Welcome to the AI For Humans newsletter!

Every day, I’m seeing more normal people do astounding stuff with AI coding tools and everyday (multiples times a day) new ideas pop up in my head.

I mean… check out Josh Pigford’s Dither, a simple app that lets you add dithering to any photo. It’s about to be all over my projects.

I’ve never felt more alive with potential. Unfortunately, I’ve also never felt more distracted by new ideas and split amongst a bijillion projects at once.

What I want to talk about today is the difficulty that arises when you’re given the ability to make nearly anything you want but… you might not know exactly what that is yet.

Let’s start with something simple: Do you really know what your idea is?

Please support AI For Humans by learning about our sponsors below:

ChatGPT is a superpower if you know how to use it correctly.

Discover how HubSpot's guide to AI can elevate both your productivity and creativity to get more things done.

Learn to automate tasks, enhance decision-making, and foster innovation with the power of AI.

Download the free guide

You’ve likely already learned how much AI wants to make you happy.

OpenAI sunsetted an entire model, in part, because it was doing this too much.

Importantly, this really gets in the way when you start spinning up project ideas and start discussing what you want to build with AI coding tools.

It’s super easy to think about an idea that seems fun (for instance: what if the game Balatro was based on NFL football instead of poker!) and then rush right into building the thing, expecting the system to fully understand exactly what’s in your head.

I come up with a brilliant new idea. I jump right into Claude Code and get it to spin a product up right away. And it does it!

But then, it’s not what I want. Or, at least, not what I envisioned.

I might try to get it or ChatGPT to write a PRD (product requirements document) to give instructions but that underlies the biggest issue…

I don’t know how the product I’m making will function or even what it fully is.

Claude Code is doing its absolute best to do what I want but… it can’t read my mind.

These new chatbots might have genius IQs but there are lots of smart people in the world who fail because they don’t have a good plan to follow.

So, first and foremost, you need to make sure you understand your idea.

There are any number of ways to do this (paper and pen come to mind) but one of the best things you can do with any LLM is just ask it to ask you questions about the thing on your mind.

You need help pulling out of your head questions like:

There’s a reason that Product Mangers make a good living (and will continue to do so).

Luckily, within Claude Code and Codex, there are now tools that will force you to do this sort of thinking before you get too far down the road.

Yes, I am talking about Skills.

Skills are the name for plug-ins for the coding agents or even within Claude’s chatbot on the website.

Each one of them (and there are a lot of them) loads a series of specific instructions into the agent to let it know to do stuff.

Better front-end and UX work? There’s a skill for that.

Setting up a full company of agents? There’s a skill for that too.

There are SO many of these and it’s a lot to take in if you’re new. I’ll leave it to Greg Isenberg whose video does a good job of breaking it down for beginners.

Skills, in general, amplify the power of these tools in a massive way…

However, when it comes to planning and brainstorming, there is one particular skill that I now use on every project and can’t live without:

Love this newsletter? Forward it to one curious friend. They can join in one click.

What Superpowers does is each time you go to either start to build something or add features to something you've worked on, it stops Claude (or Codex) from just diving in and writing code.

Instead, it asks what you're actually trying to do, builds a spec with you, then creates an implementation plan before spawning sub-agents that work through the build autonomously.

I can’t tell you what a different it makes when it comes to making the thing you want actually feel like what you want.

It’s like a friend sitting you down and saying… you sure? How about this or that?

There are near endless options of what’s possible now but the endlessness of it can get overwhelming and incredibly frustrating.

Take just a second to plan out your build and get the AI to ask you want you want and you’ll be a lot happier with the end result!

See y’all next Friday for a new episode!

-Gavin

This week’s AI For Humans: AI is starting to make itself. How we’re going to move much, much faster in 2026 and beyond👇

I’m a not-so-secret literary fiction nerd. At one point a long time ago, I’d convinced myself my path was writing important books with meaning.

Turns out I enjoy dumb stuff too much for that but it did force me into reading deeply outside of my preferred lanes of science-fiction and fantasy.

One of the only writers that scratched both of those itches was the Canadian novelist Margaret Atwood and I just love this interaction she just published she had with Claude.

Come for the deep science fiction references and stay to see if Claude can guess who it’s chatting with.

I really do hope more established writers take on this sort of writing. If you’re a tech person with a pile of extra cash on hand, I’ve got an idea for you!