How high are you?

Happy Saturday! Here are a few things on my mind...

Adam Wathan has a new podcast where he walks and talks about what's on his mind. And it's great.

ADAM'S MORNING WALK • EPISODE 1

The Tailwind business ain't what it used to be

29:13

MORE INFO

View the podcast website

How high is customer demand?

Successful startups don't create demand. They position themselves in the path of existing demand. In a good market, customers provide their own motivation.

I made a video about this:

Your job isn't to create customer motivation; it's to find where it already exists and then build something that satisfies it.

AI prompting still feels like rolling the dice

Despite how good the new LLMs are, it still feels like with many tasks, I'm stuck in this loop: prompt, re-prompt, re-prompt, re-prompt, trying to get the output I actually want.

This is especially evident when you're trying to do anything with graphics, video, images, or animation.

In the video above, I was trying to make these little animated interstitials. I wanted to show a man busting through a brick wall to buy a product (to illustrate how, when someone really wants something, they'll overcome enormous friction).

I had a decent starter image. But I spent hours trying to get Grok to animate this scene. Sometimes it got the wall right but messed up the other side. Most of the time, it just failed.

It feels like this happens partly because the "AI chat" UI is such a blunt instrument. You type a prompt, wait for it to grind, get the output, and it's wrong. So now you have to regenerate the whole thing. You're just rolling the dice again, hoping you get lucky.

Even when you try to change just one or two variables and re-prompt, you never know what you'll get. Sometimes it goes in a totally different direction, and you're like, "No, what you had before was pretty good! I just needed to adjust a few things."

Opportunity? Better UIs for AIs?

It seems a ton of money is being spent improving the models themselves, but there hasn't been much innovation on the UI side.

Little apps are trying to help—I found one called f-stop that gives you multiple UI inputs for better prompts—but even then, you're still delivering everything in a single chat input.

It would be cool to be able to provide multiple distinct inputs and receive multiple output parts. To lock certain variables and only change others. To iterate without having to regenerate everything from scratch every single time.

Not sure if any of this makes sense. I sent Adam Wathan a bunch of voice memos about this: