Skip to content

Agentic testing

AI agents write and run the tests for each user story as soon as it is built. They cover the UI, APIs, databases, ETL pipelines, performance and security, and log each failure as a bug your developers can act on.

  • UI
  • APIs
  • Database
  • ETL
  • Performance
  • Security
Stage run, 15 Nov 02:00To: QA team, development leadsSample
  • 212Passed
  • 3Failed
  • 3Bugs filed
  • UI118 of 119 passedFail
  • APIs56 of 57 passedFail
  • Database18 of 18 passedPass
  • ETL12 of 12 reconciledPass
  • PerformanceCheckout p95 1.8 s at 200 usersPass
  • Security1 high finding: session not closedFail

10,000 test records seeded and removed

One set of agents for the whole stack. We start with the UI and add layers as you need them.

The Consortium for Information & Software Quality estimated that poor software quality cost the US at least $2.41 trillion in 2022. Our agents write the tests while each story is fresh and run them on every build, so a defect turns up in dev or stage within a day of the change that caused it.

  • User interface

    Playwright tests follow each journey in the acceptance criteria. The agent also tries bad input, the wrong user role, expired sessions and values at the edge of what each field accepts.

  • Backend and APIs

    Contract tests call each endpoint and compare the response with your API specification, including status codes, error messages and authentication. Service-level load tests show where each API starts to slow down.

  • Database

    Integrity checks look for broken references, duplicate records and values that break your business rules. Query times are measured while the database is under load.

  • ETL and data pipelines

    The agent tests transformation logic and data quality rules, then reconciles source with target by comparing row counts, totals and individual fields.

  • Performance

    Virtual users run your busiest journeys at the same time. Page load and response times, including the 95th percentile, are compared with the previous release.

  • Security

    Checks cover injection through forms, session and cookie handling, what each user role can reach, and data exposed in pages. For APIs the agent also tests access control and configuration.

How it works

From a finished user story to a retested fix.

  1. Story ready

    The agent reads the user story, its acceptance criteria and the code change.

  2. Tests written

    It writes tests for the positive and negative paths, and your engineers review them.

  3. Data seeded

    A data agent creates the users and records each test needs.

  4. Tests run

    Functional, load and security tests run in dev, stage or production.

  5. Results sent

    A report goes out by email and each failure is logged as a bug.

  6. Fix retested

    When a fix is committed, the agent sees it in git and reruns the failed tests.

When tests run

A schedule for each environment, and a retest whenever a fix lands.

  • Dev

    Every merge, plus nightly

    New tests for the latest stories, then the full functional suite.

  • Stage

    Nightly and before each release

    Full regression, a load test at your target number of users, and the security scan.

  • Production

    Hourly, or at the interval you set

    Smoke tests on critical journeys with synthetic accounts. Load tests and intrusive security tests stay in stage.

Retest on every fix

  1. A developer commits the fix for BUG-218.

  2. The agent sees the commit and finds the tests linked to that bug.

  3. Those tests run again, along with the checks around them.

  4. The result is posted on the bug and sent to the developer.

Reports and bug tickets

Written for the people who have to act on them.

After every run, a report goes by email to the people you choose, usually the QA team and development leads. It gives the totals, one line per test type and a link to each bug filed.

Each failure becomes a ticket with its severity, the environment, steps to reproduce, the expected and actual result, and the evidence a developer needs to replay it.

BUG-218High · Stage · SecuritySample

Session stays open after sign-out

Description
After signing out, the browser Back button shows the previous page with customer data, without a new sign-in.
Steps to reproduce
  1. Sign in as a finance user and open the invoice list.
  2. Click Sign out.
  3. Press the browser Back button.
Expected
The sign-in page is shown.
Actual
The invoice list loads with customer data.
Evidence
  • Screenshot
  • Playwright trace
  • Console log

Outcomes

What the automation gives your team.

  • Tests ready when the code is

    Every story that gets built has positive and negative tests.

  • Every build checked

    Dev, stage and production are tested on the schedule you set.

  • Bugs ready to fix

    Each bug comes with steps to reproduce, a screenshot and a Playwright trace.

  • Less manual regression

    QA time goes into reviewing the tests and exploring edge cases.

Questions

Agentic testing, answered.

What can the agents test?

The user interface, backend APIs, databases, ETL and data pipelines, performance under load, and security. We usually start with the UI and add the other layers one at a time, using the same agents.

Who reviews the tests the agents write?

Your engineers. The agent proposes its tests as an ordinary code change, so nothing joins the suite unreviewed, and a test that turns out to be unhelpful can be edited or removed.

Is it safe to run tests in production?

In production the agents run smoke tests on critical journeys only, using synthetic accounts, hourly or at the interval you set. Load tests and intrusive security tests run in stage.

What happens to the test data?

A data agent creates the users, accounts and transactions a run needs, up to thousands of records for a load test, and removes them when the run ends.

Which AI models do the agents use?

We pick the model for each engagement on cost, security and where your data has to stay. The agents run inside the AI harness, so permissions, audit logs and human approval apply to everything they do.

How is this different from your test automation and code security service?

That service reviews every pull request for security issues and can block a merge until agreed checks pass. Agentic testing writes and runs tests for each user story across the stack. The two can run side by side.

Can our own engineers run the agents?

Yes. Our one-day, in-person course teaches developers, tech leads and QA automation engineers to work with AI coding agents: setting permissions, reviewing the agent's work, and using subagents, hooks, MCP and headless runs.

See the agents test one of your stories

Tell us about your application and we will show you the tests the agents would write for it.