My thoughts on AI

I've been thinking a lot about where the opportunities are in this AI wave we're riding, and I think most people are looking in the wrong direction.

Every gold rush has two types of people. The ones panning the river everyone already knows about, and the ones who spent a season learning which bend has gold nobody else bothered to check.

AI right now is the river. Everybody's panning the same water, all trying to get better at prompting the same general model.

I think that's the wrong bet.

AI is the most generalist tool anyone's ever built, and that's exactly the problem. Ask it to do something specific to your world and it needs you to hold its hand through every step.

Someone has to already know the terrain, the resources, the workflows, the actual tools, before AI can do anything useful in it. That's the bend in the river, and knowing it takes a season nobody else spent.

We're seeing this in design already. People are staking narrow claims in narrow niches (rams.ai is one), and none of it is simple to dig. It takes real, layered work that only someone who's walked the ground can do.

That's basically the early SaaS story again. The winners back then weren't the ones waiting for the simple version. They were the ones willing to build a complicated rig just to pull what's now a basic CRM out of the ground.

It was hard, and the hard part was the moat around the claim.

Two ways this river runs

I keep flipping between two futures for this.

Scenario 1: the river runs dry

Every technology follows some kind of S-curve eventually. It gets very good, then plateaus before it can find the good bends on its own.

If that's what happens, the people who staked a claim on a specific bend right now have already won. They'll have the workflows and the infrastructure built before the river can wash out from under them.

Scenario 2: the river keeps rising

It never runs dry. It keeps getting better until it can find every good bend itself, from nothing but a plain description of where to look.

That's most of the way to AGI. If we get there, every domain gets panned out the way code is getting panned out right now. Anyone who can describe a bend gets the bend.

So who wins in scenario two? Maybe just whoever pointed at it first.

The cascading effect

Except it doesn't stop there. If design gets panned out, "make it a little better" stops mattering, because everyone already has good design pulled straight from the river.

The advantage moves to whoever can sell what they pulled out best. And the second the river gets good at selling too, it moves again, one bend further upstream.

You can chase this upstream forever, always asking what the next bend is.

So what actually decides the winners?

The only thing that stops the whole chase is the river running dry somewhere. And it will, because everything in nature runs on some version of a log curve.

Nothing gets an exemption from a ceiling like that, not even engines. The most efficient petrol engine ever built tops out around 48%, and most road cars run between 20% and 40%_engines).

Nothing gets to 100.

So the real signal is watching where the river runs dry. Whoever plants their flag on the bend right before that happens, in whichever domain it happens in first, keeps that bend.

I don't know where the water stops yet. Nobody does. But being good at panning the same water everyone else pans isn't going to be the skill. Owning the bend is.

At least for now.

What I'm actually doing with this

I've started staking my own bend instead of panning the general river with everyone else. Not something anyone could use. Something built around ground I already understand better than most people currently building AI tools do.

Because whichever way the river runs, the version of me who knows this bend, and knows how to point the water at it, stays useful.

The question was never "will the river wash over my job." It's "can I stake my bend before somebody else notices it's there."

That's the race.