Agentic Development · Adversarial QA

Adversarial QA: challenge your software before your users do

The industry keeps accelerating, but our controls are not keeping up. AI produces copious amounts of code, and we need even better tools to ensure quality. That’s what our Adversarial QA (AQA) does.

Discuss your testing needs

01In plain words

What is adversarial QA?

Quality assurance (QA) checks whether software does what it should. Adversarial QA is about using different AI models in an adversarial role to thoroughly check AI generated code.

We use 6 different models, running many agents each, to explore all aspects of a code base:

Security checks: no one wants to be hacked
Architecture checks: is the code structured to scale
Performance tests: is the code and data speed the best it can be?
Feature bugs: what might have slipped through the net in testing

An AI model reviewing its own work tends to miss its own mistakes. We pit up to 6 different models against each other. The models that review your code aren’t the ones that wrote it, and each one looks at it from a different angle.

02AQA as a service

We offer the AQA as a service

All our projects are checked by the AQA, and we offer the AQA as a service to others. Sleep easy after knowing exactly what the state of your AI code is, and particularly all the security findings.

Contact us to discuss the size and scope of the codebase you want checked, and we’ll work with your software team to provide this in-depth analysis of the state of your software, AI-generated and manually written. A simple run will cost around R10K. Can you afford not to check?

03Findings you can use

Findings you can act on

Testing gives evidence within the agreed scope. It can’t prove that a system has no defects, or that every AI answer will be right.

The AQA report includes a prioritised list of findings, from critical to low impact. This is supplied as a spreadsheet to track progress on addressing items, and the cover report. We also provide Skills Files for the team to use, where patterns are visible in the code that can be corrected by better Skills going forward.

A Skills file is an instruction to the AI that teaches it the skills you need, such as setting standards, architectural rules, UI standards etc.

04Who it’s for

When adversarial QA helps

Your team uses AI to write code and wants an independent check
You’re adding AI features and need to know how they fail
A major release or migration is coming up
You want a second opinion on code another supplier built

Miengha Behardien’s post on AI code review Why we don’t vibe code

What do you need more confidence in?

Tell us about the software, how it should behave and what’s worrying you. We’ll tell you whether adversarial QA is a fit and what a useful review would cover.

Discuss your testing needs Call +27 (0) 21 681 2900

info@krs.co.za · +27 (0) 21 681 2900