QAk-QAk about AI: quality has not gone anywhere, it has become more important (Category AI4SDLC)
Out. finale The fifth season of the podcast QAk-QAk – and in Production, where I was invited to talk about AI, quality and how to fit into the new pace of change. We recorded it back in May and I think it was quite interesting. AI is changing the definition of working within SDLC. In the past, IT professionals often told other industries about digital transformation. (digital transformation)How to digitize processes, move data online, speed up business. Now a similar wave has come inside the development itself. We have begun to transform ourselves.
I talked about how AI affects quality. (This is a podcast for quality assurance engineers.) And it seems to me that AI doesn't negate quality, it raises the price of engineering maturity. If a team already has contracts, tests, normal architecture, understandable project context, and a habit of checking the outcome, agents can really give a strong boost. If a team has chaos, weak boundaries, incomprehensible demands and tests somewhere near religion, AI often just scales that chaos.
In the conversation, we discussed the value of experience and years lived – they do not in themselves interfere with the use of AI tools. On the contrary: experience helps to formulate intention, to tell the agent where to look, to evaluate the plan, to understand that the result looks wrong, and in time to stop the beautiful but wrong generation. The problem is that an experienced engineer often has the pain of changing his or her routine. People want to do a good job, and the industry has changed its definition.
Separately, I tried to talk about the engineering binding on the example of my pet project.System Design Space" The story was about how the AI project gradually grew binding: visual checks, light and dark themes, contrast, Lighthouse, glossary of terms, rules for text, technical debt for the checks themselves. It's not a story about how I plugged in a bunch of MCPs, skills or other fancy tools and then went to do business logic. Quite the opposite: the tie-up grew out of real points of control where there were errors in the project or I knew that a potential mistake would be expensive.
In general, if earlier quality was often perceived as a separate phase after development, then in agent development, verification increasingly goes to the left. (shift left). The trend has been in mature engineering systems before, but now there is no pipe at all. In general, now you first need to understand what exactly was asked to do, how to check it, what restrictions to give the agent, what contracts do not break, what tests should be red before implementation. In a sense, the old ideas of TDD/BDD are returning not as a methodological dispute, but as a practical language for managing agents.
It's not enough for teams to "try AI" - they need to figure out what part of the job they want to improve: write code faster, read requirements better, maintain legacy cheaper, verify changes more accurately, and disassemble incidents faster. And then build the outline: baseline, metrics, checks, responsibility, human review. Without this, the conversation quickly turns into a competition for the number of generated text and green ticks.
In the release, we covered broader risks: what will happen to the labor market if part of the work really becomes autonomous; why strong teams can accelerate faster than weak ones; how trust changes when text, voice and video become easily generated; why security in the AI world becomes even more tense. There are no easy answers, and to be honest, I don’t really like confident predictions about this topic.
For me, the practical conclusion is calmer: do not wait for someone from above to bring a map of the new world. You should do small experiments yourself, collect your harness and evals, learn to delegate work to agents, but not delegate thinking and responsibility to them. AI amplifies the system well. So the big question remains an old engineering question: what kind of system are we allowing it to strengthen?
#AI #AI4SDLC #Engineering #QA #Management #Agents