I am a professional AI practitioner and I feel that I understand the field well enough to see multiple possible paths towards actually creating AGI. They are certainly out of my own personal reach and/or skillset right now, but that doesn't mean they're impossible. And yeah, "AI will create AGI" is kind of purposefully vague, but I think it's still valid. I think the flaws we unconsciously introduce into AI through our biases as human beings are what holds it back, so the more layers of stable, semi-stochastic abstraction we can place between ourselves and the final product, the more likely the model will be able to optimize itself to a place where it is truly free of the shortcomings of being "designed".

Edit: realized I came off as bit cocky there, apologies. I value your opinion and appreciate you taking the time to share it. I also think I see where you're coming from and partially agree -- the AI systems that are popular right now probably won't create AGI, but I still believe that AGI will be created with the help of non-general AIs.