This week the calendar says one thing: one machine, eight pillars. Today is the first pillar, branding, and we are going to build a brand the way we build everything here, with rules a machine can follow. A brand is not a feeling. It is a small set of decisions, written down, that you make once and then never argue about again. By the end you will know how to write those decisions so a model can keep them for you.

Step one. Open a plain text file and call it the brand. Not a slide, not a folder of images, a file. Inside it you put the few things that never change: who you serve, what you say no to, and the three words you want stuck in their head. Everything else, every post and ad, gets generated from this file. When the source is one file, the agent always knows where the truth lives.

Step two. Write your colors as named tokens, not random hex codes scattered around. Red is one value. Black is one value. White is one value. You give each a name and a job. Then in your terminal you feed that list to Claude Code and tell it to use only those three, ever. A model will never accidentally pick a fourth color if a fourth color does not exist in the file.

Step three. Pick one typeface and put its name in the file. We use Tahoma, because it is plain and a child can read it. The point is not the font, the point is that it is decided. When the rule is written, the agent sets every headline in it without asking. You stop choosing fonts and start shipping.

Step four, the one most people skip. Make a list of words your brand will never say. Ours bans the empty hype words everyone reaches for, the ones that sound big and say nothing. You hand that banned list to the model as a rule. Now every line it writes gets checked against the list before it ever reaches you. The brand polices its own voice.

Step five. Describe your voice as if it were one person at a table. We tell the model ours is brash but calm, dry, a little dark, and it explains instead of selling. That short description sits in the file too. When you ask for a caption, the agent writes in that voice, not in the flat tone every other brand shares.

Step six. Decide how the mark behaves and never bend it. Our rule is simple: the logo stands alone, and we never type the name across an image. We wrote that rule as a check the machine runs, so any asset that breaks it gets flagged before it goes out. One sentence in the file protects the mark on every single post.

Step seven, where it pays off. Point the agent at that one file and ask it to build a week of posts. It reads the colors, the font, the banned words, the voice, the logo rule, and produces work that already obeys all of them. You are no longer briefing a brand every time. You wrote it down once, and the machine applies it forever.

Here is our honest proof. We did not write a brand bible nobody opens. We wrote one source file and pointed the studio at it, and the whole public calendar you are watching came out of it. Same three colors, same plain font, same banned words, every day. You are looking at the output of the exact file we just described.

So that is branding the studio way. Not a mood, a file, with rules a model can hold. Write your source file, name your tokens, ban your words, set your logo rule, then let the agent generate. Your next step is small: open one text file today and write down three colors and one font. You can watch ours run, live and in public, at office dot temerarii dot xyz.