Quality Assurance·

How to Get Your QA Team On Board With AI Testing

How to get your QA team on board with AI testing: the objections that are right, what QA owns afterwards, what stays human, and a first-month plan.

QA.tech team

Get QA On Board With AI – a sculptural brass control wheel connected to a compact glass testing mechanism.

To get a QA team on board with AI testing, give them control of it before you ask them to trust it. Let QA choose where the agent runs first, ask them to be critical, and put in writing what they'll own afterwards: the rules, the product knowledge, the critical-path plan, the test data, triage and the release call. "This isn't replacing you" only lands when the plan shows it.

Why it's harder to get your QA team on board with AI testing than your developers

You have a QA team and you mean to keep it. Without one, and with no plan to hire, the route is different: how to remove the dev-to-QA handoff without hiring QA.

The developers we work with tend to like agent-run testing from day one. It takes away work they never wanted. QA takes longer, and there's no mystery about why. A tester hears "AI testing" and feels they have to defend their role before they've seen a single run.

They have grounds. In PractiTest's 2026 State of Testing report, 65.6% of respondents said they were "Very Concerned" about the future of their profession. It's a vendor-run survey with no sample size on the page, so read it as a temperature check. The split inside it matters more to you. In the same report, 68.9% of practitioners were very concerned, against 55.6% of strategy leaders. The person proposing the tool is, on average, less worried than the people it lands on.

Some companies have let their testers go, and your team knows those stories. Announce a pilot into that and part of the room hears a headcount plan.

When testers first get their hands on an agent, they tend to split into two camps. One sees doom, digs in and doesn't want to touch it. The other starts out sceptical, pokes at it for a while, and then thinks: hang on, I can actually do something with this. The job changes, but toward work they wanted to be doing anyway.

The survey data points the same way. Testers who actively use AI were 17% less anxious and four times more likely to report "Zero Concern" than those who don't, PractiTest found. That's a correlation, and calm people may simply be the ones who try things. It still matches what we see in pilots.

None of this is special to testing. When BCG surveyed about 500 technology leaders in late 2025, change management came out as the top barrier to GenAI adoption across the software lifecycle.

The objections QA people raise, and what's true in each

Much of what a QA team says against AI testing is at least partly right, and the part that's right should change your plan.

ObjectionWhat's true in itWhat to do
"This is here to replace us."Testers have been let go elsewhere, and the click-through part of the job does shrink.Write down what QA owns afterwards before the pilot starts. No headcount target on the pilot.
"I can't trust a result I didn't produce."An agent can misclick, misread a page and be wrong in both directions. Its runs aren't identical step for step.QA audits the first weeks of verdicts against the evidence, with a checklist to grade the tool.
"We lose control over what gets tested."On a pull request the agent picks the test cases from the change. Nobody on the team chose them.QA owns a small critical-path plan that runs on every change, plus the written rules every run follows.
"We've seen tools like this before."Plenty of products were an AI feature added to an old recorder, and the demo beat the second month.Run it on your own product, in an area QA picks, with stop criteria agreed up front.
"It doesn't know our product."Correct. Much of what to test lives in testers' heads, and no agent can read intent nobody wrote down.Make writing that knowledge down part of QA's job, and count it as real work.
"It'll bury us in generated tests."It can. Left alone, generated test cases pile up until the suite is noise.QA decides what joins the regression suite and what gets thrown away.

"This is here to replace us"

The fear is rational, so answer it with specifics. What you can say is that the repetitive pass through the same screens goes to the agent and the decisions about what matters stay with people. What you can't promise is that the team will look the same in three years. Testers know nobody can, so don't try.

"I can't trust a result I didn't produce"

A good tester distrusts a green light they can't inspect. That instinct is what you hired them for. An agent can pass something that's broken or fail something that's fine, and anyone who tells your team otherwise is selling.

So don't ask for trust. Ask for an audit. For the first weeks, QA checks the agent's verdicts against the evidence behind them and keeps score. We've already published a checklist to put to any AI agent that runs your tests. Hand it over and let the team grade the tool with it. Expect technical questions about reproducibility, test data and regulated environments. Each has its own section in the questions QA engineers asked us about agentic testing.

"We lose control over what gets tested"

This one is correct as stated. On a pull request, an agent decides what to test from the change, and if you run it again on the same change its picks can differ a little.

What gives control back is a floor that doesn't move. QA names the flows that must never break, and a small plan covering them runs on every change as a blocking step, whatever the agent does on top. QA also writes the standing rules the agent follows, and decides test by test which existing checks stay scripted, using the buckets in how to replace your regression suite. Automation that already runs reliably stays.

"We've seen tools like this before"

They have. Record-and-playback tools promised that nobody would maintain tests again, and the maintenance arrived anyway. Some of what's sold as AI testing today is an add-on bolted onto an older product, and a tester sees through that soon after they start using it.

Let a trial on your own product settle it, and give QA the authority to say it didn't work.

What changes in the QA role, and what QA owns

The job moves from running the checks to deciding what a pass means. In a public webinar this autumn, Vilhelm von Ehrenheim, our co-founder and Chief AI Officer, said the role is changing from "sitting down and clicking around manually in an interface" to being part of the strategy behind how a team verifies its work.

Testers see a strategy gap more than their managers do, by the way. Nearly 40% of individual contributors in the PractiTest survey said test strategy was missing or underdeveloped on their team, and only 25% of leaders agreed. So far, though, AI for QA engineers has mostly meant the other end of the job: 70% use AI for test case creation and 19.9% for risk identification, the same survey found. Nobody was short of test cases.

Once agents run the tests, this is the QA team's list.

QA ownsWhat that means in practice
The rulesStanding instructions every run follows: what counts as a failure, which devices and roles to cover, what to ignore. One good rule improves every later run, which is worth more than ten extra test cases.
Product knowledgeThe naming that differs between ticket and screen, the edge case that matters to one customer segment, the thing "everybody knows".
The critical-path planThe short list of flows that run on every change and block a merge or a release when they fail.
Test data and accountsA login per role, seeded data, the inbox for the verification email.
Acceptance criteriaChecking that a ticket says what "done" means. See how to write acceptance criteria for agentic testers.
What joins the suiteWhich generated test cases become permanent. See how PR test cases become your regression suite.
TriageDeciding whether a red result is a product bug, an out-of-date test case, bad data, a shaky environment or the agent's own mistake.
Exploratory testing and risk assessmentLooking where no test case exists yet, and saying which changes deserve the most scrutiny.
The release callReading the evidence and deciding whether it ships.

This lines up with the split in building a balanced QA team, where risk assessment, exploratory testing and release judgement stay with people. Settle it before the job titles, reporting lines and transition paths in restructuring your QA team for the AI era.

Developers get something out of it too. They see feedback on their own pull request while the change is fresh, so the obvious breakages are gone before the work reaches a tester.

What not to hand to an agent

Some things stay with people, and your team should hear you say so.

The release decision is the first. An agent gathers evidence and gives a verdict. A person reads it and signs. The same goes for deciding what's critical, because how much risk the business will carry is a business call. If an auditor expects a full regression pass with a named person's signature on it, that stays too.

Subjective judgement stays. Does this look right? Is it on brand? An agent can tell you the button works. It's a poor judge of whether the page is any good.

Checks that must run the exact same steps every time belong in a script. And a product with hardware in it still needs hands on the hardware.

One more, and testers will raise it before you do. Whatever grades a result shouldn't be able to quietly rewrite the test it's grading. Any change to a test case should be visible and yours to switch off.

A first-month plan that starts with QA choosing where the agent runs

  1. Week 1: QA picks the ground. The QA lead chooses one area of the product for the agent to run on, ideally one they know cold. Agree together what would make the month a failure, such as missed bugs or too many false alarms. QA sets up the test accounts and data and drafts the first rules.
  2. Week 2: QA triggers every run. Keep it conservative, more so with a team of mostly manual testers, whose day changes the most. A tester starts the agent by hand on the pull requests they'd review anyway, does their normal pass, and compares. Every verdict gets checked against its evidence and logged as right or wrong. Say plainly that you want the criticism.
  3. Week 3: corrections become rules. Each wrong verdict from week two is something the agent didn't know, a missing account or data set, or a limit of the tool. Write down the first, add the second, note the third. QA defines the critical-path plan and it starts running on every change. They also decide which of the generated test cases are worth keeping.
  4. Week 4: QA reports. The testers who ran the trial present it: what it caught, what it missed, what it got wrong and what it cost them in time. Then comes the decision, which is theirs as much as yours. Widen it, change it or stop.

Make room for doubt early. A pilot can have everyone keen except one person who seems keen and isn't, and a worry that stays unspoken tends to come back as resistance. Talk to each tester privately in week one, and ask what they'd need to see.

Expect the resistance to move, too. After a good pilot it may sit outside the QA team, in parts of the organisation that haven't touched the thing and are nervous about AI in general. The testers' own account of what they saw does more there than a slide from you.

How to talk about it

"This is not replacing you" is a sentence testers have heard shortly before a reorganisation. So start with the plan, with QA's ownership in writing and a pilot decision that's partly theirs.

Say what's changing. If the manual regression pass is going away, say that, and say what each person does with the time. And don't steamroll. Ask for resistance. Tell the team to be critical and to skip whatever they think you want to hear. A team that folds straight away hasn't tested anything.

Some people won't want the new job. A staff engineer quoted by LeadDev in April 2026 described "a genuine pushback of 'I don't want my job to be an orchestrator of agents,' which is an identity concern". That was said about software engineers, and we'd expect it to hold for testers who love the hands-on hunt. Some of your best exploratory testers will want to keep doing exactly that. There's room for it.

Decision rights belong on paper, and for a QA team that's one page. Who can change a rule? Who approves what joins the suite? Who can block a release?

Where QA's controls sit in the product

The product we build won't make your release call or judge whether a design looks right. Its runs aren't identical step for step, so a scripted suite you already trust should stay where it is. On the web it drives a Chromium-based browser and no other engine. And rules can steer a pull request review, but nothing in it guarantees that a named test case runs.

QA.tech, an AI testing platform, tests your running app through PR reviews, where an agent chooses the test cases from the change, and through test plans, where your team chooses them.

Ask how agentic QA platforms improve automated testing workflows and the usual answer is speed. For a QA team the more useful answer is where the controls are, and here they map onto the ownership table above.

Rules live under Settings → Knowledge and come in three kinds: for the chat assistant that drafts test cases, for PR reviews, and for the agent while it executes. Each test account is saved as a config with a note on which role it is, and a test case written in chat stays a draft until someone on the team activates it or asks the assistant to (the first suite created during onboarding is the exception).

For the critical path, the docs suggest a smoke plan your team keeps to 5 to 10 critical test cases. It runs on every pull request next to the agent's review. On GitHub that's a blocking Test Run Action, and the PR review can also be a required check before merge. On GitLab and Azure DevOps the pipeline calls the Start Run API and fails the job on a bad result.

Every PR review ends in pass, fail or unable to verify, and a tester can audit any of them: the pull request links to the run, with its screenshots, logs, step trace and any recording that was captured. On test plan runs, a review agent looks at each test case that didn't pass before anyone is notified, and separates product bugs from setup problems, flakes that passed on retry and test cases that need updating. It reports what should change. It doesn't edit test cases during the review.

What joins the suite is QA's to set as well. Once a pull request merges, a post-merge agent can promote the useful test cases from its review, working from a prompt QA can edit or narrow, and QA can switch the agent off. Whatever it promotes is labelled regression and auto-added, so a tester can filter for exactly what arrived, and a merge that adds nothing is routine. With the default prompt the same agent also updates test cases the merge changed and moves clearly stale ones to draft. It never archives on its own.

On GitHub the agent is switched on for new integrations and stays off for existing ones until someone enables it. On Azure DevOps it's off until enabled, and GitLab doesn't have it. The page written for testers is AI testing for QA teams.

FAQ

Can AI replace a manual QA team?

It can take over the repetitive part of manual testing, which is re-walking the same flows on every change. It can't decide what's critical, judge design and usability, explore what was never written down or own a release decision. A company that removes QA entirely also removes the people who knew what to check, and may not learn what it stopped seeing until an incident shows it.

Does a QA engineer have a future?

Yes, though the day changes. Less time goes on executing checks and more on rules, test data, risk and triage. In PractiTest's 2026 survey, senior testers who prioritised leadership and strategy skills showed a 10.6% income premium, while those relying solely on technical execution skills showed a 13.8% income penalty. That's correlational salary data from one vendor-run survey, so take the direction and ignore the decimals.

What does the QA team own when agents run the tests?

The rules agents follow, the product knowledge they're given, the critical-path plan, test data and accounts, the decision about which generated test cases join the suite, triage of failures, exploratory testing and the release call. The agent executes and reports against a bar the QA team sets.

How do you prepare a QA team for AI testing?

Start with a pilot the QA team designs. Let them choose the product area, agree the failure criteria in advance, trigger runs by hand at first and audit every verdict against its evidence. Turn each mistake they find into a written rule or a missing piece of test data. Have the testers present the results themselves.

Does an AI QA agent replace exploratory testing?

No. An agent can explore an application and find broken flows, and automated exploratory testing of that kind is useful coverage. What it lacks is a tester's sense of what would annoy a real customer and of where this product has hurt before. Use agents to widen the ground covered, and keep human sessions for the areas where judgement matters.

Our QA team is overwhelmed with manual testing. What works?

Move the repeatable checks first: the regression pass on stable flows and the smoke check on each pull request. Keep people on new features, risky changes and anything subjective. Pick one product area, prove it there, then widen. If the root cause is that development got faster and QA capacity didn't, that's the argument of QA can't stay in the slow lane.


Your team moves fast. Can your testing keep up?

Our QA agents test your product autonomously, so moving fast never means shipping broken. Start free or get a 30-minute demo.