Approxima Opens Your App in a Real Browser to Check That Your Pull Request Actually Works

The Macro: Nobody Writes Enough Tests and Everyone Knows It

Every engineering team agrees that automated testing is important. Almost no team has the test coverage they wish they had. End-to-end tests are the worst offender: they are slow to write, brittle to maintain, and break every time the UI changes. The result is that most teams either have minimal E2E tests or spend significant engineering time maintaining a flaky test suite.

The test maintenance burden is particularly painful. A team spends weeks writing Cypress or Playwright tests. Then the product changes. Buttons move. Pages restructure. APIs update. Half the test suite breaks, not because there are bugs, but because the tests are coupled to implementation details that changed.

The alternative is manual QA, which does not scale. Every PR should be tested, but manual testing of every PR is impractical for teams shipping multiple times per day.

What if testing did not require writing or maintaining test scripts at all? What if an AI could look at a PR, understand what changed, open the app in a browser, and verify that things work?

Approxima, backed by Y Combinator, does this. Their agentic QA platform opens your application in a real browser on every PR and checks that changes work, with no test scripts to write or maintain.

The Micro: Jane Street and Citadel Alumni Killing Test Suites

Ethan Pronev (ex-Jane Street and Citadel, CS ‘25 at UWaterloo) and Ashish Selvaraj (ex-Databricks and Microsoft, CS/BBA ‘25 at UWaterloo) built Approxima to eliminate the test maintenance problem entirely.

The approach is conceptually simple but technically ambitious. On every PR, Approxima spins up the application, navigates it in a real browser, and validates that the changes function correctly. It uses AI to understand what the PR changed, determine what to test, and verify the results, without any predefined test scripts.

Visual regression testing, pull request validation, and self-serve QA automation are the core capabilities. The visual regression component catches UI changes that break layouts, misalign elements, or produce unexpected visual artifacts.

This is a fundamentally different approach from traditional test automation. Instead of maintaining a parallel codebase of test scripts, you let AI observe and validate the application at runtime. The AI does not need updating when the UI changes because it is not following a script. It is understanding and evaluating.

Competitors include traditional E2E testing frameworks (Cypress, Playwright, Selenium), AI-assisted testing tools like Testim and Functionize, and visual testing platforms like Applitools. Approxima’s differentiator is eliminating test scripts entirely rather than making them easier to write.

The Verdict

Approxima is attacking one of the most persistent pain points in software engineering. Test maintenance is universally hated and universally under-invested.

At 30 days: how many PRs are being validated by Approxima, and what percentage of real bugs does it catch?

At 60 days: does the no-script approach work reliably for complex applications with authentication, state management, and API dependencies?

At 90 days: are teams reducing their traditional test suites in favor of Approxima, or using it as a complement?

I think the approach is right. If Approxima can deliver reliable PR validation without test maintenance, it solves a problem that every engineering team has and nobody has solved well. The challenge is handling the complexity of real applications. A browser test that works on a simple CRUD app needs to also work on a complex SaaS product with dozens of features and edge cases.