Thoughts and Threads on AI

Thoughts and Threads on AI is a library of my talks about what changes when thinking gets cheap: whose work goes first, and what stays scarce. The talks share one pool of slides, and each talk is a path through it. When I fix a slide, every talk that uses it gets the fix.

AI moves too fast for a talk to stay true for long. So the library dates everything.

The first talk

"When thinking gets cheap" (Five things that stay scarce, and what work becomes) is the first talk and the featured one. I first gave it at Troglodyte Society #3 on October 13, 2026, and revised it the next day. It runs 43 slides and about 47 minutes. The short version takes 35.

The argument runs in four parts:

The past. Every new medium moved more of the mind outside the body, and every era made one input cheap. Value moved to whatever stayed scarce. The decades in between were the hard part.
The present. Intelligence is getting cheap first wherever work can be graded, and that already changes what a day of work looks like.
The situation. Most of this is already here. Capability runs years ahead of the world. Five things stay scarce: verification, context, atoms, legitimacy, and direction. The players are already fighting over them. The part ends on alignment, which means getting AI to do what we actually intend. Markets show the human side of the problem: a system that chases one number, and that nobody can switch off. The people building the models say the machine side isn't solved either.
The questions. How we get through the transition, and who decides what good is: for the models and for our institutions. Then the fork. Work could shift from doing things to choosing things and standing behind them, or the gains could collect with whoever owns the compute. Then what each of us does now. The talk ends on fifteen questions the future is waiting on, one for each scarcity at three levels: you, your company, and society. It asks the room one thing: which of the five would you least like us to get wrong, and what's your say in it?

The last slide: Not a forecast. A choice. Use the agency you have to build the future you want to see.

What else is in it

12 short talks. Each takes one idea through then, now, and next, and ends on a question. "Reading the clock" asks how fast this is moving and how you'd know. "Machines doing science" starts from a plain fact: producing results got cheap, and checking them didn't.
24 themes. Nobody hand-orders these. The library assembles each theme into a Then, Now, Next walk from the slides tagged with it, and a new slide slots in without reshuffling the rest.
73 open questions. Each one opens its case: what the slides show, and why that doesn't settle it.
The material underneath. 145 slides, 102 of them deep dives, and 62 Background cards on the thinkers and history the talks draw on, from economics, history, philosophy and politics, computing and AI, and movements and memes.

How it keeps itself honest

Every slide says what kind of claim it makes: evidence, history, interpretation, opinion, forecast, question, lens, or guide. Every slide also says when it was true. Opinions and forecasts keep a dated history, so when I change my mind, the old view stays on the slide next to the new one. Each time I give a talk, the library records where and when. If the talk changes afterward, the library marks it "Changed since it was given."

Ways in

The library. The whole network of ideas fills the screen and turns slowly. Point at a talk and its route through the network lights up. Choose it and you fly into the talk without a page load.
The map. A talk runs left to right. Questions rise above it, deep dives hang below, and the Background sits underneath. Press G and every idea in the library becomes one 3D network. Press T to pick another talk, or A for a slide's notes, sources, and related slides.
The deck. Any talk as plain slides, with related ideas on a rail under each one. Click one to read it without losing your place.

Start with the talk. Then follow whatever thread pulls at you.