Working with AI faster?

I see one obvious issue with this funny video I came across the other day:

The guy is not wearing a VR headset so naturally his setup is far from optimal (sorry bro).

Of course there is an even simpler solution to this whole problem - Neuralink (another funny video):

But drilling a hole in one's skull is an acquired taste and probably also frowned upon as a hobby.

The underlying problem however is still, well, a problem. And that problem is - how do I work with AI faster?

“But Alix, AI is pretty slow today?”

It is I guess, but the question is not about how slow AI at producing stuff, the question is how fast I am at interacting.

My big problem with AI today is that it produces a lot of information that is hard to comprehend easily but it also produces information that is not complete enough to precisely describe something. An easy example of it is coding. So say you ask Codex to code up something for you and it produced a bunch of code. Maybe you're fancy and use a software factory for it and part of your pipeline is also diagrams and video/screenshots of it working. But those artifacts are a far cry from a mental model you would've had if you wrote that software yourself.

So you technically can go through all of that code and all of these diagrams and get a better mental model. Setting aside the fact that no one actually does this, you would still not have a full mental model because one missing piece would be the “rejected pathways” - things you would have thought about but haven't implemented. Another one would be trade offs you would have taken. Yet another might be small optimizations you implemented or haven't implemented. There are probably more things in the category that a diff between AI produced mental model and human produced mental model would have.

So going back to “work with AI faster”, this information density or mental model robustness let's call it to me is one aspect of “faster collaboration”.

Second and a more trivial aspect is purely velocity of comprehension - how quickly can I understand what AI is communicating to me, the wall of text or so-so looking diagrams that supposedly represent its internal mental model.

Last and probably less trivial thing to solve here is context switching. CPUs are good at it. Mushy brains - less so. And since AI is not fast enough yet to fly at the speed of thought and I am not fast enough to rapidly load and unload my brain with all information related to what we are working on, I end up having many AI conversations covering all sorts of things and projects that all require different context.

Hopefully a nascent field for AIEx (or agent UX or HAI) can solve for these 🤞