After launching two summits at the start of this year, I needed a break.

We worked incredibly hard to make those events as valuable as possible. By the way, almost all of the sessions are now available on the BigMoves AI YouTube channel.

There’s a ton of great stuff there.

But after all of that, I decided to take a month off.

Kind of.

Because apparently my version of taking a month off is deciding to build a production-ready app from scratch.

It started with an idea from a friend A friend gave me an idea that made me think about how people actually want to learn online.

Not everybody wants a chatbot.

Sometimes you don’t want to ask an AI a bunch of questions. You want to learn from a real person who has experience doing the thing you’re trying to do.

You want that creator’s perspective, their process, and the little details they’ve picked up along the way.

The problem is that YouTube isn’t really built for focused learning.

Here’s what I hate about it right now:

1. There’s too much fluff. Creators have to make videos that hit the right keywords, cover broad topics, and keep the algorithm happy. If you’re already more advanced, you can spend half the video waiting for the one thing you actually need.
   ​
2. A chatbot doesn’t replace learning from a person. AI can give you an answer, but sometimes you want to learn how a specific creator thinks. That’s a different experience.
   ​
   You still want AI to save you time. If 3 creators teach the same concept, you shouldn’t have to watch the same explanation three times. AI should be able to spot the overlap and help you focus on what’s actually new.
   ​
3. YouTube also doesn’t give you a structured curriculum.

Rabbit holes are great when you’re relaxing.

They’re terrible when you’re trying to stay focused and learn one specific topic.

So I built a tool that solves those problems. I’ve shared it with a few friends, and they love it.

I’ll be giving it to this community for free very soon. You’ll just need to pay for your own AI usage.

But the app itself isn’t really the point of this email.

The way I built it is.

What my “workday” looked like Even though I took most of each day off last month, I still built an entire production-ready app in my spare time.

A normal day looked something like this:

• I’d wake up, open Claude, and give it something to work on.
• Then I’d go eat or go to the gym.
• I’d come back, check what Claude had done, make a few UI tweaks, and give it another prompt.
• Then I’d go to breakfast. I’d come back an hour later.
• Sometimes Claude was finished.
• Sometimes it was still working.
• A task might take anywhere from 30 to 90 minutes.
• I’d spend about five minutes checking the output and give it the next prompt.
• Then I’d go play golf.
• When I got back, I’d check Claude again and keep things moving.
• Then I’d go to lunch.
• After that, I’d do something fun, come back, check the work, and prompt it again.

That was basically the pattern for the entire month.

Go do something fun.

Come back.

Check Claude.

Give it the next task.

Repeat.

When I add it all up, I probably spent about 20 to 25 hours actually sitting in front of my computer over the entire month.

There were a few integration steps Claude couldn’t do for me. I had to handle permissions, connect certain tools, and write some of the important prompts myself.

But outside of that, Claude built almost the entire application while I was enjoying my month off.

We’re not using these tools to their full capability This experience made something really obvious to me.

So many people still aren’t using AI coding tools anywhere close to their full capability.

If I can build an entire production-ready app during a technical “month off,” why aren’t we building an app every day or two?

Why aren’t we automating more of the repetitive workflows inside our businesses?

Why are so many useful ideas still sitting in a notes app?

Ultimately, I think it comes down to time.

These tools are powerful, but there’s still too much complexity in the process.

For example, I’d have Claude make a change and push it to GitHub.

Then I’d wait for GitHub and CodeRabbit to review the code.

I use CodeRabbit because it catches problems I don’t want to find later.

If CodeRabbit found a mistake, I’d take its suggested prompt, bring that back to Claude, wait for Claude to fix the issue, push it again, and start the review cycle over.

Claude pushes.

CodeRabbit reviews.

Claude fixes.

CodeRabbit reviews again.

It worked, but there was a lot of waiting around for changes.

I knew there had to be a faster way. I just hadn’t taken the time to dive into it yet.

Now I have.

I found a much faster way to build, but before I tell you about that tomorrow, there are three big mistakes I see people making with Claude Code.

Three big Claude Code mistakes

1. Trying to rebuild things that already exist A friend told me he was building a complete funnel builder.

My first question was, “Why?”

You don’t need to build a funnel builder to build a funnel anymore.

Just build the page.

AI makes it easy to create software, but that doesn’t mean every project should become a giant platform.

Use the tools and infrastructure that already exist. Build only the piece that creates the unique value.

2. Letting AI write every prompt For AI workflows inside your app, AI still isn’t very good at writing the final prompts you actually need.

The curriculum prompt inside my app is a good example. I had to write that myself.

Claude is great at building software, but it doesn’t fully understand the outcome you want from an AI workflow. It can connect tools like Trigger.dev and OpenRouter, but connecting the pieces isn’t the same as designing the intelligence behind them.

Knowing how to write strong prompts yourself is still one of the most valuable skills you can have.

3. Building a workflow that makes you babysit the AI

This was my biggest mistake.

I had automated the coding, but I hadn’t automated the momentum.

Every handoff depended on me.

I had to check whether Claude was finished, read the review, copy the feedback, write another prompt, and restart the process.

That’s better than writing every line of code yourself, but it still turns you into the project’s notification system.

The goal shouldn’t be to make AI faster while you keep managing every step.

The goal should be to create a system that keeps moving while you’re doing something else.

In the last week I’ve found a better way And honestly, it has changed my entire philosophy on how we should be building with AI.

In the next email, I’m going to break down exactly what I found and show you the better way to build.

I’ll explain how I solved the GitHub waiting problem, how I removed so many of those “waiting around for changes” loops, and how you can keep AI moving without constantly babysitting the process.

It’s actually easier than any other tool I’ve used during my entire 3-year AI career.

Watch for the next email.

Talk tomorrow,

Mitch