WhoBuilt

Who Tests AI-Written Code?

Oct 12, 2026· 13 min· 622 views·Watch on YouTube

Speaker

At AI T Conf in Warsaw, Szymon Rybczak, CEO of TesterArmy and a former Callstack developer, talks with Callstack’s Luke Farrell and Ola about his startup journey and AI-assisted software testing.

YC’s interest in AI and hardware

Rybczak estimates that about 90% of startups in Y Combinator are focused on AI. He says YC is also investing in hardware and robotics, in part because the field has not yet had a ChatGPT-like breakthrough.

From open source to TesterArmy

Rybczak and co-founder Oscar built open-source developer tools while working at Callstack, including React Native AI and React Native Bottom Tabs. After trying several B2C ideas, they focused on a problem they saw in their work: manual testing could not keep pace with AI-assisted coding and frequent releases.

How the testing platform works

TesterArmy tests web and mobile apps, including native and React Native apps on iOS and Android emulators. Users can start with a website URL or app binary, then connect source code so the system can read pull request diffs and descriptions to create a tailored test plan. Teams can also provide business rules to help distinguish intended behavior from bugs.

Hosted product and open-source SDK

The hosted product manages the platform and selects models for users. Rybczak says the team plans to open-source a TypeScript SDK called E2E, intended to support tests for web, mobile, desktop, and other software; its mobile support will use Agent Device and let users bring their own AI key and cloud.

Branding and QA teams

Rybczak says the first TesterArmy logo was AI-generated, then founding designer Veronica created the pixel-art logo and animation. He sees agentic testing tools as a way to equip QA teams, rather than replace them.

Takeaways

  • Look for bottlenecks that remain after AI speeds up another part of the workflow; Rybczak saw manual testing lag behind faster coding.
  • Use pull request diffs and descriptions to focus test plans on the changes being shipped.
  • Let teams provide business rules when automated tests need context about intended behavior.
  • Offer a quick-start path, such as a URL or app binary, so users can try a testing product before a sales process.
  • Frame QA automation as tooling for QA teams, not as a replacement for their expertise.

Mentioned in this talk

Search inside this talk

What was claimed 14

Related talks