AI-Driven Test Management
msg.ZenTestAI is built so that you can use as much or as little AI autonomy as you want — and the part that runs your tests stays predictable no matter how much AI you layer on top.
A stable core, with agentic freedom on top
Think of the platform as two layers.
A stable, deterministic core. Every test, however it was created, runs the same way: the same input produces the same result, runs are reproducible and auditable down to the individual step, and strong caching means unchanged steps are reused rather than re-reasoned. This is the foundation you can rely on for regression testing and CI/CD — it does not change when you add AI on top.
A dynamic, agentic layer. On top of that core, AI agents help you create and maintain tests. They can explore an application on their own, propose what to cover, draft and fix tests, and answer questions — but everything they produce lands as ordinary, reviewable test cases that run on the same deterministic core.
The autonomy spectrum
Between those two layers there is a spectrum, and you choose where to sit on it:
- Human designs, AI executes. You carefully model test cases by hand; the AI only carries out the execution. Predictable and easy to audit.
- AI designs, AI executes, human orchestrates. Agents explore the application themselves, suggest coverage, write tests and run them. You orchestrate and approve rather than author every step.
You can mix freely, per application and per test, and move along the spectrum over time.
What's in this chapter
| Topic | What it covers |
|---|---|
| The AI Assistant | The in-app chat assistant that helps you create, fix and reason about a specific test. |
| Agent Home | The autonomy cockpit: plan health at a glance, what needs you, and what the AI agent did. |
| Tasks | Batch AI work — tasks and subtasks, their statuses, and how the view keeps itself up to date. |
| Enabling AI features | How an administrator switches the AI assistants on and chooses the models they use. |
| The idea behind onboarding flows | Why AI-driven onboarding exists and how scan → propose → review → build works. |
| Application Map | The AI-generated map of an application's screens, and the onboarding wizard that produces it. |
| Knowledge Base | A per-application library of articles the AI uses as lasting context. |
| MCP (Model Context Protocol) | Connect external AI tools — Claude, Cursor, Copilot — directly to your tenant from your editor. |
New to the AI features? Start with Enabling AI features to switch them on, then read The AI Assistant.