The Centaur Strategy

Dear Reader,

Garry Kasparov's hands begin to sweat.

Garry is the world's greatest chess player, but he has run out of moves.

His opponent sits there, quietly waiting for Garry to hand him the victory.

Garry realizes it's over...not only is he going to lose this game, chess has just changed forever.

That's because Garry's opponent was an AI.

On May 11, 1997, the world's leading chess player got beat by IBM's Deep Blue chess program, and it shocked the world.

The summer after that defeat, Garry did some soul-searching in the wake of his defeat by AI.

I'm sure you can relate.

Garry didn't give up. He ran some experiments with another chess player and ended up inventing something that would be called the Centaur Strategy.

Garry used a basic chess software to help him compute moves before he made them. This allowed him to find blind spots and offload the grunt work of calculating moves to the computer.

Garry didn't have supercomputers or IBM compute tokens, so he made do with what was available.

Years later when other chess players started using the strategy, they realized that he could beat the fancier chess programs and other chess masters.

human + computer was greater than either the human alone or the computer alone.

That's why the centaur name is so perfect because a centaur could totally kick the ass of both a human and a horse. The sum is greater than the parts.

Amateur players using the Centaur Strategy and mid-tier computers beat both elite grandmasters and massive supercomputers. They won because their collaborative process was superior.

You're probably already doing this without realizing it, but here's how to lean into the Centaur Strategy as a designer/builder with AI.

Process beats genius and compute

The beauty of the Centaur Strategy is that a competent human with a good process for AI beats either a lone design genius or a lone AI most of the time.

So think about your process and iterate on it.

Here are some things you might try:

• Have AI help with divergence and generating options
• You own convergence and choosing options
• Run 3-4 design directions in parallel
• Build pauses and decision points into your process

2 tests for a healthy centaur

The centaur strategy is about splitting up the work smartly, but AI is getting better every day, so the split will evolve. But there are two tests you can run on a task.

Test 1: Are you above average or do you want to become above average at something?

For example, I don't like AI to write for me, but some may prefer to hand over copy to AI. I've spent a lot of time trying to improve my writing, and I think it makes me a better thinker, so I will continue to struggle with writing copy.

That test may not look the same for you, and that's ok. That test may also look different in five years, and that's ok, too.

Test 2: Does it need a human touch regardless of "skill-level"?

AI will probably get better than humans at many things, but that doesn't mean you should let AI do everything it's "better" at.

For example, you will probably want to handle client and stakeholder communication yourself when real money is on the line.

Strategy passes both tests for almost everyone which is why it's worth treating strategy as a domain you own no matter what.

I like to think of Strategy as the Human Layer orchestrating AI and ensuring that the process is a human-driven collaboration.

Here's 2 ways to start:

• Start with the "why" (vision, goals, context), not the "what" (redesign this checkout flow)
• Avoid becoming that person who presses the approve button all day (you won't enjoy that or be good at it)

Cory Doctorow (author of Enshittification) has a new book that talks about what happens when you aren't in control of how you use AI, something he calls a "Reverse Centaur."

The process for AI-human collaboration is still being determined, so now is the right time to start taking ownership of your process.

The Centaur Strategy works best when you own the collaboration.

TL;DR: Don't be a purist designer who gives up on AI and don't turn into a lazy slopper that treats AI output like an oracle. Be the Centaur. Guide the technical powers of the machine with strategic thinking.

No More Mosquitoes

Tornyol, a Y-combinator startup, has invented a drone that kills mosquitoes mid-air for $50/month. I guess you'll need another drone to clean up the mess.

The internet is realizing that LLMs sound annoying because they're based on Reddit answers.

​Source​

​Source​

Pre-Call Intel Prompt

Use this to get some info and talking points before a call:

Adapted by Jeff Humble from this Source​

Talk to you in 2 weeks! ✌️

Jeff Humble ​Designer & Co-Founder ​The Fountain Institute

​