How We Use AI in Our QA Delivery Process
As software delivery continues to accelerate, the role of quality assurance is evolving with it. QA today is no longer only about finding defects. It is also about enabling teams to move faster, more consistently, and with better coverage across the entire delivery process. For us, AI is no longer an experimental add-on. It is increasingly becoming part of how we work every day.
As a QA Lead, I see the biggest value of AI not simply in saving time. More importantly, it helps us make our testing operations more scalable. We can get automation-ready test cases faster, create initial scripts sooner, reduce manual administration, and allow the team to focus more on the testing tasks that truly require human expertise.
AI in Test Case Preparation
One of the most important areas where we use AI is Zephyr test case generation. The goal here is not to generate large volumes of test cases without control. The goal is to move faster and more consistently from functional specifications to a test set that is ready for automation or for manual test cycles.
In practice, this means that AI creates Zephyr test cases based on Confluence specifications. These generated test cases do not go straight into final use. They go through a review process where we clearly separate specification issues from AI tool-related issues. This distinction is essential, because it is the only way to integrate AI into QA operations in a sustainable and reliable way.
Our experience shows that AI is especially strong when it comes to well-structured, well-specified functional flows. It performs particularly well with base flows, key alternative paths, and standardized test case structures. This gives the team a faster way to build a sufficient number of Approved test cases that are suitable for automation or for manual QA work.
AI-Supported Script Generation in Test Automation
The next major step is using AI not only for test case preparation, but also for scripting itself.
In our automation process, we use an AI-based solution that generates an initial automation script based on the content of Zephyr test cases and project-specific knowledge. This is not a black box that replaces the tester. It is a strong starting point. The automation tester still plays a critical role by executing, refining, validating, and correcting the result when needed.
We see the best outcomes where recurring patterns, well-defined framework rules, and similar previous solutions are already available. In those cases, AI significantly reduces the time needed to start scripting from scratch. The direct result is simple: more new AutoTCs can be delivered within the same period.
When AI Can Also Execute and Attempt to Fix
One of our most exciting development directions is a new autotest skill that goes beyond traditional script generation.
This solution not only creates the initial script, but can also execute it automatically. In certain cases, it is able to identify smaller issues, attempt self-healing corrections, and rerun the script. When it encounters a critical problem that requires real judgment or deeper intervention, human involvement is still necessary.
In practice, this creates an important balance. We do not take control away from the tester, but AI removes a number of repetitive, lower-value cycles from the process. The early feedback has been very positive. We can already see that the workflow is becoming faster, not only in the creation of new automations, but also in the maintenance of existing scripts.
Lower Maintenance Effort, More Focus on Valuable Work
One of the biggest hidden costs of test automation is maintenance. Creating a new script is only the first step. The real challenge is keeping it stable, readable, and executable over time.
AI helps here as well. It supports the maintenance of existing AutoTCs and also contributes to meeting code quality expectations. For example, our SonarQube-related AI skill can automatically fix certain categories of Sonar issues. This is particularly useful on projects where new automation scripts are actively being developed and newly created scripts need to comply with quality standards and coding guidelines.
The overall impact is that the team spends less time on technical cleanup and more time on work that creates real business value: improving coverage, securing critical flows, and providing faster regression feedback.
Review and Governance Still Matter
It is important to emphasize that AI does not replace professional control in our process. On the contrary, the more AI we introduce, the more important review and governance become.
Generated test cases must be reviewed. Script review is mandatory. Traceability must be maintained. We have clear rules for when an issue is a specification problem, when it is an AI tool issue, and when a test case can truly be considered approved. The same applies to automation status handling and Zephyr administration. Without reliable data, there is no meaningful performance measurement and no sustainable automation operation.
From a leadership perspective, this is one of the biggest lessons. AI should only be scaled within a controlled process. Being faster is not enough. We also need to remain reliable.
Where We Are Heading Next
What we have today is only the beginning. Our next important direction is specification-based support for E2E test case generation, as well as the development of automated test coverage docs.
The second area is especially promising, because in the longer term it can make it possible to see, with minimal human intervention, which specifications already have Zephyr coverage, where automated coverage exists, where gaps or duplicates remain, and where additional AI-generated test cases would bring the most value.
This is not only an efficiency topic. It is a strategic QA capability. The better we understand coverage, the more deliberately we can prioritize automation.
Final Thoughts
For us, AI in QA is not a flashy extra. It is becoming an operational layer. It helps prepare test cases, generate scripts, execute them, fix certain issues, reduce maintenance effort, and build a more effective automation process.
But the most important point is not the technology itself. It is how we integrate it into delivery. For us, AI creates value when it accelerates work while remaining transparent. When it reduces manual effort without weakening professional control. And when it leads not only to more automation, but to better QA operations overall.
That is what we are building now — and this is only the beginning.
