The archive · Developer & Business Tools · Product decision · 2026
TesterArmy (YC Spring 2026) bets plain-English QA agents, not scripts, guard every PR
Three Polish founders from YC's Spring 2026 batch launched agents that test web and mobile apps in plain English; Launch HN drew 132 points and 69 comments.
TesterArmy
What the business is
Managed AI QA service whose agents click, type and navigate real web, iOS and Android apps like users, run on every GitHub pull request and on schedules, and return screenshots, recordings and bug reports — no selectors or test scripts live in the repo.
How it started
Polish engineers Szymon Rybczak, Oskar Kwasniewski and Piotr Matyjasik, who had spent years shipping web and mobile products, founded TesterArmy around a problem they knew firsthand: teams ship fast, but writing and maintaining automated tests is painful, so QA gets skipped. They went through Y Combinator's Spring 2026 batch and launched publicly on June 18, 2026.
What happened
The Launch HN drew 132 points and 69 comments, and the YC launch post said the service was live with 20+ companies in production, offering 30% off with a launch code. The product runs real browsers, handles captchas, OAuth and email OTPs, and reports video, screenshots and network logs; Claude Code, Codex and Cursor can drive it through MCP, and results post as GitHub PR checks. Through August 2026 the team shipped an Issues tab that deduplicates agent-found bugs (Aug 20), open-sourced unbox-ai, a trace explorer for its own QA agent's tokens (Aug 26), and benchmarked ten vision models on clicking accuracy (Aug 18); site testimonials claim regressions caught before production, including 20 issues in three months at Copyfy.
No ending yet — it is still running.
Background
TesterArmy is a Y Combinator-backed agentic QA service. Its agents click, type and navigate real web, iOS and Android apps like human users, run on every GitHub pull request and on production schedules, and return screenshots, recordings and bug reports; no test scripts live in the customer's repository. The company was founded by three Polish engineers and appears in Y Combinator's Spring 2026 batch.
The bet is that teams will hand QA to always-on agents rather than maintain scripted frameworks: you describe a journey in plain English, and the service runs it, fixes it when the UI changes, and owns auth — captchas, OAuth and email OTPs arrive in per-agent inboxes. The founders positioned this as a different layer from Playwright or Cypress: the same browser primitives underneath, but the maintenance burden is outsourced.
TesterArmy launched on June 18, 2026 with a Launch HN that drew 132 points and 69 comments; the YC launch post said the product was live with 20+ companies in production. During August 2026 the team shipped an Issues tab for deduplicated bug reports, open-sourced unbox-ai, and published a benchmark of ten vision models on click accuracy, with testimonials claiming production regressions caught before users saw them.
What has to be true
- Framework-era QA punishes every UI change with brittle selectors, so suites rot and get skipped; an agent that maintains natural-language journeys removes that upkeep tax.
- Black-box delivery — a URL or a compiled build, no SDK, no repo wiring — plus MCP prompts for Claude Code, Codex and Cursor made a first test run fast enough to feel magical.
- Solving auth, OTPs and captchas removed the blocker that historically killed automated testing of logged-in flows, opening production monitoring rather than just smoke tests.
- Every run ships video, screenshots and network logs, so a gate on pull requests does not ask engineers to trust an invisible agent.
What can be applied
Sell the maintenance, not the test: TesterArmy attacked why QA automation gets abandoned — upkeep, logins and brittle evidence — rather than adding features to test frameworks.
Aftermath
As of September 5, 2026 TesterArmy is an active Y Combinator Spring 2026 company based in San Francisco, listed with a four-person team, and still live with a free trial and paid subscriptions. The site advertises GitHub PR checks, scheduled production monitoring for web and mobile, and test authoring by Claude Code, Codex and Cursor, with recent blog posts covering its deduplicated Issues tab and the open-sourced unbox-ai trace explorer. No funding round amount, revenue or verified customer count beyond the founders' launch claims has been disclosed.
Sources
spotted an error? The archive wants to know.
Your turn
You just read one. Describe what you are building, and see who is betting on the same thing.
Free account · 3 free questions · no card