DEV Community

Cover image for I built PROOF: a tool to verify AI-generated apps with real browser tests
Eduardo Dos Santos
Eduardo Dos Santos

Posted on AI-assisted

I built PROOF: a tool to verify AI-generated apps with real browser tests

Building software with AI coding agents is incredibly fast.

But I kept running into a frustrating problem:

The AI says the task is complete. The code compiles. The tests pass. But does the application actually work?

That question inspired me to start building PROOF, a tool that turns real browser tests into visual evidence.

๐Ÿ” What is PROOF?

PROOF is a local web application that uses Playwright to verify user interactions and generate reports showing exactly what was tested.

Instead of relying only on statements like "all tests passed," I wanted a way to see the actual results.

The current prototype can:

  • Execute configurable browser tests.
  • Interact with real web applications.
  • Verify specific success conditions.
  • Capture screenshots during execution.
  • Identify failed verification steps.
  • Generate standalone HTML reports.
  • Preserve evidence from completed test sessions.

๐Ÿงช Testing a real application

For the first independent test, I used TodoMVC, an open-source example application.

I configured PROOF to:

  1. Open the application in Chromium.
  2. Create a task called "Review the release checklist."
  3. Verify that the task was created.
  4. Mark the task as completed.
  5. Confirm that exactly one task was completed.
  6. Verify that the counter displayed "0 items left."

Result: PASSED.

PROOF recorded the execution, verified the expected conditions, and generated a report containing the evidence.

I also built a separate demonstration involving a broken contact form. PROOF detected a validation error, and a second execution verified the corrected behavior.

๐Ÿ› ๏ธ The technology

The prototype uses:

  • React and TypeScript for the interface.
  • Node.js for the local verification service.
  • Playwright with Chromium for browser automation.
  • HTML reports with embedded screenshots.

The goal is to make test results easier to inspect, understand, and share.

โš ๏ธ What PROOF cannot do yet

I want to be transparent about its current limitations.

PROOF is still an early prototype that runs locally.

It doesn't automatically discover every bug.

Tests require configured interactions and expected outcomes.

It currently supports authorized local applications rather than arbitrary public websites.

And a passing test proves only the specific conditions that were checked.

๐Ÿš€ Where I want to take it

I'm exploring whether PROOF could become a useful product for:

  • Developers building applications with AI.
  • Freelancers delivering websites to clients.
  • Small development agencies.
  • Teams that need visual evidence before releasing software.

My goal is to make software verification more accessible, especially for people who build applications quickly with AI-assisted tools.

๐Ÿ’ฌ I'd love your feedback

I'm currently validating the idea before investing in a larger product.

If you build applications with AI, how do you verify that they actually work before shipping?

Would you use a tool that automatically runs your configured tests and creates shareable, evidence-backed reports?

And what would it need to offer for you to consider paying for it?

I'm also looking for a few developers interested in trying an early version.

Let me know what you think!


Disclosure: PROOF was developed with AI-assisted coding tools. This article was also prepared with AI writing assistance and reviewed against the prototype's demonstrated capabilities.

Top comments (0)