Isaac Wostrel-Rubin
I spent three and a half years talking to AI ten to twenty hours a day, trying to get it to do what I actually meant.What I found wasn't a prompting trick. It was that almost everything we say — to each other, to machines — is a compressed transformation we hand over as if it were simple. You can always open one up and look at what's inside. What survives being opened and closed again is the part that was real.Then I hit the problem that cost me four years. When you build this way, a true idea and a beautiful wrong one feel identical from the inside. So I built a machine to tell them apart.It worked, which was worse. It killed most of what I loved. Then it told me that no machine can finish the job — that somewhere this has to touch someone who isn't me.That's the work: how patterns become real enough to act, and how to build the ones that make it easier to be a person without costing anyone else theirs.
What do you do?
I teach you how to think about the Claude Code ecosystem so your AI systems actually compound.
Why should I listen?
I learned to code entirely through AI. No CS degree. No bootcamp. That constraint meant I couldn't work around the limitations—I had to solve them at the root.Everything I made is battle-tested with AI. Not "here's a framework I thought up"—I test every piece: Can the AI talk about X better with this context? Can it do the workflow? Where does it get tripped up? If I break this skill into multiple steps, does it perform better?Everything does exactly what I say because I tested it until it did.
I didn't write down what I think would help you and try to sell it. I built it, ran it, fixed it, ran it again—until it worked.Two years later, I came back with a working artifact: an agent that makes its own tools, skills, and other components and remembers what it built, and extends itself.
What problems do you solve?
When an AI doesn't understand something, it sometimes tells you and sometimes pretends to. Especially dangerous is when it has just enough context to talk about it right—but then you go to actually do something and it's immediately wrong.Why does that happen? The concept is too complex for the LLM to understand all at once AND do something with it.You need to progressively disclose information. And you need to do this multiple times across a workflow.Without a scaffold for this, you're stuck driving every step. Re-explaining. Re-instructing. Re-contextualizing. Every session starts over. Nothing compounds.Most AI education gives you frameworks. "Here's how to make an MCP." "Here's how to write a skill."Then you try it—blank screen. What do you even write?The implicit contract: figure out everything the course didn't say.
How do you solve my problems?
- Step 1: Enable composability- Step 2: Enable AI understanding of it- Step 3: Talk to AI and get good at your own systemMeta-architecture.The agent suggests ways of thinking about it, you agree on one, it gets built, tested, installed, and deployed to your own agent. Automatically.
The artifact makes MCPs. Makes skills. Makes workflows. Remembers what it made. Knows how to extend itself. Suggests what's next.You don't walk my path. You use the artifact. The artifact is the compressed path.My system is an observation-to-agent pipeline. You see potential. You test it with AI. You iterate until it works. Now it's a product.Your knowledge forms it to your niche. Your work scales it into a funnel and beyond.
Why do you do what you do?
When people finish journeys, they come back with boons.
Most offers don't have boons. They have frameworks and blank screens and implicit contracts. I'm doing the opposite—giving you an artifact you can study, use, and replicate.
Don't trust offers without boons. Trust offers you can study.I provide the journey map that reliably causes you to come back with boons. In this case, those are awesome AI systems.
Should we work together?
I'm not selling a system that does your specific thing. I'm teaching you how to systematize your thing into AI. And then I'll build that with or for you.