Me, happily using AI in my own way
"All happy families are alike;
each unhappy family is unhappy in its own way."
"All unhappy sessions with AI are alike;
each happy session is happy in its own way."
Reverse Tolstoy Principle
All unhappy sessions with AI are alike
If you're having a bad time using AI, you're cognitively overloaded.
You had a very limited budget of attention, and it got misallocated.
(Semi-related: why do experts get way more out of AI systems? Larger attention budget within their expertise!)
There's no recovery from this once it happens. Your pre-LLM instincts will tell you to stop everything and tackle the misunderstandings one at a time. You can still do this, but not if the system is generating things to confuse you faster than you can work towards making sense.
Each happy session is happy in its own way
If you're having a good time using AI, you've somehow spent your attention budget just right.
For me, it's usually dumb luck. But maybe you asked a really excellent question that sends the LLM off looking for just the right resources which elucidate your concern at just the right level of abstraction. Or maybe your Acceptance Criteria was rock-solid, and ambitious but not too insane, so the AI spends a lot of tokens and brings back something delightful to verify.
When this happens, you really are harnessing (it's a good word!) the power of AI to traverse the latent space of possibilities. You're going places you never would have been able to go, on your own digital magic school bus. "Verification" at its best feels more like discovery:
I'm washing the dishes and I hear a clatter from the attic. I had set up a camera and told claude to knock over a shoebox with a toy robot, but the loop was so slow I just walked away and washed dishes until the clatter (I couldn't find footage of teh clatter, but here's the robot "pushing" the box real good).
I'm playing chess in a web browser, adjusting the sliders until I can predict the engine's move every time.
I'm making coffee and I hear my thermal printer kick off a morning report of what happened over the last 24 hours, which I can linger over without even opening my laptop.
And what makes it the Reverse Tolstoy Principle is that each of these happy sessions is totally unique! Making a chess engine is nothing like getting a robot to knock over a shoebox, and neither of those share anything at all with looking over a thermal printed report while drinking my coffee, thinking about what I want to do in the world of bits before cracking open this laptop I love and loath so much.
Reverse Tolstoy Principle says: are your days blending together? That's not good. You are not using AI right. Was yesterday totally different than today? Good. Your harness (it's a very good word!) is taking you places.
(I will be writing & sharing more about the thermal printer soon, hopefully. It was less than $100 on amazon, a breeze (with Claude) to install, and now it becomes another gate on my attention.)