Workshop: The Full Agentic QA Loop for Web Applications

In this workshop, 𝐒𝐢𝐝𝐝𝐡𝐚𝐧𝐭 𝐒𝐢𝐧𝐡𝐚, Lead Member of Technical Staff at TestMu AI, points out that most "AI testing" demos stop at generating a test, while real QA is a loop: author coverage, capture what actually happened, assure it against intent, and keep it from rotting as the product changes. Learn to run that full loop end-to-end, hands-on, and leave with a working pipeline on your own machine. Siddhant also walks through each stage with kane-cli, an agentic test tool that drives a real browser and authors runnable tests from plain-language requirements - turning a one-line requirement into concrete scenarios, capturing every run as a portable .evidence pack with DOM, actions, network and assertions, and moving from brittle exact-match assertions to intent-level verification. 𝐇𝐢𝐠𝐡𝐥𝐢𝐠𝐡𝐭𝐬: 0:37 Speaker Intro - Siddhant Sinha, Lead Member of Technical Staff at TestMu AI 2:10 AI Writes Code in Seconds - So Who Tests the Whole Web Flow? 2:56 The Same AI That Wrote the Code Also Writes Its Test 3:11 Brittle, Slow Tests Never Proven in a Real Browser 4:12 The Second Problem: PMs, Devs and QAs Working From One PRD 5:29 Drift, Stale Tests and Coverage Numbers That Are Really Guesses 6:01 One Loop, One CLI: PRD In; Criteria, Scenarios and Tests Out 6:32 Tests Authored in Plain English, Replayed in CI With Evidence as Proof 7:33 Claiming 10,000 Free Credits to Follow Along 8:18 Installing the CLI and Logging In to TestMu AI 8:34 Test Manager: Where Authored Tests and Runs Are Audited 11:08 The Sample GitHub Repo and the Five Things the Workshop Covers 12:24 What Context Means Here, and Maintaining It as the PRD Changes 13:26 Running context ingest on the PRD 13:58 The Demo App: A Local V16 Food Ordering Website 14:59 Four Use Cases Extracted From the PRD 16:30 Changing a Use Case's Risk Level in Plain English 17:47 context list and context view: The Graph of Linked Use Cases 19:34 Designing Criteria and Scenarios Before Tests 20:20 Running design tests on Use Case One 20:52 The Review Gate: review and approve Commands 21:53 Sessions Let You Resume Instead of Designing Everything at Once 22:09 Thirteen Acceptance Criteria From a Single Use Case 23:25 Not Just UI - Asserting on Network and Console Too 24:57 The Agent Asks Two Questions Before Building Scenarios 25:43 Choosing a Scenario Budget - and Why No Limit Avoids Gaps 26:46 Four Scenarios: Two Happy, One Negative, One Boundary 28:17 Wiring Scenarios to Acceptance Criteria 31:04 Four Tests Created, Each Verifying a Set of Criteria 32:52 The Graph View, Lineage and the maintain evolve Command 34:55 Opening the Plain-English test.md in VS Code 37:12 Listing Tests and Running One in the Browser 38:29 Behind the Scenes: The Agent Reads Screen, Tree, Network and Console 39:30 After the First Pass, Execution Is Just Replay 41:20 The Run Completes: Artifacts, Version History and Test Manager Link 43:25 Re-Running the Authored Test as a Fast Replay 44:27 The Evidence Pack - Served Locally, Shared via evidence.lambdatest.com 45:13 Coverage: 12 of 15 Acceptance Criteria From One Test 48:18 Q&A: How Agentic Review Detects That a Test Has Drifted 49:36 Q&A: Preventing an Updated Test From Masking a Real Bug 50:54 Q&A: Who Is Responsible When an AI Agent Causes a Security Incident Register for TestMuConf 2027: https://www.testmuai.com/testmuconf-2027/?utm_source=youtube&utm_medium=organic&utm_term=&utm_campaign=full_agentic_qa_loop_for_web_applications #TestMuConf #TestMuAI #AgenticTesting #TestAutomation #WebTesting #Workshop #AITesting #QualityEngineering