How one engineer at Younite built 1,500 regression tests with Testsigma.

The challenge: the project had no tests, and a codebase that kept breaking them.

Most automation stories start with brittle scripts that somebody has to keep alive. This one starts earlier than that.

Aleksi Nokela from Younite picked up the work all alone. There was no automated suite. There were also no manual test cases, written down anywhere, for anyone to run.

"In this project, there weren't any tools," he said. "There hasn't been a written down manual test or anything like that."

The app was in active development and shipping constantly. So the backlog of things nobody had ever tested kept growing while he worked through it.

Then the part any automation engineer will recognise. The code had been written without automation in mind, so the elements he needed mostly had no IDs or names.

"We have to rely pretty much on the xpaths, and those change when the UI changes. And then we have to make the changes to the test."

One person, an untested app, and a locator strategy that breaks on every redesign. That is the tax that usually ends a one-person automation effort. Every interface tweak breaks tests that passed yesterday, and the person repairing them is the same person who was supposed to be writing new ones. The backlog grows from both directions.

Why using a framework was out of the scope.

Building and maintaining a framework involves keeping its dependencies alive. Fix it when it breaks. Then start writing tests, alone, against a moving target, with a backlog already deep enough to worry about.

Testsigma let him skip it. Tests get written in plain English, so authoring tests was the work rather than a thing he had to build the capacity for first. When the step library ran out, he went and found what he needed instead of filing a request.

"For my tests, I have found all the needed steps, and for some steps I have been using the community add-ons."

The platform also covered web, Android and iOS from one place, which is what the team had originally bought it for. In practice this project went deep on web and API rather than wide across devices.

The result: roughly 1,500 tests, running while nobody is watching.

From zero to roughly 1,500 test cases in about 18 months: One person, on an application that never stopped shipping.

A nightly regression pass nobody has to start: Core suites run every night. Bigger suites run over weekends and during any longer break in the build. Deeper runs go out ahead of releases. Results land in Slack when a plan finishes.

The routine checking stopped being a person's job: It freed the only person on the job to spend his attention where judgment was actually needed.