Exploration and the 21st Century

Although I fully embrace AI and will include AI writing in different sections of this newsletter, this article is written by human, and might include human mistakes. I invite you to revel in the imperfektion of my human writing.

Did you ever wish you were a settler in the old West forging your way to a new land? Or a mapmaker, documenting the roads connecting different settlements? Or maybe a great sea explorer searching for new land or resources or a great white whale? But alas, it's been done. We live in a world where every nook and cranny has been named, cataloged, discovered, and described. Stories made up and disproven by science. And so, maybe it all feels done. BUT it is not.

If you are looking for adventure, exploration, experimentation, and discovery. We are living through a renaissance of AI. A generation from now, it will all be figured out, which means in our time it will be the time to experiment, explore, and discover is right now.

I'm writing this newsletter to document my exploration at the forefront of AI, which today means, Agentic AI, but could mean something else tomorrow.

Most people know AI as LLMs and use it as a way to more or less replace a Google search, maybe you have it write an itinerary, or a silly photo, or proofread an email. The way I'm using AI is to create and write code… go ahead and say it, ViBe CoDinG, it is. But don't let the attention-grabbing posts and headlines deceive you. It's harder than you might think.

You can send a prompt and build something, but when you look at modern software, there are many things that you need to prompt to build a tool that actually replicates that functionality and form or creates a new one. I'm not saying the Agentic AI isn't impressive. It's absolutely the most meaningful tool in our latest tool arsenal for building and creating and if you follow along I think you will agree.

WHAT AI HEADLINES DON'T TALK ABOUT

1.	You still need a couple of active brain cells. It's not that easy. At least right now, the best models (frontier models) are like a really helpful assistant that doesn't take coffee breaks or need rest but also sometimes sort of just does silly things. Misinterprets a request or makes things up, or gets lost in its own spiral of problem-solving.
2.	You need ideas. If you're building, you have to know what you want. A lot of times, you hear "AI is amazing and it can do anything," but then you ask people what they use it for and get a lot of vague answers. This is because YOU NEED IDEAS. AI needs an operator, and without it, it's like the world's most powerful calculator. Sitting in wait for the operator to present it with an important equation.
3.	Prompting is trial and error. Do you need a class? Maybe, but "prompting" is just assigning a task. Pretend you are talking to your newest employee; that's the right level of explaining you need to use. And, some models are still better than others at interpreting your prompt into action. If you were able to send the same prompt to the last year's best models and compare it to now, the new models will do much better. But the prompt stayed the same, so what really is happening is the model is getting better at deciphering what the heck we are asking it to do. And even now, different models will do better than others.
4.	Anyone can do it. If you can check the boxes for the first 3 tips then you can do this. You can experiment with this technology and now is the time. For me, the first step in experimentation right now is setting up Openclaw. Stick with me and I'm going to tell you what that is and why you want to do it in just a bit.

A SOFTWARE REVOLUTION IN NUMBERS

With AI, if you blink, you miss something big. It's probably the most rapid news cycle that I have ever followed for changes and new developments. To put numbers to what is happening right now, GitHub, a repository for open-source software, broke a new record in 2025 when it hit "1 billion new software commits". That absolutely mind-blowing number was a short-lived record. Where we currently sit in April is with 275 million software commits per week — that puts us on track for 14 billion in 2026. From Sept 2025 to March 2026, the number of commits has increased 400%. So where are all those software commits coming from? You probably already know, it's being written by AI. Parabolic software growth isn't all slop; it's solving problems and improving systems.

EVERYONE NEEDS THEIR OWN AGENTIC AI

Right now, agentic AI means one thing to me: Openclaw. Openclaw, the open-source framework for agentic AI, was made famous under the name Clawd (Nov 2025 – Jan 2026) but received a legal request from Anthropic asking them to change their name due to trademark infringement (Clawd? Claude? Confusing). So now it's been renamed to Openclaw and everyone is happy.

So what is it? It takes your AI model from your provider of choice (Google, Meta, Anthropic, OpenAI) and plugs it into a persistent environment. This environment, along with Markdown files (text files), gives your agent context and direction on "who am I? What do I do? What have I done in the past? Learnings, etc." Then you plug in tools to the environment that allow the agent to do things you want it to do. For example, you can have tools that interact with social media posts, YouTube videos, or check and send emails. You can give it keys to the kingdom and let it edit files, settings, network configuration, and write code, or you can keep it contained to specific tasks in a contained environment. Right now, the big AI providers are not giving you the "everything agent"; they are slowly releasing agentic features and what they eventually build will probably be better — or maybe it wouldn't. But right now, if you want to taste the rainbow of what AI has to offer, Openclaw should be your first stop.

CHART YOUR OWN JOURNEY

This newsletter has an optional DIY journey below the fold. I hope to entertain you with what I'm doing and creating, but if you want to take the plunge and join me, I'm going to give you a roadmap for what you need to do to get started. If you aren't ready to take the plunge and taste the rainbow, that's fine too — and here's where you check out and we catch up next week.

FEEDBACK

Optional feedback… What questions do you have about AI? What are the buzzwords you're hearing about or topics that concern you? I'll try my best to address them all or write about them in an upcoming newsletter.

// until next week

◆  Below the fold  ◆

This area is so you can follow along and setup your own Agentic AI.

THE HARDWARE

The CPU requirements are very low, and the GPU is not necessary at all because the compute is all in the cloud. Go read around, and you will find people running Mac mini computers for $600+; you don't need that. Today, we just don't have good enough small enough models that run on a $600 computer to replace what you can tap into with a $10–$20/month subscription. If you have the money and want to buy a Mac mini, go for it, but realistically, that money is better spent with a subscription and just finding some old, out-of-use laptop or other computer in your house, ideally with 16GB of RAM but 8 will work too.

THE SOFTWARE

Install Ubuntu Desktop, it's free and has all the capabilities you need.

https://ubuntu.com/download/desktop

Then load up OpenCLAW in the terminal.

curl -fsSL https://openclaw.ai/install.sh | bash -s -- --beta

There are lots of good guides out there, so you do this part on your own. If you get stuck, you can reach out to me for help, but this is really the first step down the rabbit hole, and you need to try it alone.

COMMUNICATION CHANNEL

The computer you choose is the home for your Openclaw instance but you will be communicating with it via a chat app. The options for chat interface are basically all of them. But the easiest to setup is Telegram, so if you get stuck here on choosing, just go with Telegram and you will be ready to roll in 10 min or less.

CHOOSE YOUR AI MODEL

You need to plug in a communication protocol for your AI model to connect to Openclaw. So any model that has API access will do. Openrouter has a lot of different options if you want to try out all the different models and compare them, but my recommendation is using the Minimax Starter Plan. For $10 a month you have basically endless access to their frontier model called M2.7. Use my referral code for 10% off.

View comparisons information below. More info here.

I also pay for a $20 a month Anthropic Pro membership so I can use Claude Code. It's the gold standard for coding and when I'm not working with Openclaw I'm working with Claude Code and feeding things back and forth. You could also skip the Openclaw instance altogether and just use Claude Code. But making the switch to Openclaw really gets interesting.

My current setup $10 / month — Minimax subscription plugged into Openclaw $20 / month — Anthropic Pro membership to use Claude Code Old computer + $30 a month gets you started on your journey.

“Imagination is the only weapon in the war against reality.”
— The Cheshire Cat