Write a user story and get executable test cases from it within minutes. That is no longer a thing of the future today.
AI can generate surprisingly good first drafts. The question is no longer whether it works, but how you turn an AI-generated first draft into a reliable automated test.
How the process works
from user story to test
◉
AI analyses
scenarios & criteria
</>
Gherkin or test code
generated
◎
Human review
mandatory check
Some tools use Cucumber-like language as an intermediate step: you describe a scenario in plain language, and the AI translates it into automation code. Others generate Playwright or Selenium code directly from a description. Both approaches lower the barrier to entry: even people who don't write code daily can put together a first test setup.
AI can write a test case. It cannot judge whether that test case covers the right risk.
Where it works well
- Happy-path scenarios: standard user flows are predictable and closely resemble the examples AI models were trained on.
- New features: generating a first draft takes seconds and avoids repetitive work.
- Migrations: AI can often convert existing Selenium tests to Playwright or other frameworks surprisingly well.
Where human review remains essential
Edge cases and negative scenarios
AI-generated tests tend to follow the obvious path. The scenarios that actually catch bugs (boundary values, error handling, unusual combinations) still require domain knowledge.
Assertions that are too weak
A login test that only checks that no error message appears does not yet prove that the user was actually logged in correctly.
Business logic the AI doesn't know
A model does not know your specific business rules unless you explicitly provide them. Without context, it generates generic tests that miss what makes your system unique.
The teams that get the most value from this don't use AI to replace testers. They use AI to speed up repetitive work, so testers can focus on the scenarios where experience, context and critical thinking make the difference.
An example from practice
At a client with a password-reset flow, an AI tool turned the user story "As a user, I want to be able to reset my password" into a working Playwright test within two minutes. The test was functionally correct and fully green.
// AI-generated: happy path only
1. request a reset link via email address
2. open the link from the email
3. enter a new password
4. log in with the new password
// Missing, and only surfaced during review:
// - is the old password still valid after reset?
// - what happens with an expired reset link?
// - is there a limit on reset requests per hour?
None of those three scenarios show up in an average user story, yet each is exactly the kind of gap an attacker or a frustrated user finds first.
A review checklist we apply by default
- Count the negative paths. Does every happy-path scenario have at least one matching failure scenario? If not, coverage is one-sided.
- Check the strength of every assertion. "No error message shown" isn't proof of success; check the actual outcome (redirect, database state, email received).
- Watch for hard-coded waits. AI-generated tests reach for a fixed sleep() surprisingly often instead of an explicit wait condition, a direct source of flakiness that only shows up weeks later.
- Ask about the test data behind the scenario. Does the test generate its own unique test user, or does it reuse a shared account that conflicts under parallel runs?
// Start faster, stay thorough
Curious how AI test generation fits your workflow?
We show concretely what works for your stack, and where human review continues to make the difference.
Book a call