Workflow guide · 3 tools
How to test browser workflows with AI
Test a browser journey when its value depends on several steps working together: a user can submit a form, see the resulting state and return to it later. Use AI to navigate changing interfaces, but make pass or fail depend on an outcome you can observe.
Compare the tools
| Tool | Best for | Tradeoff |
|---|---|---|
| Momentic | Teams that want hosted browsers and natural-language testing of deploy previews. | Credit-based usage can grow with test volume and is not a bring-your-own-model setup. |
| e2e | Keeping TypeScript browser tests in the repository while mixing Playwright-backed steps with agent actions. | You configure the runtime and model yourself, and the framework is still pre-1.0. |
| Agent QA | Natural-language browser scenarios that need to survive routine UI changes. | Not a turnkey hosted execution service; adaptive navigation still needs assertions to catch semantic changes. |
A practical workflow
- Pick one critical journey and use a disposable test account with seeded data. Keep real customer data out of model context.
- Use stable selectors for predictable steps and put a concrete assertion after any exploratory AI step.
- Run the scenario on an unchanged build, then break the expected behavior to check that the test detects it. Measure runtime and flaky reruns before broadening coverage.
Try one on your project
-
Momentic
Use a preview deployment for a single form flow. Compare the hosted run and evidence with your existing smoke test.
-
e2e
Keep login and submission deterministic, and use an agent only for the tricky interaction before an explicit confirmation assertion.
-
Agent QA
Rename a nonessential control and verify the journey still works. Then remove the final confirmation and verify the same test fails.