AI works great for code, but what about Mendix?

Why AI coding assistants struggle with Mendix (and how we can fix that)

If you build software in Mendix, you’ve probably seen colleagues working in Java, TypeScript, or Python use AI tools like GitHub Copilot, Cursor, or Claude. In those text-based stacks, you can point an AI assistant at a file, ask it to write a unit test, and get a sensible test script in seconds.

Try doing that in a Mendix project, though, and you quickly hit a wall.

The reason is pretty straightforward: “Mendix applications aren’t folders full of text files.”

Your microflows, domain entities, associations, security rules, and pages live inside a single `.mpr` project file. Because standard AI tools only know how to read text files, they can’t see what’s going on inside your Mendix model. If you ask an off-the-shelf AI to “test my calculate discount microflow,” it usually makes up random Java code or guesses what your parameters might be.

From clicking through forms to AI-assisted test creation

Why setting up tests in MTA used to take too long

For teams already familiar with Menditect Test Automation (MTA), the execution engine has always been solid. It runs microflows directly, supports frontend UI tests, and automatically rolls back database changes so your test data doesn’t clutter your environments.

The pain point was never the test runner; “it was the time it took to build tests by hand.”

In MTA 3.1 and earlier, setting up a test meant doing everything manually in the web interface: creating a suite, adding test steps, selecting entity types, and typing in parameter bindings and assertion rules one by one.

If you had a couple of microflows to test, that was manageable. But if you were maintaining a large Mendix 9 or 10 app with dozens of business-critical flows, manually scripting tests felt like a full-time job. Most teams simply didn’t have the hours to spare, so test debt piled up, making refactoring or upgrading to Mendix 10 feel risky.

With MTA 3.2, that manual hurdle goes away:

• The AI handles the setup: Your AI assistant reads the microflow logic and calls MTA tools to configure the test steps, parameters, and assertions for you.
• Full test levels covered: You can set up unit tests for calculation microflows, component tests for module logic, backend process tests across multiple steps, and UI tests for web pages.
• Built-in data seeding and cleanups: Tests can seed necessary test objects beforehand and use MTA’s rollbacks to leave the database clean after execution.

What this means for teams

• For developers: You can generate a test suite for a complex microflow in a few minutes right from your editor, instead of spending an hour clicking through web forms.
• For testers: You get structured, repeatable tests across unit, process, and UI levels without having to write code or manage fragile test data by hand.
• For product owners and leads: You can finally get decent test coverage on older Mendix 9 or 10 apps, making future upgrades to Mendix 10 or 11 much safer.