January 16, 2026
With AI's Help, PM and Engineering Can Now Own Tests
Fine PMs ship.
Great PMs obsess over the product, stress the key flows, and notice what should be better.
They audit the key journeys constantly. More than anyone, they know what actually matters.
Onboarding. Checkout. Upgrade. Billing. Account setup. The paths that drive revenue, retention, and trust.
And when you live in those journeys, you eventually hit an issue. A button does nothing. A modal never loads. The flow breaks.
The old world: the feedback loop is slow, lossy, and brittle
When a PM finds a problem in the product, the system around them often looks like this:
The PM files a ticket with the best description they can write.
It gets triaged and prioritized.
Engineering attempts a fix, often without perfectly reproducing the original path.
If you are lucky, QA later turns that flow into a test.
The UI changes slightly, the test breaks, and it gets ignored.
Everyone involved has good intent. The process is the problem.
The “ticket to test” translation step is slow. It loses nuance. It creates distance between the person who knows what matters (PM) and the artifact that keeps it from breaking again (the test). And because those tests are typically brittle, every UI change comes with a maintenance tax.
So teams end up in a familiar pattern:
Bugs recur because nothing durable got encoded.
Tests exist, but they drift away from reality.
Flake trains engineers to stop trusting the suite.
The new world: agentic test generation closes the loop
Agentic test generation changes the operating model.
Now, when the PM hits an issue while auditing a key journey, they can turn what they just experienced into a durable test immediately.
Not by writing Playwright scripts.
Not by learning selectors.
Not by opening a PR.
They simply record the journey in the browser exactly as it happened and add a couple natural language assertions that capture intent, like:
“User reaches confirmation”
“Plan updates successfully”
“Receipt downloads”
“Error message is shown”
“User remains logged in”
That journey becomes a durable, agent-powered test.
And the most important part is what happens next.
Self-healing makes the test suite worth owning
Traditional UI tests are fragile because they anchor on implementation details. A small UI shift can break selectors even when the user experience is still correct.
Agentic tests behave differently.
Because the system learns your product and the intent of the journey, tests can self-heal as the UI evolves. That keeps the suite high signal and low flake without becoming a maintenance burden.
This changes the default assumption inside a team.
When a PM hits a bug while using the product, they can assume engineering is already seeing it, because the test is failing.
That is the outcome teams actually want:
Faster detection
Clearer reproduction
Less back and forth
Less test babysitting
More trust in what CI is telling you
What changes when PMs can own tests
PMs already own outcomes. Agentic test generation lets them also own the durable guardrails around those outcomes.
PMs can:
Codify the most important journeys while they are fresh in their head
Turn “this feels broken” into a reproducible artifact
Keep a living suite of product-critical workflows without needing engineering time for every test
Instead of writing tickets that describe behavior, PMs can publish tests that assert behavior.
That is a meaningful shift.
What changes when engineers can create tests in minutes
Engineers are closest to change. They know exactly what they just shipped and what could break.
When test creation is fast and maintenance is handled by the platform, the workflow becomes natural:
Ship a feature
Record the flow once
Add a few natural language assertions
Move on
The result is that “done” can actually mean “protected” without introducing a testing tax.
Where QA fits in this world
Agentic test generation is not about removing QA. It is about removing the bottleneck that forces QA to be the sole author of automation.
In this model, QA becomes a force multiplier:
Defining standards and best practices for assertions
Reviewing the highest-risk journeys
Monitoring signals and reliability
Helping teams scale coverage without scaling maintenance
PMs and engineers create and own the tests in the moment. QA helps the system stay healthy.
This is what we built at Decipher AI
At Decipher AI, we built agentic test generation so teams can create durable E2E tests quickly, with almost no maintenance.
The workflow is simple:
Record steps in the browser
Add natural language assertions that match how humans describe success
Let the system generate the test and keep it stable as your product evolves
No brittle selector whack-a-mole. No scripting marathon. No growing test suite that quietly turns into a second codebase.
This approach is already helping some of the fastest-growing, AI-native companies ship faster and more reliably.
A practical way to roll this out
If you want PM and engineering owned testing to work in practice, start small and make it real:
Pick 5 core journeys
Onboarding, checkout, upgrade, billing, and one workflow that is uniquely critical to your product.Assign explicit owners
A PM owns the intent of each journey. An engineer owns updates when functionality changes.Define assertion style
Keep assertions human-readable and outcome-focused. “User reaches confirmation” beats “Button X exists.”Make tests part of shipping
When a feature changes a journey, record the updated path and ship the updated test as part of the work.Let the platform handle stability
Self-healing and low flake are what make this sustainable over months, not days.
The real takeaway
Testing stops being a separate world.
It becomes something PMs and engineers can own in the moment, with the platform doing the hard parts.
If you want to introduce PM and engineering owned testing in your organization, reach out and we can design the right setup for your team.
Written by:
Michael Rosenfield
Co-founder
Share with friends:

