The New Skill Stack for SDETs in the Agentic Era
summary
SDETs already keep automation frameworks, risk models, CI/CD, debugging taxonomies and code review; the talk argues those skills stack into four new, concrete abilities: prompt engineering as a versioned prompt library (start with ~10 prompts), data curation as a labeled corpus tied to risk tags, model literacy that maps failures to mechanisms (temperature, retrieval, context window), and observability that captures prompts, tool calls and decisions. Repetitive test-writing shrinks; instrument one agent, run an eval in CI, and track the cost.
AI-written summary · about
chapters (41)
- 0:00The New Skill Stack for SDETs in the Agentic Era
- 1:18Speaker Intro - Adeel Mansoor, Director Engineering, Quality at Centerfield
- 1:48The Magic Wand Question - and Why It Isn't Hypothetical Anymore
- 2:18Three Questions: What Shrinks, What Magnifies, What You Stretch Into
- 2:48The Story: An AI-Built Chatbot and the Question Nobody Asked
- 3:48The Person Who Caught It Knew Software, Not Models
- 4:51The Whole Talk in One Picture: Four Columns on a Slab
- 5:52Skill One - Automation Engineering Was Never About Writing Scripts
- 6:54Kill Your Flakiness and Give Your Framework a Socket
- 7:24Skill Two - Test Design and Risk-Based Thinking, the Most Undervalued
- 7:54Get the Risk Model Out of Your Head and Onto Paper
- 8:26Skill Three - Debugging: Refusing to Stop at a Symptom
- 8:56The New Problem: Confident AI-Generated Root Causes That Are Still Wrong
- 9:58Write Down Your Failure Taxonomy
- 10:30Skill Four - CI/CD Is an Information Problem, Not an Infrastructure One
- 11:31Skill Five and Six - Collaboration and Quality Advocacy
- 12:36Reading Code You Didn't Write and Forming a Fast Opinion
- 13:07The Whole Industry Is Arriving at the Condition You Already Live In
- 13:38Risk-Based Sampling When You Can't Read Everything
- 14:08The Honest Part: What Actually Shrinks
- 15:19The Stretch Half - Four New Skills
- 15:50Prompt Engineering for Testing Built on Automation Frameworks
- 16:50A Prompt Library Is a Framework, Not a Pile of Scripts
- 17:50Data Curation Built on Test Design and Risk Thinking
- 18:22Equivalence Partitioning Is Corpus Design
- 19:22Model Understanding Built on Debugging - Lower the Bar
- 20:54Don't Accept "It Hallucinates" as a Root Cause Either
- 21:57A Real Visual Testing Bug and the Context Window Overflow
- 23:00Observability for Agents Built on CI/CD and Fast Feedback
- 24:33A Wall of JSON Is Not Observability
- 25:03The Four Skills Add Up to One Artifact: An Eval
- 26:03An Eval Is a Test Suite for a System That Doesn't Repeat Itself
- 26:34Three Steps, and the Third Nobody Warns You About: Watch the Cost
- 27:35The Roof: Becoming a Quality Architect
- 28:05Governance Sits on Collaboration and Advocacy
- 28:36Three Questions: Staging vs Production, Who Reviews, What Forces a Human In
- 29:36Closing the Loop and Picking One Column
- 30:40Q&A: How CI/CD Pipelines Should Change When Agents Write Code
- 32:42Q&A: How the SDET Skill Stack Evolves Into 2027
- 33:43Q&A: Enforcing Security Policy at the Tool and Infrastructure Level
- 35:46Q&A: Upskilling With 15 Years of Experience - Get Comfortable With Evals
related talks
43mFrom Clickers to Coders: The SDET Evolution Story
27mSkills are the New SDKs
16mThe Era of Intelligent Testing Transforming Testers into Skilled SDETs
29mStop Prompting, Start Composing: Agent Skills in Practice
01h 00mFrom Copilots to Agentic SE Teammates: Engineering the SE 3.0 Era with AI-Native...
14mPrompt Engineering Impacts to Software Testers : Staying in Demand
52mMaking SDETs from ppl with 0 programming experience
09mFoundational Skills for Testers