From AI exploration to stable test cases: Best practices for test generation in QF-Test 11

QF-Test 11 lets AI agents explore your application and generate a first draft of a test suite. A question we hear a lot from customers evaluating these features is: how do you turn that AI-driven exploration into stable, maintainable test cases, rather than ending up with a skeleton suite full of “AI Instructions”?

AI Instruction nodes are a great prototyping tool, and they can help you get unstuck in a pinch, but they’re not a good foundation for long-term, robust UI testing — you should migrate them to native QF-Test test steps wherever possible.

This post walks through the workflow that works well today, what’s changing with QF-Test 11.1 later this year, and where the current limits are when it comes to AI reading or editing existing suites.

The recommended workflow today (QF-Test 11.0)

The current recommended approach combines the AI-based test planning and generation dialog with QF-Test’s classic recording mechanism:

1. Use the “Plan and generate tests with AI” dialog to generate an initial suite of test cases with “AI Instructions” from a prompt and a connected application.

2. Clean up the generated suite: Remove unnecessary steps, clarify instructions and add missing information where needed.

3. Convert the individual “Test Step” nodes into recorded sequencces via record & replay: Press the “Start recording” button, run a test step, then stop and save the recording. Repeat for every test step. You can use the option “Insert recording at current selection” to streamline the process.

4. Review and refine the results: You may want to refactor repeating actions into Procedures, put repeating values into variables using Set variable, or insert conditional logic using if and else nodes, or error handling using try/catch. QF-Test offers all kinds of different nodes for structuring and enhancing your test suite.

This “rerecording” approach can be quite powerful when combined with the Procedure call recording mode introduced in QF-Test 11. With it, you can turn AI instructions directly into matching procedures from your existing library:

Who ya gonna call? Auto-generate procedure calls for clean and maintainable tests

In this webinar, we’ll show you how QF-Test 11 can recognize matching procedures from your library while recording sequences and automatically insert the corresponding calls.

Close

The dialog for planning and generating tests with AI.

Close

A freshly generated suite, made up of “AI Instruction” nodes.

One limitation to keep in mind: UI actions performed by the AI agent like clicks, navigation events, keyboard inputs, are picked up automatically by a running recording, but performed checks can not yet be captured this way in QF-Test 11.0.

You don’t have to treat exploration and recording as separate phases. If you keep a recording running during the exploration phase of test generation, every interaction the agent performs is captured as it happens. They are not automaically organized into individual tests, but you can copy and paste them into your generated test where they apply.

What’s coming in QF-Test 11.1 (later this year)

This workflow will become much nicer and more automated starting with QF-Test 11.1:

• The suite generator will directly produce recorded nodes where possible, and only fall back to AI Instructions if the required interaction is too complex.
• You will be able to convert AI instructions into recordings in bulk, instead of requiring the manual record-and-replay dance for each test step.
• Most assertions performed by the AI agent will also be converted into native checks.
• New MCP tools (and matching internal tools) will let the AI start and stop recordings, and organize the results into simple procedures and test cases automatically.
• A new script API will let you start or unpause recordings programmatically, in case you need more control.

If you’re interested in trying these features early, a pre-release preview build may become available before the general release — reach out to us if you’d like to get involved.

Reading and editing existing test suites

Recording covers adding new content to a suite, but our customers also want the AI to read, explore, and edit existing suites — for example to understand what a suite already tests before extending it, or to refactor it. We’re investigating dedicated solutions for this (MCP tools and internal APIs for structured suite access), but do not have anything available yet.

In the meantime, suites can be explored by external AI agents via their underlying XML file. This works reasonably well for suites that aren’t too large, but it doesn’t scale to big or complex suites.

When experimenting with editing suite XML keep in mind: a suite that has been modified externally — for instance by an AI agent editing the XML outside of QF-Test — can no longer be opened with a free QF-Test evaluation license. Opening an externally modified suite requires a commercial license.

Today, the best path from AI exploration to a stable test suite is: generate a draft with AI, then record alongside it (or convert it afterwards) to get native, replayable steps (excluding checks). QF-Test 11.1 removes several of these manual steps: bulk conversion of AI instructions into recordings, AI-controlled recording via MCP, and checks captured during recording. Broader support for AI reading and editing existing suites is still being investigated; until then, the underlying XML is your best option for small suites.

Interested in an early look at QF-Test 11.1? Contact us and let us know you’re interested!

Best practices for designing your own MCP toolkit with QF-Test

How to turn your own QF-Test procedures into a custom, self-documenting MCP toolkit for AI agents.

Max Melzer

25/09/2026

AI-driven test automation with QF-Test

With QF-Test 11, you can integrate your own AI models into agentic workflows and have them create and execute tests autonomously. Additionally, the integrated MCP-Server allows you to make the powerful QF-Test tools available to external agentic AI systems as well.