Course Launch, Sleep Book, and What I'm Doing With OpenClaw

Lots of things to share today from what I've been building.

Ever since bringing AI into my daily workflow, my output has been compounding week after week. With OpenClaw, the pace is picking up again.

So I’m experimenting with a new newsletter format. Occasionally, instead of focusing on a single theme, I’ll cover a few ideas that have been on my mind.

Today I’m writing about the AI Game Dev Course that I just launched, which is a free course for my subscribers. And the sleep book that I’ve been writing for the last two years, which is finally coming out in April.

Then on OpenClaw, which is this personal AI assistant system that has been in the news everywhere for the past month, especially after OpenAI bought it after three weeks of the project going live. I’ve been setting up a system and I want to share my learning so far from the first two weeks of building a cluster of OpenClaw bots.

AI Game Dev Course

Last year I started putting together a course on building mobile games with Claude Code. It didn’t come together, because I felt the tooling was changing so fast. Every week something changed in how Claude Code worked, and what felt like a solid approach on Monday looked incomplete by Friday. I decided to wait.

Now, in Feb 2026, things have stabilized. Using AI to build mobile games has gone from something experimental to something with a clear, learnable workflow. That’s the right moment to teach it.

I’ve spent 15 years making mobile games. I’ve watched the industry go through several major shifts: from premium to free-to-play in the early 2010s, from feature phones to smartphones before that, from small teams to large studios and back again. Each time, the shift looked uncertain from inside it and obvious in retrospect. Each time, the people who figured out the new way early had a real advantage. Claude Code feels like one of those shifts, if not the biggest, and 15 years of building games gives me a useful frame for evaluating what’s actually different now.

I’ve been deep in it. Learning what works, testing what breaks, figuring out where the real leverage is when you’re collaborating with an AI on a codebase. The workflow is different from anything I’ve used before.

That experience is what the course is built on. 12 lessons across 4 parts, starting from the first prototype and ending at a published App Store build. The case study is Read Clash, a speed reading game for iOS. Every lesson uses real working sessions, real decisions, and real code.

7 lessons are live now and the rest will follow over the coming weeks. What the course is actually about isn’t which buttons to press in Claude Code. It’s how to think about building with AI: how to break down a problem, how to communicate intent clearly to a model, how to stay in control of everything. Those skills hold regardless of what ships next month.

You don’t need to be a developer to follow along. You need to want to build something, even if it’s not the right thing, and be willing to work in a new way. The tools are capable. The workflow is learnable. The gap between an idea and a working prototype has never been smaller.

The course is free. Sign up here:

https://joakimachren.com/courses/ai-game-dev

More lessons are coming. I’ll keep you updated here as each one goes live.

Sleep Again book

My sleep book comes out April 9. I’ve been working on it for two years and I want to tell you what it is, because I haven’t said much about it here.

In 2019 I burned out.

My sleep had been fragile for years, but that was when I could no longer manage around it. I was waking at 4am, lying there running through decisions I could not act on until morning. I had been building companies for 15 years. Sleep was always the constraint I worked around instead of fixed. During burnout recovery, I slowed down and started sleeping again.

Then we started F4 Fund, and it deteriorated again. Everything that had worked stopped working. So I went deep into the research. Sleep science, chronobiology, and how founder stress disrupts sleep in ways standard advice does not address. I interviewed other founders. I ran experiments on myself. This book is the result.

It covers why entrepreneurs sleep worse than most people, and what to do about it given the life founders actually lead. There is a chapter on chronotypes and scheduling around your biology. One on the 3am wake up and what is happening physiologically. One on caffeine timing, because most founders shorten their sleep window without realizing it. One on overtraining, which took me three years to understand in my own case.

This is for founders whose brains do not switch off at night and who have already tried the usual tips. Most sleep advice assumes a predictable schedule and a stable cortisol curve. Founders do not have those.

I am looking for first readers. If you read it before launch and leave an honest review on Amazon or Goodreads by April 9, I will send you a free ebook copy now. A few sentences is enough. A new book lives or dies on its first reviews, and early ones from real readers matter more than anything else I could do.

I’m looking for 40 ARC readers and its first come, first served. Sign up at https://BookHip.com/SWHWWTL and I’ll send a copy in early March.

If you want to wait for the book to come out, sign up on the Sleep Again website to get notified. Hardcover, paperback, ebook and audio, all coming April 9th.

Your AI Co-Founder

It feels like the entire OpenClaw discussion is orbiting the wrong center.

People are excited about email triage, scheduled automations, timed events. But none of that is new. We could do most of it with Zapier and cron jobs years ago. Automation isn’t the breakthrough.

The real shift is this: we are getting close to fully autonomous AI founders.

Running agents 24/7 isn’t new either. I’ve had Claude Code grinding away on my machine for hours, executing long task chains without me touching the keyboard. That was already possible. OpenClaw doesn’t invent autonomy. It lowers the friction and expands the surface area.

The inflection point is when we stop thinking about AI as a productivity assistant and start treating it like a lean startup operator.

AI is natively built for experimentation. It can run A/B tests, collect data, optimize funnels, rewrite landing pages, iterate pricing, test positioning, all without ego and without fatigue. That’s where the leverage is. A founder that never sleeps and optimizes purely on signal.

My focus now isn’t on more clever workflows. It’s on turning these systems into AI founders. Entities that launch, test, measure, and adapt continuously.

There are two bottlenecks:

1. I’ve experimented with OpenClaw as an AI founder already. Here’s one of the many systems I’ve been experimenting with for the AI founder: ideate mobile games, many of them. In each stage, the AI founder does the work with several sub-agents.

   

   But the struggle is real. It can come up with ingenious ideas worth prototyping, but it doesn’t feel like it is inventing anything new. It’s very much like throwing stuff at the wall and seeing what sticks. This is where the bottleneck is: you need to human in the loop to point the direction.
   
   In the latest Naval podcast episode, “Motorcycle for the Mind,” Naval Ravikant talks about AI’s creative abilities. AI is good at connecting existing things, but Naval says that real creativity is something else. It is arriving at an answer that is not predictable or foreseeable from the known elements. Something so far out of bounds that you could search forever and not find it by accident. Naval points out that this kind of creativity is rare even among humans, which is why the human in the loop becomes more important. I can definitely feel that with game ideation.

2. The other bottleneck for this early wave of “AI founders” will definitely be the token cost. A 24/7 autonomous founder cluster running experiments nonstop won’t be cheap. But that feels like a temporary constraint, not a structural one, since the race for AI dominance isn’t only giving us better LLMs but the affordability has been coming in “all-you-can-eat” packages like Claude Code for $180 a month, which I use and have not ran into the usage limits for months now, even though I constantly have 3-4 Claude Code sessions running simultaneously for 8-10 hours a day.
   
   Memory is another similar complaint, the model forgets what it was supposed to do. This can be a bottleneck like price, but at the pace we are moving, all the memory issues will be solved soon.

What I am focused on now is not the OpenClaw cluster itself or getting the infrastructure to work. That part is increasingly solved. What I am actually building is the system, and that’s not something that needs OpenClaw. I can prototype and experiment on the system inside Claude Code, locally on my machine. The goal is an AI founder that can execute large tasks end to end. Creating documents, writing, coding, shipping, doing stuff online like filling forms.

The execution layer is no longer the bottleneck. The real human edge, for a long time still, will be in creative direction. But the leverage to build AI co-founders is already here.