Bruno Legeard

Head of AI Lab 

Join my presentation on: AI Test Execution Agents vs Scripted Test Automation: A Practical Decision Framework

AI test execution agents can now execute natural-language test scenarios (manual or Gherkin) directly on a GUI, without automation code. Interacting visually, they perform functional tests like a human, generating structured evidence (screenshots, explanations, PASS/FAIL verdict). This capability challenges the long‑standing assumption that functional GUI testing must be fully scripted to be automated. 
 
 
 
Read More

Watch The Presentation

Who is Bruno Legeard ?

Head of AI Lab

I have been working at Testinium as a Senior Test Solutions Architect for over 7 years. I lead automation and performance testing projects and manage projects and teams. I specialize in mobile, web, and API test automation and provide mentorship to my teammates.
 
Together with our team, we evaluate the feasibility of new technologies and integrate them into our projects. My goal is to optimize processes, provide innovative solutions, and achieve successful results collaboratively.

What will Bruno Legeard be discussing?

AI Test Execution Agents vs Scripted Test Automation: A Practical Decision Framework

AI test execution agents can now execute natural-language test scenarios (manual or Gherkin) directly on a GUI, without automation code. Interacting visually, they perform functional tests like a human, generating structured evidence (screenshots, explanations, PASS/FAIL verdict). This capability challenges the long‑standing assumption that functional GUI testing must be fully scripted to be automated. In many contexts, AI agents can replace scripted automation for functional test execution, significantly reducing the cost of test creation and the ongoing maintenance burden caused by fragile locators in frequently changing UIs.

I propose a strategic shift: use AI agents as the default functional GUI test execution engine during the high-volatility development and qualification phases to enable fast feedback. Scripted automation, ideally AI-assisted, should be reserved for a smaller set of stable, high-value tests run frequently in the CI/CD pipeline. Based on experiments across twenty projects, I will introduce a Decision Radar to help teams select the right approach to test execution using three dimensions: execution cadence, UI evolution rate, and oracle strictness. I will also demonstrate how to measure and govern AI agent reliability using true/false PASS/FAIL metrics against human references to avoid false confidence.

Key Takeaways:

  1. How to decide “agent vs script” using three practical questions: how often the test runs, how fast the UI changes, and whether the oracle is visual or rule‑based.
  2. How to measure and control trust in AI test agents with accuracy, correct FAIL detection, and correct PASS detection—so agent execution can be used safely in production test processes.
  3. How to build a cost‑effective QA test execution strategy: replace a large portion of scripted functional tests with AI agent execution, and keep only high‑frequency CI/CD gate tests as scripted automation for speed and determinism.