't Remove the Human - It Changes the Work
  39. 28:10Measure Outcomes, Not How Busy the Agent Looks
  40. 28:41Four Measurement Areas: Flow, Effectiveness, Trust, Economics
  41. 29:43Autonomy Comes From the Loop, Not From One Chatbot
  42. 30:45Without Governance, It Creates Chaos That Costs More Than Doing It Right
  43. 31:17Q&A: Practical AI Security Checks for Every CI/CD Pipeline
  44. 32:47Q&A: How Agents Should Learn From Historical Production Incidents
  45. 34:52Q&A: Architectural Safeguards Against Hallucinated Test Steps
  46. 35:22Never Let the Model That Generated It Certify Its Own Output
  47. 36:23Cross-Model Validation: Generate With One, Verify With Another
  48. 38:24Q&A: Telling Genuine Regressions From Environment-Induced Noise
  49. 39:24RL Makes It Easier - Trust Still Comes From the Systems You Build

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.