On Augmenting Experts

I’m the slowest of my bike-riding friends – I’m a swimmer; I can deal with waves, currents, and even too much chlorine – but there’s something about a 45-degree incline wall of dirt where you have to fight for both forward motion and balance that makes it less fun for me.

Now, give me an e-bike, and it’s a whole new ballgame. I suddenly keep up with the best of them. And I’ll argue that I’m having more fun than all of them – combined.

But give them an e-bike (at least the ones not purist enough to accept it), and the gap doesn’t reset; it actually widens. Now Im back to being maxed out, running out of battery earlier than everyone, and further behind than with the original bike. Why?

An expert, given an augmenting tool, will take even more advantage of it than a normal person. This happens with e-bikes, and it happens with AI tools.

Give me any LLM-enabled code editor, and I’ll crank out Python scripts and stitch together a workflow in a few hours – compared to the days it used to take me. Give the same editor to Gerardo – same biking friend; bastard is also a far better programmer than me – and he’ll produce code that handles edge cases I never envisioned and is optimized for bigger datasets.

Same e-bike/LLM, but his output is 10x mine.

This is my framework for user-facing AI tools right now. Use it as a replacement for an expert, and you get faster – but expected – results. Augment experts, and now you have exponential outcomes.