The Elephant

I don't know how to code.
I am 100% AI native.
It works.

359K
AI messages
132
repositories
581
deep sessions (100+ msgs)
$200K+
revenue generated
github.com/mordechaipotash →

The Arc

Nov 2022Couldn't write "Hello World." Used low-code/no-code.
Dec 2022AI taught me to use VSCode and run a script in terminal.
2023Hundreds of scripts, broad exploration.
2024Production systems. $200K revenue generated.
2025Brain MCP. 106K embeddings. 82-tool server.
Jan 2026Querying my own thinking at 256ms.

35 months. Zero to cognitive prosthetic.

The Work

brain-mcp
82 tools. 106K embeddings. Query 3 years of thinking at 256ms.
Python, DuckDB, LanceDB, MCP Protocol
wotc-app
$200K in tax credits. 72x efficiency gain. Built in 1 week.
Next.js, FastAPI, Supabase, PostGIS
youtube-pipeline
32K videos. 41.8M words. Zero API costs at scale.
Python, Whisper, LanceDB, yt-dlp
qinbot
Android AI agent on a 3" phone.
Kotlin, Termux, Claude API

How I Build

Context engineering > vibe coding

The 2026 paradigm. Specification matters more than syntax. AI writes the code. Humans specify what to build and verify it works.

First principles > frameworks

No copy-paste tutorials. Think from scratch. The same wiring that blocks rote thinking forces first principles.

Hyperfocus sessions

581 sessions over 100 messages. Deep dives, not chat. "When I very deeply get into something, I almost like hack it, crack it, understand it, and go very, very far in building it out."

Ship at 80%

Rather than die at 0%. Perfectionism kills. Done beats perfect.

Communication Pattern (from 359K messages)

"give me" (5,735×) — Direct. Want output, not process.
"no no" (853×) — Fast correction. Fix immediately.
"step by step" (686×) — Sequential depth, not parallel breadth.

5.63% correction rate = 1 in 18 messages redirects the AI. This is fluent AI. Not prompting—thinking.

Platform Agnostic

129+ technologies across 36 months. Not locked to any stack. Pick the right tool, learn it fast, ship, move on.

DatabasesFirebase → MySQL → PostgreSQL → Supabase → DuckDB
AIOpenAI → Anthropic → Gemini → OpenRouter → Ollama
FrontendWordPress → React → Next.js → Streamlit → plain HTML
DeployHeroku → Vercel → Railway → Docker
"I am skilled to the point that I built a full stack application the other day in two and a half hours, something that ChatGPT said should take four to six hours for a team of four... should cost somewhere between $30,000 to $120,000 to do."
— 2024-03, Beijing Borja session

The 2026 Thesis

AI commoditizes single-platform expertise. Anyone can ask Claude to write React code.

The skill is knowing WHEN to use React vs Streamlit vs plain HTML.
The skill is judgment. The skill is specification.

I've been practicing this for 35 months.
359,167 messages of practice.

Verify it yourself →
← back to facets
SYSTEM ONLINE
> Optimizing query path... _
T-