Momentic, a Y Combinator W24 startup that has raised $19.2 million, launched pull-request testing for Vercel and Cloudflare deploy previews on November 2, 2026, and published QA Bench, an open benchmark for its Mo AI bug-finding agents. The company announced both in a Hacker News post.
The release is Mo's second public push, following coverage by SiliconANGLE on September 28. It moves the agent beyond exploratory testing and into the pull-request workflow, where, as many in the industry argue, AI-written code often arrives faster than teams can check it.
An agent that uses the app like a person
Mo looks for bugs by operating an application the way a user would, instead of running a maintained suite of scripts. The founders titled their post "Launch HN: Momentic (YC W24) – Agents that catch bugs by using your app."
The company draws a line between exploratory discovery and repeatable regression checks. Under Mo's documentation, one session tests one target, so web and mobile work needs separate sessions. The documentation lists web, iOS and Android as supported targets, and the founders described desktop as forthcoming.
Mo replaces Explore, an earlier agent that debuted in a June 23, 2026 platform update. Momentic's September 2026 changelog marks Explore as Legacy and directs users to Mo for autonomous bug finding.
Reading the QA Bench numbers
The new PR testing feature runs Mo against deploy previews from Vercel and Cloudflare. Alongside it, the founders released QA Bench, with results tied to label revision e77c72b in the benchmark repository.
The figures reward a slow read. The QA Bench dataset card lists 84.2% precision for Mo, a measure of how many of its reports are valid. Recall, the share of labeled bugs the agent actually found, is 19.3%. Emergent's November 3 headline said Mo "Catches 84% of Bugs," though the article's own body reports the 19.3% recall figure.
The comparison with OpenAI's Codex is mixed. According to the benchmark's published table, Mo surfaced more bugs. The leading Codex configuration, though, had a lower cost per matched bug and slightly higher precision, per the QA Bench repository.
Momentic's YC launch text makes bolder claims: 3.8 times more bugs than Codex, nine times more bugs per hour and half the cost per bug. That text is undated and describes a different evaluation from the November benchmark, so the two sets of numbers should not be read as one result.
There was also a brief discrepancy. Paras Madan wrote in a November 2 post, updated November 3, that older README numbers in the repository initially differed from those in the launch post. A rerun on November 2 resolved the mismatch, he wrote.
Pricing and early customers
New signups receive $250 in Mo usage, the founders said. That is trial credit rather than unlimited testing. Momentic's session documentation describes usage-based spending and notes that higher effort settings can raise total spend even when cost per discovery improves. The command-line tool includes a qa cost command that itemizes AI and runtime charges.
SiliconANGLE's September 28 coverage tied Mo testing to Notion, Superpower, Iris, Committee for Children and Boundless. In its June 23 release, Momentic named Notion, Xero, Bilt, Webflow and SPS Commerce as customers of its broader platform.
Founders and funding
CEO Wei-Wei Wu and Jeff An founded Momentic in late 2023, according to the company's March 4, 2025 release. Y Combinator's profile lists Harj Taggar as primary partner.
The founders' launch post cites developer-tooling and infrastructure experience at Robinhood, Retool, WeWork and Qualtrics, without assigning each employer to a particular founder. YC's profile separately lists cloud and test-infrastructure experience at Splunk, Google and Robinhood.
Momentic announced a $15 million Series A on November 24, 2025. Standard Capital led the round, with Dropbox Ventures, YC, FCVC, Transpose Platform and Karman Ventures participating. It followed a $3.7 million seed round announced March 4, 2025, led by FundersClub and FCVC. SiliconANGLE put total funding at $19.2 million. YC's jobs page listed 18 people as of November 4, 2026, though the page is undated.
A crowded field
Mo competes with conventional regression frameworks such as Playwright, Selenium and Cypress, plus Maestro for mobile. Momentic pitches itself against the upkeep those suites demand. QA Bench names Codex, Claude Code and Devin as comparators.
QA Wolf launched Automation AI on September 8, 2026. It generates Playwright tests from prompts or features, including in preview environments, which means it leaves executable regression scripts in place. BrowserStack and mabl appear in Momentic's own comparison navigation.
What the benchmark leaves open
Momentic's public command-line repository does not make the hosted agent service open source. The company's AI data-use policy also says constraints on the agent do not guarantee correct decisions.
The Hacker News thread showed 158 points and 124 comments on November 4. Those are engagement counts, not adoption figures, and the comments do not independently validate the benchmark.
