I totally get that. Yeah, for one not having access to the very best models will produce that frustration. Also experience leads to better prompts (to avoid past frustrations).
For now, there are some specific limitations, like being able to think about a specific large code base all at once. But there are paths to overcoming that, some of which are live if difficult to use at the moment, like vector embedding databases that really start to overcome that. Another path would be automating the creation of these councils/swarms of AI agents allowing you to spin your an entire corporation of LLM individuals in seconds.
But yeah basically every day I write at least one prompt which gets me exactly the kind of “previously had to be in the hands of humans” result. I don’t know if we really need another 100x in capability or just more mature infrastructure to use what we have. Blows my mind. And also convinces me that humans are on our way out of the economy unless we stop playing with this new kind of fire. 😬