Brand file 2.0: getting nine agents to stop contradicting each other

The problem with the first version was blunt: the competitive analyst called you a "specialty coffee brand", the video director wrote a voiceover about "good coffee at a fair price", and paid ads targeted "price-sensitive commuters".

Three agents each read the website once and each reached a conclusion. None of them was wrong. Together they were three different brands.

The file is the single source of truth

2.0 turns this around: crawl the site once, produce one brand file — business, selling points, tone, category, competitors. From then on every agent reads that file instead of going back to the website.

The story is consistent now, but that creates a new problem: the file changes.

So the file carries a version

When the brand strategist produces a new positioning line and you confirm it, the file moves up a version. Every agent run records which version it consumed.

That gives "positioning changed — do the old content scripts still hold?" a mechanical answer: compare versions, and anything produced below the current one is flagged as stale. You can rerun it or keep it — but at least you know it is old.

What belongs in the file, and what does not

The test is simple: it goes in only if several agents need it.

"Tone of voice" goes in, because writing scripts, briefs and support replies all use it. "Is this video 9:16 or 16:9" does not — that is one parameter belonging to one agent.

A bloated file costs you on every agent call, in both spend and surface area for mistakes.