Panel Discussion: Scaling Enterprise Practice in the Agentic Era
summary
Enterprise pilots that looked impressive often fail to scale because operational details break first: one case creates 1,500 epics and 15,000 user stories plus test assets with 100% traceability and 70–80% reusable automation but then test data provisioning became the binding constraint. Panelists outline five signals that a pilot will never reach production, recommend a hub-and-spoke governance and continuous, four-layer AI assurance, and note commercial models split into human cost plus AI token cost.
AI-written summary · about
chapters (40)
- 0:00Scaling Enterprise Practice in the Agentic Era
- 1:41Panel Intros - Mhahesh Muraleedhara, Dror Avrilingi, Vikul Gupta, and Richa Agrawal
- 2:13The Hardest Assumption to Kill: More Scope Meant More People
- 3:49Capacity No Longer Equals Headcount
- 4:52What That Does to the Testing Pyramid and the Commercial Model
- 5:53Pricing Splits Into Human Cost Plus AI Token Cost
- 7:28A Real Enterprise Example: 1,500 Epics and 15,000 User Stories
- 8:30Test Assets Written Before Quarterly Planning Even Finished
- 9:31What Broke: Test Data Provisioning Became the Binding Constraint
- 10:02Every Bottleneck Cleared Revealed the Next One
- 11:03AI for IT vs AI for Business - Two Buckets of Adoption
- 11:33The Airline Chatbot Case: 12 Domains, 15 Backend Systems
- 12:36Testing What the Agent Is Not Supposed to Do
- 13:37Bringing the Focus Back to AI Assurance and Observability
- 15:37Five Signals That a Pilot Will Never Reach Production
- 16:38Signal 1: A Real Workflow and Measurable Business Outcome
- 17:08Signal 2: Reliable Context - the Fastest Way for a Pilot to Fail
- 17:38Signal 3: Verification Designed From the Beginning
- 18:41Signal 4: A Clear Operating Model and an Owner After Deployment
- 19:11Signal 5: What Happens Beyond the Pilot
- 21:15Does the Organisation Trust It Enough to Delegate?
- 23:51The Sponsor Test: Can They Change the Definition of Done?
- 24:54Where Is the Budget Coming From - Innovation or Delivery?
- 25:25Is Anyone Measuring Things That Could Embarrass Them?
- 27:26Rolling Out Across Uneven Maturity: The Multi-Speed IT Analogy
- 27:56A Shared Enterprise Foundation Plus Graduated Autonomy
- 29:28Autonomy Earned Through Evidence, Governance Federated Hub and Spoke
- 31:36Building the Workflow, Not Just the Use Case
- 32:39Trust Signals: Accuracy, Consistency, Error Rate, Predictability
- 34:44Explainability and How to Trust Someone Else's Model
- 35:16The OpenAI Hugging Face Incident as a Watershed Moment
- 36:18Continuous Validation of Decisions, Policy Adherence and Tool Usage
- 38:22LLM as a Judge and Grouping Metrics Into Truth, Evidence, Clarity, Auditability
- 40:25The Four-Layer AI Assurance Pyramid
- 41:26Where Enterprise QE Moves in the Next Three Years
- 42:29Chief Quality Officers and Trust Engineers
- 44:03The Apprentice Model Disappears - Pyramid Becomes a Diamond
- 45:37Engineering Trust as the Next Decade's Discipline
- 46:40We May Stop Calling It Testing Altogether
- 47:40Continuous Trust Engineering as the New Practice Name
related talks
59mPanel Discussion: Strategies for Scaling AI in Quality Engineering
20mFrom Demos to Production: Why Agentic AI Breaks at Enterprise Scale
56mPanel: Scaling AI successfully
32mScaling a Responsible Agentic Workforce
01h 02mThe Enterprise DevOps Challenge
19mScaling & Industrialization of AI
21mMost Enterprise Agentic Projects Are Doomed, Here's Why
16mLaunching AI Pilots: Insights for Startups and Enterprise