Canary (YC W26) launches an AI agent that automatically tests code changes by generating and running end-to-end user workflow tests directly on pull requests. Founded by Aakash and Viswesh, who previously built AI coding tools at Google and other startups, Canary addresses the gap where code changes break real user flows like checkout or billing. The system connects to your codebase, understands changes, and comments PRs with test results and recordings.

Beyond PR testing, Canary moves tests into regression suites and allows English prompts for test creation. Canary's purpose-built QA agent outperformed GPT-5.4, Claude Code, and Sonnet 4.6 by 11, 18, and 26 points respectively in the first QA-Bench v0 benchmark measuring coverage of affected workflows. This highlights the challenge of multi-modal QA requiring custom browser fleets and specialized harnesses for second-order effects. The company aims to solve the critical problem of verifying real user impact before code merges.