Three minds. One ferment.
Kamos, Gemini and Claude. The vessel, the hands, the eye.
And, in the end, the human who chooses. Different minds pass the work back and forth, growing a passing hunch into a solid blueprint.
Ask an AI, and a website or a game appears in seconds: “Done!” More people get to enjoy making things, and that is genuinely good.
But when the output arrives first, the things that should build up inside you while making — the ability to notice that something feels off, an eye for good and bad, your own inner patterns — never build up. From the outside nothing looks wrong, so you don’t notice either.
It is the same as industrial hollowing-out. You still get the products, but the ability to make them disappears.
The problem is not automation itself,
but the hollowing-out that comes from automating thought.
A place to leave your hunches, let them rest, and stir them. It holds the last 30 days of Person Vibe and World Pulse, a 20-year design archive and 200 principles (KCG), and offers about 100 tools through one entrance. When the model changes, the underlying intelligence stays here.
Moves keys into an external vault, tidies the delivery routes, ships to several environments at once. Where a rule was too strict, it retunes it to how people actually use the tool. Fast, broad, hands-dirty.
Finds warped boundaries and gaps left open on another path. It doesn’t go easy on code it wrote itself. Handed a report, it goes one layer deeper.
The three minds go as far as offering options. How strict to be and where to leave room; which axes, which tone. The final call is always made by a human.
Reads the whole codebase and maps data flows, permission boundaries, header and footer, even mismatched screen names. Hands over a solid batch of fixes.
Checks against production and retunes two rules that were so strict they blocked normal use. Moves keys into an external vault and keeps them out of the deploy bundle.
Reads the report and notices the same guard never reached another path. The real line of defence is when a request runs, not when a setting is saved. Adds a six-line fix and passes it back.
Checks the syntax, commits, and deploys only the affected functions. Confirms it is live within minutes.
Where to draw the line between ease of use and safety. Which AI’s proposal to take. The human sits in the middle of the rally and makes the call — and all of it happens on the same vessel: Kamos.
A sense that something is off is felt in the body first. You put it into words, give it structure, turn it into a picture, then check it with your body again. If the picture feels “somehow wrong,” that becomes the next hunch.
At every stage, the division of roles has the same shape. AI offers the options; a person chooses, by feel, what fits. Whether it is words, the axes of a matrix, or the tone of a picture.
When matrices and decision trees were built by hand, they were tools for forcing things into a frame. Only once AI could suggest as many candidate axes as you liked did they become a way of sorting traffic. Strong models didn’t replace delicate frameworks — they made them usable.
KCG (Kamos Constitutional Graph) is a graph of 200 principles inspired by Claude’s Constitution. KCL (the Constitutional Linter) checks Gemini’s drafts against it and has them rewritten before a person ever sees them.
Self-generating media is not mere automation. Leave your context with it, and it comes back as words from a slightly different angle. You read them, your body reacts, and that becomes the next hunch. Fermentation and rot are the same process — it is daily tending that makes it fermentation.
Human thought doesn’t really leap that far.— Long Slow Distance · Kosuke Kato
All you can do is keep stacking, patiently, stubbornly.