Despite the current trend across the industry of cutting QA headcount, I think that in the long term, we will see an increase in specialized QA demand.
I’m saying this after creating four applications I use daily using LLM-generated code with Claude Code and Google’s Antigravity.
While the most cutting-edge LLMs are extremely proficient at generating new code, they are also bug-generating machines. I think this happens because LLMs lack human-level taste and criteria.
On some occasions, I have spent several hours explaining why a specific behavior is not desired in a product or application, to the point that it would have been faster to fix it myself.
On the corporate level, working and QAing AI-generated code, you can absolutely tell the difference between that and purely dev-written code. On testing sandboxes, you can see the type of bugs I haven’t seen in years. This is not a problem with the devs; they are using AI tools like the industry is currently demanding from them. This is the expected outcome of using a probabilistic-based tool like an LLM.
If developers keep building using AI, then bugs and other undesired behavior will proliferate. Not only that, but even if there are no functionality bugs, I have noticed that LLM-generated code also affects performance and user experience. Some “working” code ends up being a performance nightmare because it introduces hidden inefficiencies like redundant API calls or unoptimized database queries.
The last thing LLM introduces: Devs and QAs that don’t know how things work (known as cognitive debt) because they have created code and tests that are 90% AI. This is also a root for regressions and unexpected behaviour.
QAs that have been around for a while have developed a sense and a taste for where and how bugs or defects happen. If we pair those skills with AI-powered tools, then some volume of defects will be caught, but I don’t believe it will be enough.
Companies will eventually realize this, and they will need to double down on the QA headcount. I really don’t know when, but I think it will happen.
This is not a post to bash on AI. I’m just trying to say that every tool or technology brings not only benefits, but also downsides that we must be careful with. We have to be mindful of this to make sure we can get as many benefits as possible without falling into the potential downsides.