Not saying krafton deserves the benefit of the doubt but:

Understand that “agentic AI” is almost entirely a buzzword that means “Microservices with an LLM somewhere in the mix”. Which… is what people are already doing.

Yes, there are some (idiots) who think that means EVERY single node in the graph needs to be an LLM and fuck the planet, Jensen needs a new zipper. But, by and large, what that means is they are using the exact same infrastructure they were last week but MAYBE added an LLM for preprocessing or postprocessing. It makes management happy because “We are using AI” and it makes everyone else happy because they can keep using the tools that actually work.