The regression pass is the part of QA that nobody defends and everybody keeps. A list of flows, a build, and a team working through the list on a calendar that is not yours. It gets outsourced because it is repetitive, and it stays outsourced because writing an automated suite for it costs more time than the release has.
Two things changed. Agents can now explore a product and file a bug with a recording. And the tests they write can live in your repository, so the pass you automate stays yours. This page lists the choices and what each one is best for.
What the offshore pass actually delivers
An offshore pass is good at the work that needs a human eye. A person notices what a script never asked about. A vendor takes on work that nobody owns. The trade is that the pass runs on its own calendar and reports days later. Coverage lives in a document rather than in your codebase, so the same list gets run by hand at the next release.
Five ways to cover the pass
An offshore QA team: best for judgment on flows you cannot script
Keep people for payments in a real account, localization, accessibility, and hardware. You are buying a cycle, not a suite.
A crowdtesting vendor: best for breadth on real devices before a launch
Many testers, many devices, one window. It fits a launch, but it is expensive as a weekly habit and leaves nothing that runs on your next merge.
A managed QA service: best for a suite you do not want to own
The vendor writes and maintains the tests in its workspace, on its schedule. That is the same coupling as the offshore cycle.
An in-house automation team: best for a product whose testing is the hard part
You get full control, but you also carry the whole bill: authoring, flake triage, browser and device infrastructure. See the in-house cost model for pricing.
An AI QA agent plus a suite in your repository: best for turning the pass into something that runs on every build
Mo runs the pass on demand. Agents explore the product, work cases in parallel, and file reproduced bugs with a recording and repro steps. Flows you want checked forever become Momentic tests: plain-English YAML in your repository, run by the CLI in CI.
In one customer session, 116 agents returned 27 bugs from one bug bash and covered two days of manual QA in one run. The customer was a Fortune 1000 file-collaboration platform with more than 700M registered users, as described by a Senior Engineering Manager.
What a session looks like in place of a pass
- You point Mo at the build and give it the scope, from the web app , the
moCLI, a schedule, or a pull request through the GitHub App. - Each agent gets its own browser, simulator or emulator, and a hundred or more run at once, so a pass over dozens of flows is not the sum of their run times.
- Findings arrive while the session runs, and the first bugs can arrive within minutes.
- Every report has the same fields, so this week's verdict reads next to last week's.
The roles matter, because the agent that found the behavior is not the agent that decides it is a bug. A bug reproducer confirms it from a clean start, and when the sandbox cannot reach what a case needs, the report records a gap instead of a bug.
"I don't trust an agent with the bug bash"
A customer said exactly that, and it is the right instinct. An agent that reports everything it noticed is a second inbox, not a QA pass.
A finding arrives with a recording and repro steps, so you judge it in a minute rather than believing it. A separate reproducer confirms it from a clean start before it is filed. An unreachable precondition is reported as a gap, not a bug.
Run it as a shadow pass first. Keep the human pass for one release, run the session on the same build, and compare the two lists. That comparison is cheap, and it settles the argument with data.
Can AI replace a manual QA team?
It replaces the repetitive pass: the regression list, the dogfood doc, the browser and device matrix. It does not replace the person who decides what is worth testing, and it does not cover payments in a real account, localization judgment, or a defect that only appears on hardware. The teams that get the most from this keep a little human testing and stop paying for the checklist.
Retool increased its release cadence 8x, saved more than 40 engineering hours a month , and automated 100% of its manual QA checklist, according to its published case study.
Where to start
Pick the next release. Run the human pass as planned and start a session on the same build. Compare the lists. Then take the flows that both found, write them as Momentic tests in your repository, and let CI run them on every merge from then on.
See how Mo runs a session and read the Mo docs for setup . Talk to sales for a walkthrough .