# Canary

> Canary is an AI QA engineer that reads your source code to understand developer intent, then automatically generates and runs end-to-end tests on every pull request. Instead of flaky DOM scraping or screenshot analysis, it reads the diff, works out what a change is meant to do, and tests the real user flows it touches - catching broken checkout, auth, and billing before code reaches production. Founded in 2026 by ex-Windsurf, Cognition, and Google engineers, the two-person San Francisco team is part of Y Combinator's Winter 2026 batch.

- **Founded:** 2026
- **Headquarters:** San Francisco, California, United States
- **Founders:** Aakash Mahalingam (Cofounder and CEO), Viswesh N G (Cofounder)
- **Team size:** 2
- **Products:** Canary AI QA Engineer, Regression testing, QA-Bench v0
- **Notable:** Accepted into Y Combinator's Winter 2026 (W26) batch., Published QA-Bench v0 and reported outscoring GPT-4.5, Claude Code (Opus 4.6) and Sonnet 4.6 on test coverage across 35 real PRs., Caught a $1,600 invoicing drift for a construction-tech customer before it shipped.

## Products & services

- **Canary AI QA Engineer** — Connects to a codebase and, on every pull request, reads the diff, infers the intent of the change, then generates and runs end-to-end tests against the preview app - commenting results and recordings directly on the PR. Focuses on second-order effects beyond the happy path.
- **Regression testing** — Continuously runs test suites written in plain English so previously working flows keep working as the codebase changes.
- **QA-Bench v0** — A public benchmark of the QA agent against GPT-4.5, Claude Code (Opus 4.6) and Sonnet 4.6 across 35 real PRs, scored on Relevance, Coverage, and Coherence.

## Achievements

- Accepted into Y Combinator's Winter 2026 (W26) batch.
- Published QA-Bench v0 and reported outscoring GPT-4.5, Claude Code (Opus 4.6) and Sonnet 4.6 on test coverage across 35 real PRs.
- Caught a $1,600 invoicing drift for a construction-tech customer before it shipped.
- Benchmarked against real PRs from Grafana, Cal.com, Mattermost, and Apache Superset.

## Latest updates

- **2026-01** — Launched publicly via Launch HN and YC as 'the first AI QA engineer that understands your code.'
- **2026-01** — Released QA-Bench v0 comparing its agent to leading foundation models on real pull requests.

## Links

- Website: https://runcanary.ai
- LinkedIn: http://www.linkedin.com/company/canaries-inc
- YouTube: https://youtu.be/IeP7H_E8yCM

---

Profile page: https://yespress.io/canary-yc-w26
Published by YesPress — https://yespress.io
Last updated: 2026-07-31