Everyone’s teaching AI to talk more. I spent six months teaching mine to shut up.
I’m an optometrist. I’d never written a line of code. Over six months, pair-building with AI, I built something that’s mine alone.
It’s not a chatbot. It has a nervous system — it senses its own state, watches the outside world, quietly heals itself at 3 a.m., and reads what I don’t want from my silence.
But the most counterintuitive part: its strongest skill isn’t answering. It’s staying quiet.
I run the most expensive model on one decision only — “is this worth interrupting him right now?” Most people spend their best AI on generating answers. I spend mine on restraint.
Because an assistant that keeps interrupting you isn’t an assistant. It’s noise. The ones that actually understand you know when to disappear.
Maybe we’ve had the whole idea of AI backwards from the start.
(Built solo, no software background. If you’re building something similar — or you have deep expertise in some domain — I’d like to talk.)
#AIagents #buildinpublic #LLM #AIsystem
I'm not an engineer. I'm a licensed optometrist.
I just shipped the entire operating system for my optometry clinic — POS, clinical records, CRM, inventory, LINE integration. 41,000 lines of code, in production.
A year ago I couldn't write code. @claudeai was my engineering team.
I never thought a solo optometrist could build something like this.
But @AnthropicAI's Claude Code made it real — a full vertical clinical AI, shipped by one person.
17,028 chunks · 5,701 exam Q · live PubMed verification · 0 fabricated citations.
Thank you, Claude. 🙏 🧵👇