7th Frankfurt Meetup – Prompts Are Not Controls: Governing Agentic QA

What happens when an AI test agent ignores its instructions but still reports a successful test run?

As AI agents take on more testing responsibilities, from generating tests and executing suites to triaging failures and reporting results, relying on prompts alone is no longer enough. Quality requires enforceable boundaries, verifiable evidence, and clear accountability.

At The Test Tribe’s 7th Frankfurt Meetup, Muthu Kumar Koodalingam will demonstrate how to make AI-powered testing agents more secure, auditable, and reliable through engineering controls rather than instructions alone.

Session
Prompts Are Not Controls: Governing Agentic QA

Speaker: Muthu Kumar Koodalingam, QA Automation Engineer at GLS IT Services

With over 14 years of experience across mobile, web, and API testing, Muthu specializes in test automation strategy, developer tooling for QA, and applying engineering discipline to AI-assisted testing. His current work focuses on making AI agents in test automation enforceable, auditable, and verifiable.

This session explores why system prompts, AGENTS.md files, and skill instructions cannot serve as security controls on their own. Through a practical demonstration, Muthu will show how to define agent permissions, enforce restrictions at runtime, independently verify test results, and introduce human approval gates to keep agentic QA accountable.

The session will cover:

Enforcing Agent Boundaries: Defining agent authority through configuration, restricting write access by path, and removing unnecessary tools rather than relying on instructions.
Runtime Controls and Verification: Checking every tool call, reading back effective permissions, and verifying that agents operate within their authorised boundaries.
Trustworthy Test Results: Deriving pass/fail outcomes from exit codes and captured output instead of trusting an agent’s summary.
Independent Review and Human Oversight: Using independent model reviews, deterministic checks, and human approval gates to catch failures before they move to the next stage.

The central question to take back to your team: If your test agent ignored its instructions, what would stop it?

Event Details

Date: 11 November 2026
Time: 05:30 PM – 07:30 PM Local Time
Venue: Ederstraße 10, 60486 Frankfurt am Main-Innenstadt II, Germany

Agenda

5:30 PM – 5:45 PM: Welcome & Introductions
5:45 PM – 6:45 PM: Speaker Session by Muthu Kumar Koodalingam on “Prompts Are Not Controls: Governing Agentic QA”, including Q&A and demo
6:45 PM – 7:30 PM: Networking over snacks
7:30 PM: Event Wrap

Stay back after the session to exchange ideas with fellow QA and Engineering professionals on building agentic testing systems that can be trusted through evidence, enforcement, and accountability.

About The Test Tribe

The Test Tribe is the world's largest software testing community, empowering testers globally since 2018. With 925+ events and 170K+ members across 130+ countries, we bring software professionals together to learn, grow, and connect through meetups, conferences, cohorts, masterclasses, and more.

Become a Speaker at The Test Tribe Events

Wish to share your insights with the community at a future event?

Submit your talk idea here, and if it resonates with the perspective we are trying to build at our events, our team will connect with you.

By RSVPing, you agree to our Terms & Conditions and Privacy Policy and consent to be contacted by The Test Tribe and our event partners.
Clarify the Frankfurt venue detailsReplace the local time label