Skip to main content

AI testing services where an engineer reviews every test

AI agents explore your app and draft Playwright tests for your critical flows. A QA engineer reviews every pull request before it's merged. It's how we deliver every QA Launch and QA Care engagement.

Our AI testing services use agents to explore your app and draft Playwright tests for your critical flows. A QA engineer reviews every one before it’s merged, so nothing an agent writes reaches you unchecked. We call this way of working agentic QA.

Agents and engineer review are how we deliver all our QA services: QA Launch, one month to cover your critical flows at a fixed price, and QA Care, optional monthly upkeep. QA automation describes the finished suite.

What an AI testing agent does

An AI testing agent is an AI model connected to a real browser and a set of tools, and given a goal such as “check that a returning customer can reorder from their order history”. It opens the app, decides what to click and type, compares what it sees with what should happen, and writes down what it did. Because it follows the goal, it can adapt when a button moves or a label changes, which would break a recorded script.

The same flexibility means an agent can misread a page, report a problem that isn’t there, or miss one that is. So we don’t sell autonomous AI testing. Agents do the work they’re good at, and a person is accountable for every result.

Where traditional test automation breaks down

Teams that try test automation without a dedicated QA engineer usually run into the same problems:

  • The suite stops working. Tests are written once. Then the UI is redesigned, dozens of selectors break, and fixing them never makes it into a sprint.
  • Red builds become normal. Nobody triages failures, so the team learns to ignore them, and the suite stops protecting anything.
  • Coverage stops at what someone had time to script. New features ship without tests because writing them is always the next sprint’s job.
  • Bug reports are vague. “Checkout is broken” arrives in Slack with no steps, no browser and no screenshot, and a developer loses an hour reproducing it.

How our AI testing services work

  1. You name the flows. Only you know which flows cost money when they break. You pick up to ten.
  2. Explore. Agents walk each flow on your staging app, step by step, guided by any docs you share and, if you allow it, read access to your code.
  3. Draft. Agents turn each flow into Playwright tests: the main path and the variations that matter. Each draft arrives as a pull request to a test repository in your GitHub organisation.
  4. Review. A QA engineer runs every draft, fixes weak assertions and approves it, or sends it back. This is where most of the value is: left alone, an agent will happily check for a “Thank you” message that also shows on the error page.
  5. Run. The finished tests are ordinary Playwright code, with no AI at runtime, so runs are fast, repeatable and cheap. They run on a schedule and after every deploy to staging, report to a shared Slack channel, and never block your merges.
  6. Repair (QA Care). When a run fails, an agent works out why. If the UI changed, it updates the test in a pull request. If the product is broken, it drafts a bug report. An engineer decides which it is within one business day, and files real bugs in Jira, Linear or GitHub Issues with steps, video and a Playwright trace.

One month first, then monthly if you want it

QA Launch covers up to ten critical flows in one month, at a fixed price agreed before we start. The month begins once your staging environment and test accounts are ready, and it ends when every flow has tests running in your CI. If that takes longer, the extra time is on us. Bugs we hit along the way are reported as a bonus.

QA Care is optional and billed monthly. Agents keep the tests current as your UI changes, an engineer checks every failed run within one business day, and we add up to two new critical flows a month. Cancel any month and the tests stay yours.

You provide a staging environment, a test account for each role and sandbox modes for payments and email, and name one person who can answer product questions within a day. We do everything else.

What you get

  • Playwright tests in their own repository in your GitHub organisation, drafted by agents and reviewed by an engineer
  • Every change to the suite as a pull request you can read
  • Runs on a schedule and after every staging deploy, reported in Slack
  • With QA Care, test updates when your UI changes, and checked bug reports with steps, video and trace
  • Tests that stay yours if you stop

Who it’s for

A good fit:

  • Teams that ship weekly and change their UI often
  • Teams with no QA engineer, or one who is overloaded
  • Teams whose test suite breaks after every redesign
  • Teams that want bug reports they can act on, with steps and video

Not a good fit:

  • If you want fully autonomous AI testing with no human review, we don’t offer it, and we don’t recommend it
  • Products without a staging environment, because we won’t run agents against production customer data
  • Safety-critical systems that require formally validated test processes

Pricing

We don’t publish prices, because scope depends on your product. After a 30-minute call you get a fixed price in writing, before any work starts. We don’t bill by the hour.

Book a call for a quote

HOW THE WORK IS STRUCTURED

  1. QA Launch (one month)

    Agents draft the tests for up to ten critical flows. An engineer reviews every one.

  2. QA Care (monthly, optional)

    Agents update the tests when your UI changes. An engineer checks every failed run.

WHAT’S INCLUDED

  • Agents that explore your staging app and draft tests for your critical flows
  • Human review of every agent-drafted pull request before it's merged
  • Plain Playwright tests in their own repository, with no AI at runtime
  • With QA Care, test updates as pull requests when your UI changes
  • Bug reports, checked by an engineer, for the bugs we find

Frequently asked questions

Is agentic QA the same as AI test automation?

They overlap. AI test automation usually means using AI to write or repair test scripts. Agentic QA goes further, with agents that also explore the product and draft test updates and bug reports. In our service, the tests that run are ordinary Playwright code, so runs are fast, repeatable and cheap.

Can AI agents replace our QA engineer?

Not on their own. Agents are fast at repetitive work, such as walking flows, writing tests and reproducing bugs, but they misread pages and sometimes report problems that aren't there, so an engineer reviews everything. For teams without a QA engineer, agentic QA gives you one, with agents doing the legwork.

How do you stop the agents from reporting bugs that aren't real?

Every report is reproduced and reviewed by an engineer before it is filed in your tracker. Reports that can't be reproduced, or that describe intended behaviour, are discarded. You only see reports a person has checked.

Do the agents touch production or real customer data?

No. Agents work against your staging environment with test accounts. If you have no staging environment yet, setting one up comes first.

Which AI models do you use, and is our code used for training?

We use Claude, from Anthropic, under commercial API terms that exclude training on your data, and we list it in our data processing agreement. Read access to your code is optional. Without it, the agents only see your staging app and the documents you choose to share. If your data must stay in the EU, we can run the models in EU regions.

What happens if we stop after the first month?

The tests stay in your repository and keep running on their schedule as ordinary Playwright tests. You lose the upkeep and the triage, but nothing is locked into our tools.

How do we start?

With a 30-minute call where you name your critical flows, after which you get a fixed price within two business days and QA Launch covers them in one month. You can stop after it and keep the tests.

Start with the flow you worry about most.

Book 30 minutes and bring a staging URL and the flow you worry about most, and you get a fixed price within two business days.