If higher bandwidth networking consisted primarily running more and more ethernet lines in parallel, you would most certainly agree that "networking has stagnated".

"Reasoning" and now "Agentic" AI systems are not some fundamental improvement on LLMs, they're just running roughly the same prior-gen LLMS, multiple times.

Hence the conclusion that LLM improvement has slowed down, if not stagnated entirely, and that we should not expect the improvements of switching to these "reasoning" systems to keep happening.