The 10X Threshold Applies To AI
Monopolies
In Zero to One, Peter Thiel argues that monopolies don't win by being a little better. They win by being 10X better. They have a secret, they're right about it, and the gap between them and everyone else is so wide that switching is obvious. The market capture is rapid, accelerating, and widespread.
That exact framing is useful for thinking about where AI tools are actually failing right now.
Never Change Process for 1.5X Improvement
Many AI tools offer a 1.5X improvement. Your email is a little more polished. Your first draft comes out a little faster. Your summary is a little tighter.
But think about what you actually have to do to get that 1.5X. You write the thing. You send it to the AI. You read what it gives back. You check whether it diluted your intent or sounds weird.
- You notice that it changed the date, not a big deal, just fix that.
- You notice it sounds harsh, no worries, fix that too.
- Why did it rephrase "meeting" as "get together". Weird, this is a company not a retreat. Oh well, just fix that.
You fix the parts that don't land. Then you use it. The email is definitely improved, but it took 3x as long to write it.
You think: "Oh well, better is better right? It sounds better than it did."
But that's a new process.
Process Change Is Expensive
Changing how you work is not free. It has a real cost in attention, coordination, and friction. And that cost compounds when things are already moving fast. If your team is shipping, iterating, responding to customers, the last thing you want is to introduce a new step into the workflow for a 1.5X gain.
The existing process works. It got you here. You know its failure modes. You know where it breaks and how to fix it. Swapping that out for a marginal improvement is a bad trade.
Compounding can help, but it's not the whole story
You might be thinking you would make a few of these changes at once. You note that there are 5 places you can get a 1.5x improvement. Together that makes 1.5^5 = ~7.5X speed increase. But this will mean nothing if it doesn't improve bottlenecks. Writing an email 7.5X faster may have almost no impact on revenue.
It's so important to optimize around bottlenecks and not to optimize things that have the opposite effect. I wrote more about this in The Bottleneck Is Never the Code.
10X Is a Different Category
A 10X improvement doesn't need a meeting. It doesn't need a pilot program. It doesn't need buy-in. It's the kind of improvement where you'd be stupid to ignore it. Where the old way looks embarrassing in comparison.
That's the bar. And most AI tools aren't clearing it for most tasks.
If you have to convince someone that the AI version is better, it probably isn't better enough. The improvements that matter are the ones where convincing isn't necessary.
Don't change your process for 1.5X. Wait for 10X. Or build it.
This post is brought to you by myself and Claude Opus 4.6.