Momentic launches AI QA agents that test apps end to end

Momentic, identified as a YC W24 company in its Hacker News launch, has presented Mo, an AI QA engineer that tests applications end to end from a prompt. It matters because coding agents can generate changes quickly, while teams still need reliable evidence that those changes work in a running app before they merge or deploy.

According to Momentic, the launch is aimed at a shift in software development where verification has to sit inside the agent loop. The company argues that agents need evidence from the application itself, not only from code, to confirm intended behaviour. Its platform is therefore built around writing user flows, running them against a live app, and reviewing recordings, findings and reproduction steps.

What is Momentic and what did it launch on Hacker News?

Momentic describes itself as an agentic quality platform and AI testing platform. Its Hacker News launch title, Agents that catch bugs by using your app, frames the product around agents that interact with software rather than only inspecting code. According to Momentic, Mo is the headline product: an AI QA engineer that explores an app from a prompt, verifies critical flows and reports bugs.

The company also offers agentic testing, where a goal is given to AI actions while preset steps and assertions are kept in YAML files in a repository. An MCP server is presented as a way to write, run and debug tests from a coding agent. The pitch is that tests can be authored in plain English, stored alongside code, and reviewed like other project files.

Momentic says its customers include engineering teams at companies such as Notion, Coframe, GPTZero, Runway and Webflow, and it publishes testimonials from people at those organisations. Those claims are presented by Momentic itself, so they should be read as company-reported customer evidence rather than independent verification.

How does Mo test an app end to end?

According to Momentic, a user describes a flow and its expected result in plain English. Steps and assertions are kept in YAML, which the company says can be reviewed and versioned in a repository. It gives an example of a checkout test file with a URL, a description, modules for logging in and filling in personal information, an action to add an item to the cart, and an assertion that the item total matches the amount listed for each item.

When a test runs, Momentic says AI-assisted steps can look up product context from a Knowledge base that contains terminology, expected behaviour and test guidance. It also describes an App graph that maps screens and user flows seen during test runs, and a review step for session suggestions before they are added to the Knowledge base.

If a step fails, Momentic says users can inspect the failing step, screenshots and traces. Its triage tool is described as classifying failures as application bugs, test flakiness or environment issues, and proposing a repair for review. The company calls one capability runtime recovery and reviewed repairs: if a control is hard to find because the interface changed during a run, the system can find it again, and permanent healing proposes a test-file repair for a person to approve.

What does QA Bench measure and who does it compare?

Momentic promotes QA Bench as an evaluation of QA agents on real web applications. According to Momentic, QA Bench measures how many real bugs a QA agent finds in a running web application within a fixed amount of time, and how many of the bugs it reports are real. The homepage summary says Mo, Codex, Claude Code and Devin were compared against 1,396 labelled bugs.

That framing is useful because it separates two questions that matter in practice. The first is recall in a practical sense: did the agent find genuine defects? The second is precision: when the agent reported a bug, was the report correct? A system that reports many false positives could create triage work, while a system that misses real bugs could give false confidence.

The source text does not provide the detailed results of the comparison in the material supplied here, beyond the benchmark description and the number of labelled bugs. Readers who want the methodology, task setup and per-agent results should consult the QA Bench report linked from Momentic's site, rather than relying on the launch summary alone.

Which platforms and CI workflows does Momentic support?

According to Momentic, web testing runs in Chromium and Chrome on local or hosted browsers. iOS testing runs on local or Momentic-hosted simulators, and Android testing runs on local or Momentic-hosted emulators. Tests can run on a team's own machines, in CI, or on Momentic-hosted infrastructure.

The company describes a CLI for authoring and running tests from a terminal, locally or in CI. It says suites can run on every push from GitHub Actions, GitLab or any CI provider, and that the CLI exit code can be used as a required check before merging or deploying. For coding agents, the MCP server is presented as the route to author a test for a changed flow, check it in a live session and save the YAML alongside the code.

For enterprise customers, Momentic lists SOC 2 Type 2 certification, a contractual uptime SLA, SSO through SAML and OIDC, admin and member roles, audit logs, dedicated onboarding and priority support. It also mentions SSO and custom data policies for testing private applications.

What scale and customer results does Momentic report?

Momentic publishes large usage figures on its homepage. According to the company, it reports 77,813 tests created, 70,607,819 test runs executed, 339,166 changes verified, 80,290 pull requests verified, 8,932,104 auto-heals and 117,010 bugs caught. These are company-reported totals, and the source does not explain the counting method for each category.

Customer testimonials in the source describe several kinds of benefit. Erdem, head of developer experience at Notion, is quoted by Momentic as saying setup is simple and the developer experience has been strong. Glavin, a founding AI engineer at Coframe, is quoted as saying end-to-end QA runs for every pull request at 6.3x parallel speed. Alex, chief technology officer at GPTZero, is quoted as saying the team writes and maintains end-to-end tests 3x faster.

Momentic also publishes outcome claims attributed to customers. For Quora, it reports daily test execution of 30 minutes, down from 7 hours, and 500 or more manual test cases replaced. For Retool, it reports an 8x increase in release cadence and 40 or more engineering hours saved per month. For GPTZero, it reports 80 percent faster release cycles, 75 percent reduction in test generation time and an 89 percent decrease in defect escape rate. Again, these are figures presented by Momentic in customer stories, not independently audited results in the source text.

How can readers explore this with Grafy?

Grafy does not offer a dedicated AI QA testing platform in the list of apps described for this article, so the genuine connection is through adjacent tools for building, reading and comparing. Developers who want to work on the code side of an agent workflow can use Grafy Code, described as an AI coding studio, to review changes, test ideas and prepare the flows that a QA agent would later verify. Teams evaluating the launch can use Grafy Reader to read and question PDFs and papers, including a benchmark report if they save it as a document, and Grafy Chat to compare how different models summarise the QA Bench methodology or draft a test plan from a prompt.

That distinction matters. Momentic's claim is that an agent should use the running application and produce evidence from it. Grafy Code, Reader and Chat can support the surrounding work of writing code, reading evidence and reasoning about results, but they should not be confused with Momentic's browser, simulator, emulator and CI execution model.

Frequently asked questions

What does Momentic's QA Bench compare?

According to Momentic, QA Bench compares Mo, Codex, Claude Code and Devin against 1,396 labelled bugs. It measures how many real bugs a QA agent finds in a running web application within a fixed amount of time, and how many reported bugs are real. The launch material describes the benchmark but does not include full per-agent results.

Which platforms can Momentic test?

According to Momentic, the platform supports web testing in Chromium and Chrome, iOS testing on local or Momentic-hosted simulators, and Android testing on local or Momentic-hosted emulators. Tests can run on a team's own machines, in CI, or on Momentic-hosted infrastructure, with a CLI for local and CI runs.

How are Momentic tests stored and run in CI?

According to Momentic, steps and assertions are written in plain-English YAML files that can be reviewed and versioned in a repository. The CLI runs the same tests locally and in CI, including GitHub Actions and GitLab, and its exit code can be used as a required check before merging or deploying.

Source: Momentic via Hacker News

---

Source: Hacker News — front page — Launch HN: Momentic (YC W24) – Agents that catch bugs by using your app