All-In on AI, Coding, Having Fun

It’s been a while since I’ve written here.

I want to be honest about that, not with an apology, but with an actual answer. Because I wasn’t gone. I was deep in something.

---

Last year, I built a product called Helm. It’s a hardware-software focus device for your phone and your desktop. The idea had been sitting with me for years: physical and digital, linked. You touch the device, the system activates.

What I did not know when I started: that product required a backend, a mobile app with native system access, a device ID system, an admin panel, and a React Native bridge connecting two native layers.

I also did not know what a backend was. That is not a metaphor. I thought it had something to do with American football.

The crisis hit in late June. I had $20,000 in a factory, a ghost developer, and 200 damaged chips that needed 200 unique device IDs in 48 hours. Ship date: mid-July. My son was born in early August.

I took the codebase into Claude Code and started asking it to explain everything to me in plain English.

The device IDs: I generated them myself. First real output. Then I kept going.

By August the product shipped. A few months later, I walked two engineers through what I’d built: the backend, the iOS and Android apps, the bridge. One of them said he’d never seen a non-technical founder figure that out. The other asked if the CTO of Cloudflare could interview me.

I tell you that not to brag. I’m still not a developer. What changed is that I can close the gap now: describe what needs to exist, understand the answer, push back when something doesn’t add up.

The old bottleneck was the execution gap. Helm made that real in the most expensive, stressful, and clarifying way possible.

---

That was the accidental misogi. Here’s what came after it.

The same method that got Helm shipped now runs my whole business. I have a team of AI agents that handle content, ops, data, and research while I sleep. I built the infrastructure (OpenClaw) that keeps them running. I’ve been coding useful things and open-sourcing pieces of the system.

One of those: a Claude Code skill called diagram-design. I wanted good-looking diagrams for my blog posts. Claude kept making generic ones that looked nothing like my site. So I built a skill that reads your website’s color palette, extracts your fonts, and generates 13 types of editorial diagrams that actually match your brand, in about 60 seconds.

It’s almost at 1,000 GitHub stars. Which I did not see coming.

---

I’m building a course on all of this, not the tools, but the method. How to go from plain-English intent to working software, running agents, and useful things you’ve actually made.

If that’s why you signed up here: we’re getting there.

In the meantime, I’m going to be writing again. Short. Specific. About what’s actually happening and what I’m actually learning.

That’s what LittleMight is for.

---

One thing: what are you trying to build? What do you want AI to help you make? What feels hard about getting started?

---