Momentic's Mo runs an agent swarm instead of a test suite

The reason your test suite is bad is that somebody has to maintain it. Momentic launched Mo today, an AI testing agent that skips the suite entirely: you hand it a URL, test credentials, and instructions in plain English, and it spins up a swarm of agents that go poke at your app. No Playwright file. No selectors that break when a designer moves a button. For a small team that quietly stopped writing tests eighteen months ago, this is the more interesting half of the agent boom.

What actually happened

Per SiliconANGLE's coverage, Mo runs from a web chat interface or a CLI. You give it the app URL, credentials, and what you want checked. You can also attach a requirements doc, a Jira issue, a Linear ticket, or a Confluence page and let it derive the cases itself.

From there it "opens up the app, spins up a swarm of agents, and then goes through the motions" — thousands of permutations and edge cases, recorded. What comes back is a blow-by-blow of what was tested, what failed, reproduction steps, and video evidence of each bug. Co-founder Wei-Wei Wu's framing: "It's just agents verifying it themselves, based off of your instructions."

Momentic named customers running Mo over the past few weeks: Notion, Superpower, Iris, Committee for Children, and Boundless. Pricing was not disclosed.

Why agentic testing matters for your business

The maintenance cost is the actual product here. A test suite is a liability that grows with your app. Every ship breaks a handful of selectors, someone spends a morning on it, and eventually the team starts marking things skip. An agent that re-derives intent from the app each run has no selectors to break. That is the whole pitch, and it is a good one.

Video evidence changes the bug-report loop. We have lost real hours to "works on my machine" and a screenshot of the wrong tab. A recording of the exact click path that broke checkout is worth more than a stack trace to anyone who is not the person who wrote the code.

Non-determinism cuts both ways. A swarm that tries thousands of permutations will find things your fifteen hand-written tests never would. It will also not run the same way twice. That is fine for exploratory coverage and bad as a merge gate — you cannot block a deploy on a check that might flag something different on a retry. Run it alongside a small deterministic suite for your money paths, not instead of one.

Test credentials on a live app deserve scoping. You are handing an autonomous agent a login and telling it to try everything. Point it at staging with seeded data, give it an account that cannot touch production payments or real customer records, and confirm what happens when it finds the delete button. "Thousands of permutations" includes the destructive ones.

Pulling cases from Jira and Confluence is the quiet win. Most small teams have requirements written down somewhere and tests written nowhere. Closing that gap without a human translating tickets into code is where the hours actually come back.

Sources: SiliconANGLE, Momentic.