~/talk

The Copilot Wrote It but Who Saves It: Surviving the AI Code Tsunami

on YouTube 31m
Talkat TestMu Conference 2026 · 2026-08-19..08-21· [ai-assisted-development, engineering-leadership]

speaker

name:Mathias Strziga
links:none on file

resources

video:https://youtu.be/tk4fMxTfRFU
slides:none on file

summary

AI-generated code arrives in bursts not sprints, so QA must become Validation Pilots with decision templates, not line-by-line reviewers. The AI Development Lifecycle runs 82 skills, 9 agents and 6 workflows that shape, build, validate and learn; each step needs scope, evidence and stop conditions. Use Jira and TestMu as event streams, not Git repos, and treat failures as classification (product, spec, test, infra or unknown stops the chain) to decide whether to fix or teach.

AI-written summary · about

chapters (58)

  1. 0:00Start
  2. 0:05Session Opens - The Copilot Wrote It, but Who Saves It?
  3. 1:05We Gave Developers Turbocharged Engines; QA Is Still on a Two-Week Sprint
  4. 1:36Speaker Intro - Mathias Strziga, Director of QA, Apps Factory Group
  5. 2:07Introducing the AI DLC - Who Understands It and Who Changes It
  6. 2:37About Apps Factory: 450 People, Mobile-First, Now Full Service
  7. 3:10Wearing Every Hat With "Quality" In It
  8. 3:40Building the QA Department From Nothing Over Seven Years
  9. 4:10Background: BMW, Audi, Mercedes-Benz, Carriad and the Aldi Group
  10. 4:43The Analog Hobby: Restoring Classic Cars
  11. 5:13Disclaimer: This Talk Will Look Different by Christmas
  12. 5:43The Legacy SDLC Curve - Effort Peaks in the Middle
  13. 6:15Code Got Cheap, So Specification and Intent Got Expensive
  14. 6:46Bursts, Not Sprints: Specify Heavily, Build in Three Days, Then Verify
  15. 7:17The Framework Today: 82 Skills, 9 Agents, 6 Workflows
  16. 7:50Three Phases: Specification, Implementation, Maintenance and Operation
  17. 8:20The Copilot Is No Longer Alone - and Neither Are We
  18. 8:50The 1979 IBM Slide: A Computer Can Never Be Held Accountable
  19. 9:20Implementation Has Never Been More Out of Our Hands
  20. 9:51The Thought Experiment: What If Output Really Goes 10x or 50x?
  21. 10:22QA's Job Isn't Only Control - It's Trustworthy Evidence and Ship Decisions
  22. 10:52Why the Act of Learning Itself Is Now the Interesting Part
  23. 11:23Defining Right From Wrong and the Measured Response to a Defect
  24. 11:55The Binary AI Jobs Debate - and What the Evidence Actually Shows
  25. 12:25Learning Is Not a Soft Skill - It's the Operational Layer of Decision-Making
  26. 12:57Where AI Falls Short: Complex Systems With No Definite Answer
  27. 13:27Job Postings Now Demanding Judgement and Accountability
  28. 13:57Half of Postings Ask for Problem Framing - QA's Home Turf
  29. 14:27The Scaffold as the Governance Backbone
  30. 14:57Classifying Findings Before They Become Repairs or Lessons
  31. 15:27Building Self-Doubt Without Poisoning the Context Window
  32. 15:59Deciding What Is a Real Lesson vs a One-Time Incident
  33. 16:31Every Step Needs Scope, Evidence, Requirements - and Stop Conditions
  34. 17:01Parallelisation Only Helps If Scope and Evidence Travel With the Work
  35. 17:32Walking the Agents Through a Single Button
  36. 18:03The Build Agent Also Asks How It Would Be Tested
  37. 18:33The Scaffold Separates Builder Context From Challenge Context
  38. 19:05Developers Ask How to Build It; Testers Ask How to Break It
  39. 19:37The Tester Walks Into a Bar Joke - and Why It Matters Here
  40. 20:07Six QA Personalities Inside the Validation Cluster
  41. 20:38Validating the Validation and Challenging the Review
  42. 21:10Validating Against Intent, Not Just Code
  43. 21:41Product, Behaviour, Spec, Test Issue, Infrastructure, Harness or Unknown
  44. 22:12Unknown Stops the Chain and Puts a Pilot in the Seat
  45. 22:42The Problem With Repositories: Git Only Looks Backwards
  46. 23:12Why You Can't Keep Every Spec and Intent in the Repo
  47. 23:43Jira and TestMu as Event Streams and Buffers for the Swarm
  48. 24:15Third-Party Dependencies With Their Own Deployment Cadence
  49. 24:46The Future Challenge Is Convergence, Not Speed
  50. 25:16Thinking, Fast and Slow - The Two Mental Systems
  51. 25:47System One Is Fast and Lazy; System Two Costs Real Energy
  52. 26:19Our Context and Experience Window Is Far Larger Than the AI's
  53. 26:50The Client Who Couldn't Define Quality but Knew When It Was Missing
  54. 27:20Why That's Exactly How You Have to Address an AI System
  55. 27:51Combining Human Instinct With Agent-Generated Decision Templates
  56. 28:57Q&A: Preventing QA Validation From Becoming the Bottleneck
  57. 29:27Teach the System the Concept of Validation Instead of Reviewing Every Line
  58. 29:58Conduct the Orchestra; Don't Play Every Instrument

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.