My AI approach at work that changed a bad situation into a workable solution

Work brings new challenges and in my case as an automation test engineer brought more complexity as I needed to join a data warehouse project. Everything is now low level, while the only tangible thing might be the input for it that I know quite well as I've tested the app producing it. But from here on there are new transformations that are taking the base data and aggregate it layer on top of layer into a multidimensional database. And this is a first point of failure in the SQL queries that are doing this transformation. And testing that alongside a terabytes of a database where any mistake pushes you back days, it is not an easy task. But there is a tool coming to the rescue and you've all guessed it by now: AI and my choice is Claude Sonnet 5+, Opus and Higher. And what AI did for me is to simplify all this complexity.

With the help of AI, I succeeded to translate requirements or issue into tangible items that I can run and verify. Wording and technical descriptions would scare any X years old by thinking that you speak alien, but with the help of Claude models those can be explained and approached with generated samples that can be used within the tests. This way a problem that I wouldn't even know from where to approach it, transforms into a puzzle that I can decipher. Not fully in control, but used as a tool, AI comes to the rescue. It is not all his doing as because I know that app that generates the input, everything seems natural to me after being explained. But the key factor here is that once I understanding, I am in control and I can run the tests with higher confidence factor. And as a software test engineer that's what's expected from me.

What I can say is that AI is not a magic solution and its use in the beginning of the project was quite faulty without any boundary in mind or considerations that the task at hand is to be run on a huge data. It misunderstood requirements, made assumptions that were not valid for specific implementations, generate SQLs that looks correct but produces wrong results or weren't even usable due to their duration. That is why I never treated its answers as absolute truth. Instead, I used the responses as a starting point for creating a realistic test plan. Every generated SQL query, every explanation and every testing idea got through my knowledge and thinking filter and I ended up making it be more precise to what I need. He was out of scope and I succeeded to steer it to simply validate what was needed without going for heuristic testing.

In the new project what changed compared with what I was previously doing, is not that AI does the work for me, but that it dramatically reduces the time needed to understand a new domain. Instead of spending hours reading documentation, jumping between meetings or waiting for someone to explain a concept that he doesn't even understand (yes, I've tried that), I can have an interactive conversation that adapts to my questions and steering until everything clicks. This allows me to become productive much faster and to test something more relevant for the problem at hand.

For me, this experience proved that there is real value in AI by using it in software testing. It doesn't replace experience or critical thinking, but it rather amplifies them. It turns uncertainty into a structured learning process and helps bridge the gap between business requirements, technical implementation and testing. In projects where complexity is unavoidable, having such a companion has become less of a luxury and more of a necessity. And to be truthful, without it I am not sure I could even do this job. I don't want to destroy my neurons with low level technical thinking, but rather with what's important and to reach the testing objectives I feel will help with a smooth production.