The missing tool in my AI coding stack was Playwright

AI is really good for frontend work. Definitely Opus 4.6. ChatGPT-5.4 is still catching up, but Claude has the crown right now,
anyways.

I use it a lot for design, UI ideas, and getting screens built fast.

But the biggest problem shows up right after that.

The UI looks done.
The code looks fine.
Nothing throws errors.

And then the actual product is broken.

I ran into this while building the application form in my open source ATS.

Some fields just didn’t work.

No big red error or obvious crash.

The AI had written something that looked correct, but it never actually tested the flow.

I used to just manually check after the AI has written the code.

That’s basically the problem with AI coding right now.

A lot of code is almost right.

And i came accross a stack overflow survey found that 84% of developers use or plan to use AI tools, but more distrust the output than trust it: 46% vs 33%.

What fixed this for me was Playwright.

I started writing tests for the critical flows.

For me that means stuff like:

create a job with all custom fields -> apply to the job -> check the application in the dashboard

Then I can be more confident when i make some changes, that the main features wroks.

I write it once, and it keeps protecting the product on every commit.

I run it in GitHub Actions on every pull request, which Playwright supports well in CI workflows.

That changed the whole way I use AI.

Now AI can help me move fast, but Playwright is the thing that tells me whether the feature actually works.

It's also well integrated in vs code, and AI is also pretty good at writing the Playwright tests.

You can basically one shot most tests, if there's any erros just write: playwright test it not working, fix. Make no mistakes!! (for the vibe coders ;)

Hope this was valuabe :)

Curious if anybody else have had this problem and how did you solve it?