I manage heavy civil projects in Hawaii. No CS degree, no dev background. November 2025 I started building software with Claude as my developer.
Operator is the latest one. It's an AI course taught entirely through conversation — no videos, no slides. You talk to Gojo, an AI teacher that asks questions back, corrects your thinking, and won't move on until you actually get it.
The meta part: an AI course, built with AI, by someone who learned to use AI by building with it. That's not a coincidence — it's the whole thesis. You learn by doing, not watching.
Lesson 1 is free for anyone who signs up. Six lessons total in the founding cohort.
Stack: Next.js, Claude API, Supabase, Resend, Stripe. I directed every product decision. Claude wrote the code.
Happy to talk about the build process, the pedagogy, or what it's like shipping software without being a developer.
Fair question — I don’t think raw AI is “reliable enough” if you treat it as a source of truth.
The goal here is the opposite: train people to work with it so outputs get pressure-tested instead of accepted blindly.
The system is structured as a back-and-forth — it asks for context, challenges inputs, and forces iteration — so reliability comes from the interaction, not the model alone.
In practice, the gap isn’t just model quality — it’s that most people don’t know how to make the outputs reliable.
34 comments
[ 3.1 ms ] story [ 82.8 ms ] threadOperator is the latest one. It's an AI course taught entirely through conversation — no videos, no slides. You talk to Gojo, an AI teacher that asks questions back, corrects your thinking, and won't move on until you actually get it.
The meta part: an AI course, built with AI, by someone who learned to use AI by building with it. That's not a coincidence — it's the whole thesis. You learn by doing, not watching.
Lesson 1 is free for anyone who signs up. Six lessons total in the founding cohort.
Stack: Next.js, Claude API, Supabase, Resend, Stripe. I directed every product decision. Claude wrote the code.
Happy to talk about the build process, the pedagogy, or what it's like shipping software without being a developer.
The goal here is the opposite: train people to work with it so outputs get pressure-tested instead of accepted blindly.
The system is structured as a back-and-forth — it asks for context, challenges inputs, and forces iteration — so reliability comes from the interaction, not the model alone.
In practice, the gap isn’t just model quality — it’s that most people don’t know how to make the outputs reliable.