Momentic takes its Mo AI QA agent out of beta, claims 3x more bugs found than rivals

Momentic, a vendor of AI-driven end-to-end app testing, says Mo, its AI QA engineer, is out of beta. Mo tests an app end to end and, in the company's words, will "verify critical flows and find bugs from a prompt." A session can be started from a coding agent, the app, a PR comment or the qa CLI. The pitch on the page is that coding agents have made building software much faster, so the bottleneck is now knowing that it works, and verification belongs in the agent loop.

The headline claim is a benchmark run by the vendor. Momentic graded Mo, Codex, Claude Code and Devin on 20 areas of four open-source apps: Gitea, Metabase, Twenty and Saleor. With 60 minutes per area, Mo found about 269 labeled bugs. The next best agent found about 86, so Mo found roughly 3.1 times as many. The page also says a separate agent reproduces each functional bug before Mo reports it.

The page gives one example of a failure found by Momentic: at a consumer fintech app, a login routing change bypassed the new-device verification step, so a security checkpoint was silently skipped.

How Mo works. Mo runs autonomous QA sessions in the cloud, on the web or on hosted Android and iOS devices, with the organisation's MCP servers and secrets attached to the session. For pull requests, it can be started from a PR comment, a label or the CLI; it tests the PR's preview deployment, posts a review with its findings and sets a Mo commit check. Each session has an effort level (low, medium, high or max), ranging from about 15 minutes of testing for a pull request to about 2 hours for a whole app. Bugs can be accepted, dismissed, fixed or reopened with qa bugs; a retest keeps the finding and its history and marks it fixed when Mo can no longer reproduce it.

The tests themselves. Beyond Mo, Momentic's older product is a test framework. Steps and assertions are written in plain-English YAML files kept in the repository, so a team and its coding agent can review and version them. The example on the page is a checkout flow with steps such as adding a backpack to the cart, filling in personal info through reusable modules, and asserting that the item total matches the listed amounts. The CLI runs the tests locally against the app being changed, and the same tests run in CI before a merge or deploy. When a step fails, the user can inspect screenshots and traces, and AI triage investigates the cause and proposes a test repair for review. Permanent healing proposes test-file edits for review, and the user applies the diff and reruns. A Knowledge base lets teams describe product terminology, business rules and expected behavior so agents can choose actions and check outcomes; session suggestions are reviewed before they are added.

Triage goes further in the changelog: it works out what broke each test, commits the fixes the user's branch caused, then opens a pull request for every other cause and asks whoever owns it to review.

Changelog items. The V3 agent is now the default everywhere; it plans the flow up front, caches the resolved steps and self-heals on failure. The knowledge base is generally available, and classification, triage and AI test selection dropped their beta labels. Other listed items include iOS test authoring on hosted simulators, Android tests on hosted emulators without local Android Studio, an MCP server for driving browser sessions from a coding agent, request mocking step types, ephemeral inboxes via email.create() and numbers via sms.lease(), momentic doctor for diagnosing setups, a REST API for pulling finished runs, and caching of AI-action branches such as a cookie banner that shows up only half the time. Assertions can be checked against frames from the run video, so a toast that flashes for a second still counts.

Customers and enterprise. The page shows a testimonial about Poe.com and three unattributed customer metrics: daily test execution down from 7 hours to 30 minutes, 500+ manual test cases replaced, and 100% of critical tests created in one month. For enterprises it lists SAML and OIDC SSO, Admin and Member roles, audit logs, priority support and a dedicated account manager, and says Momentic is SOC 2 Type 2 certified.

Key facts

Momentic says Mo, its AI QA engineer that tests an app end to end from a prompt, is out of beta; sessions can start from a coding agent, the app, a PR comment or the qa CLI.
In a benchmark Momentic ran itself on 20 areas of Gitea, Metabase, Twenty and Saleor, with 60 minutes per area, Mo found about 269 labeled bugs against about 86 for the next best agent; Codex, Claude Code and Devin were the other agents graded.
Mo tests a PR's preview deployment, posts a review with findings and sets a commit check; effort levels run from about 15 minutes for a pull request to about 2 hours for a whole app.
Tests are plain-English YAML in the repo; AI triage proposes repairs for review, and permanent healing proposes test-file edits that the user applies as a diff.
Customer figures (7 hours down to 30 minutes of daily test execution, 500+ manual cases replaced) are unattributed, and no pricing is given.

Why it matters

The page frames the problem plainly: coding agents have made building software dramatically faster, and the bottleneck is now knowing it works. Momentic's answer is to put an autonomous tester in the same loop, one that gives coding agents evidence from the running app rather than relying on their own reading of the code. The claimed novelty is the move from scripted tests to a QA agent that explores an app from a prompt, alongside the older plain-English YAML tests. The fintech example shows the kind of bug the company wants to catch: a routing change that silently skipped a new-device verification step.

Who it affects

Engineering teams that ship often and use coding agents, QA engineers who maintain end-to-end suites, and teams that test web, Android or iOS apps. Enterprises get SAML and OIDC SSO, Admin and Member roles, audit logs, priority support and a dedicated account manager, and the vendor says it is SOC 2 Type 2 certified.

How to use it

Start a Mo session from a coding agent, the app, a PR comment, a label or the qa CLI. Pick low, medium, high or max effort per session. For regular regression tests, write steps and assertions in plain-English YAML in the repository, run them locally with the CLI, and run the same tests in CI before a merge or deploy. Review AI triage and permanent healing proposals before applying them, and use the Knowledge base to describe product terms and rules. The page gives no pricing, plans or credit costs.

How solid is it

Most of the evidence is the vendor's own. The benchmark is Momentic's comparison, and the page does not say who did the grading, how bugs were labeled, or how Codex, Claude Code and Devin were configured; no independent verification is mentioned. It also does not say which agent was the next best. The customer metrics are not tied to a named customer, the Poe.com quote has no named speaker, and the fintech example names no company or date. The SOC 2 Type 2 claim comes from the vendor, which points to a Trust Center for compliance reports.

Risks and caveats

No false-positive rate for Mo's bug reports is given, though the page says a separate agent reproduces each functional bug before it is reported. Fixes are proposed for review rather than applied automatically in the healing and repair flows, but the triage description says it commits fixes the user's branch caused and opens pull requests for other causes, so teams should decide how much autonomy to allow. Changelog items carry no release dates. Sessions run in the cloud with the organisation's MCP servers and secrets attached, which is something security teams will want to review.

“Momentic gave us a fast and reliable way to validate Poe.com's AI responses, even when they weren't deterministic.”

— Testimonial on Momentic's product page; speaker not named