manifesto
agentic engineering: you don't write the code 99% of the time. you orchestrate agents and oversee what they do.
if this is new to you: a coding agent is an AI program that turns your written request into code. this site is about building real products with it, in a disciplined, verifiable way, and about learning to do the same.
what agentic engineering is
Andrej Karpathy drew the line on February 4, 2026: agentic, because you're not typing the code, you're directing whoever does; engineering, because there's a craft to it that you can learn.
We don't build agents. We build anything with agents. You orchestrate, the agent does the typing.
why we prove it by building
A claim is weightless without proof. So we don't say it works. We show the logs.
This site is the first build. 4 days since day one, 16+ agents on the job, 1 human at the wheel, ~$1 spent so far. You don't take the numbers on faith. You see them.
how the flywheel turns
Four steps, one loop: build, show, sell, loop. We build products with agents; we publish the full process of every build as a recipe; we sell the products and the packs pulled from them; the attention and revenue fund the next build.
A build qualifies if it's small, a real pain, 1-2 weeks, and either monetizable or an attention magnet. The hub itself is the reference for every build that follows.
why everything is public
Every recipe carries the same six fields: spec, workflow, token cost, calendar time, what broke, and a link to the working product.
We don't estimate the cost. We print it. We don't dress up the mistake; we log it. If a claim has no proof, we don't make it.
what the zero-contact rule means
No calls, no live events, no promised support, no consulting. Everything is self-serve and asynchronous.
One human with agents only maintains a system that carries no live obligation. The recipes and the asset packs work around the clock instead of us, with zero contact.
read the builds, use the assets
We're not asking for belief. Read the recipes, check the numbers, reproduce them on your own project.
What holds up gets pulled into an asset pack: CLAUDE.md files, skills, hooks, verification harnesses. Each one has a recipe proving it works.