Scaling QA with Automation and AI-Assisted Development
At OBDeleven, we continuously expand to support new vehicle brands, communication with hardware, user-facing features. Each addition increases the complexity of our quality assurance (QA) efforts – especially when behaviour shall remain reliable across a wide range of vehicles, firmware versions.
As our apps grew, one limitation became clear: manual regression testing no longer scaled. What was once manageable, gradually turned into a time-consuming repetitive process as regression scope expanded with every new car brand and feature. Maintaining release confidence required a different approach.
From Manual Testing To Automation
Manual regression testing became a bottleneck as:
The number of supported vehicle brands increased
New features and code changes were introduced more frequently
Existing functionality required re-validation
Relying on manual execution alone meant that regression cycles grew longer with every release, limiting how fast we could confidently ship changes.
Our first step was to move toward automation-first QA. The goal was not to eliminate manual testing, but to reduce reliance on repetitive regression work and ensure that critical flows were consistently validated across releases.
By prioritizing automation for stable and repeatable scenarios, we were able to significantly shorten regression cycles. However, as automated coverage expanded, so did the effort required to develop and maintain test code.
AI-Assisted Test Development
To address end-to-end automation issues, we integrated AI-assisted developer tools — such as GitHub Copilot and Cursor — Into our test development workflow.
These tools help our QA teams to:
Implement automated test cases more quickly
Refactor and extend existing tests as features evolve
Reduce repetitive boilerplate when onboarding new brands
Navigate and understand large test suites more efficiently
AI is used only as productivity aid. All test logic, assertions, domain-specific decisions remain fully under engineer control, which is especially important in hardware-dependent automotive context.
Results And Impact
By combining automation with AI-assisted development, we scaled regression coverage alongside platform growth without a proportional increase in manual effort.
In practice, this means:
New features are covered by automated tests written by either QA engineers or developers, ensuring that regression-related technical debt does not accumulate over time.
Regression feedback has shifted from late in the release cycle to earlier stages of development, closer to code changes that introduced it.
AI-assisted development reduced test implementation time for common scenarios; on average, a complete flow – including both happy paths and edge cases – can be covered by automation within a day.
Manual testing is primarily focused on exploratory scenarios and edge cases where human insight provides the most value.
For a long time, AI solutions for mobile testing lagged their web-focused counterparts, leaving mobile QA largely reliant on manual and traditional automation approaches. Today, the mobile testing landscape is caching up. At OBDeleven, we are experimenting with AI agents that can execute test flows on mobile apps based on provided test cases. While still in early stages, these experiments have the potential to accelerate coverage further and reduce manual effort, especially with the flows that are hard or even impossible to automate via automated tests.
Key Takeaways
Manual regression testing does not scale with rapid feature and brand expansion
Automation provides the foundation for sustainable QA and reliable release cycles
AI-assisted development accelerates test implementation without replacing human judgement
Human expertise remains essential, especially for exploratory testing and edge cases
Mobile testing is an emerging frontier where AI has the potential to further enhance coverage and efficiency
What’s Next
We are experimenting with AI agents that execute mobile test flows, which are difficult to automate using traditional approaches. These experiments run in Firebase Test Lab, where agents receive predefined test cases written in natural language and perform actions on emulators.
While promising, this approach comes with several challenges:
AI-driven interactions can be non-deterministic
Dynamic UI changes can confuse the agent
Execution cost and duration are higher than traditional automation
Hardware dependencies (vehicle control units, OBDeleven devices) cannot yet be tested
For now, AI agents complement rather than replace automation. They are best suited for exploratory testing and uncovering unexpected behaviour, while critical regression paths remain covered by stable, scripted tests and targeted manual testing.
In parallel, we are exploring the use of AI to analyse logs and metrics to identify patterns and anomalies.
Our goal is to continue combining automation, AI assistance, and human expertise to deliver high-quality releases faster while keeping QA efficient and scalable.