Overview
Leapyear runs hundreds of internal services that move money for its customers every day. Higher code coverage gives its engineering teams confidence in application stability when they add new code, but writing unit tests by hand had become one of the slowest parts of every release. The company wanted coverage to grow without slowing delivery.
Challenge
Engineers were spending a large share of each sprint writing repetitive unit tests for Java services, many of them more than a decade old. Coverage was uneven between teams, reviews dragged on, and the riskiest legacy code often had the fewest tests, which made even small changes feel like a gamble before release.
Slow, manual test writing in every sprint
Uneven coverage between product teams
Risky releases on legacy Java services
Solution
Leapyear connected Synthora agents to its repositories and CI pipeline. For every change, agents read the surrounding code, generate meaningful unit tests, run them against the build and open a pull request for an engineer to review. Model routing sends simple files to faster models and complex logic to stronger ones.
Agents write and update tests on every commit
Model routing keeps cost predictable
Reviewers approve changes in minutes
Start with the Riskiest Code
The rollout began with the services that handle payments and settlements. These had the oldest code, the fewest tests and the highest cost of failure.
Instead of asking teams to pause feature work, agents generated a baseline test suite for each service in the background. Engineers reviewed the results in small batches, corrected the few tests that misread business rules, and merged the rest within days instead of months.
Once the baseline was in place, every new commit carried its own tests. Coverage stopped drifting downward between releases and began to climb steadily across every team.
Keep Engineers in Control
Leapyear's security and compliance teams needed proof that nothing reached production without human approval. Every agent change arrives as a normal pull request with a clear summary of what was tested and why, so existing review rules still apply.
Engineers can comment on a generated test and the agent revises it, which turned review into a quick conversation rather than a rewrite. Over time the agents learned each team's conventions and naming patterns.
Measure the Impact Release by Release
Within one quarter, coverage on every critical service had doubled and time spent writing tests had fallen by more than ninety percent. Release reviews became shorter because reviewers trusted the safety net underneath.
The bigger change was cultural. Engineers no longer see tests as a chore that competes with product work, and teams plan features without reserving whole sprints for maintenance. Leapyear now applies the same agent workflow to integration tests and documentation, and leadership tracks coverage alongside delivery speed as a shared measure of engineering health.
