The Way of Friction

For builders

This site is also a guide for people who build with code and with AI.

The way is graphs and loops, not vibe coding. See the system. Find what leans on what. Make a few precise moves. Watch what happens. Adjust.

If what you build with AI finds elegant solutions, people will love using it.

The principles

1. Map it before you build.

A problem looks endless until you draw the graph: the parts, and what leans on what.

From 2 · Look Closer and 3 · Many Small Frictions.

2. A bigger model is not a better answer.

More force, more tokens and more steps tire you, and the problem stays.

From 4 · The Bigger Hammer.

3. See before you act.

Context and data first: give the system what it needs to see before you ask it to move.

From 5 · The Warrior Stops and 7 · Data Is Sight.

4. Find the highest-leverage step.

Not the biggest task or the loudest error, but the one the others lean on, first.

From 6 · The One They Lean On.

5. Fewer, precise moves.

Start with the simplest system that works, and add a step only when it earns its place.

From 9 · Fewer Moves.

6. Small things, fix fast. Big things, step around.

Clear a small error in one move, and route around a big one instead of meeting it head on.

From 10 · One Comes, You Cut. One Comes, You Slip.

7. Loop: act, observe, adjust.

Every run is a round, measured with evals, not with a feeling; when you are wrong, now you know.

From 15 · Practice.

8. Machines see, you choose.

Let the machine search and show you; keep a person at the decisions that matter.

From 8 · Many Small Eyes.

9. Elegance is the goal.

Less motion, more done, until it feels simple and the way to the next thing opens.

From 13 · The Way Opens and 17 · Mastery.

What you don't understand looks endless.
Look closer.
No wall is one stone.