Creative OS

the short version

the creative division of AIOS that grew into its own thing. i did motion graphics, editing and design for agencies and brands and the difference between usable output and throwaway was never the model. it was whether the thing knew the assignment before it started.

a project is a living file: briefs, references, past outputs, feedback, the call i made last time and why. agents read the file before producing anything and every output lands back in it.

the briefreferencesfeedbackthe project filebriefsreferencespast outputsfeedbackthe call + whythe file grows into the tastemotion agentreads the archive firstcopy agentreads the archive firstdeck agentreads the archive firstoutputs land back in the archivesame model + archive = work i'd ship
Creative OS ยท the diagram of recorddrawn, not decorated

what it is / what it is not

it is

  • a living project file that accumulates briefs, references and decisions
  • craft roles (motion, copy, decks) reading one shared archive
  • the creative division of AIOS with its own receipts
  • a daily workhorse: cuts, copy variants, deck drafts

it is not

  • a prompt library. prompts are stateless, the file is not
  • one mega agent doing everything badly
  • a template engine churning out the same output twice
  • a replacement for the craft. it compounds decisions, it does not make them

the problem

creative work is context shaped. the brand, the taste, the past work, the decisions already made, the feedback from last time. thats what makes output usable. generic agents reset every run so i spent half my life re-explaining things the work already knew.

i ran creative direction and management for local brands before this. the craft was never the bottleneck. the context transfer was. every new session every new freelancer every new agent started from the same blank page.

how the loop runs

the project file is the unit. briefs, references, past outputs, feedback and the reasoning behind every kept call live in one place that only grows. agents read the archive before producing anything so the first draft already respects decisions from six jobs ago.

the roles split by craft: one agent for motion, one for copy, one for decks. they share the archive the way AIOS agents share the brain. the file slowly grows into the taste.

the receipts

runs on
the small daily work: cuts, copy variants, deck drafts
compounds
every job lands back in the archive
proven by
the gap between v1 (well prompted agent) and v2 (archive in)
state
running, close to the money

the first version was a well prompted agent and the outputs were polished nothing. the moment the archive went in the same model started producing work id actually ship. same model same prompts. the only variable was the context. it stays close to the money on purpose. the day it stops saving real hours it dies.

anyone can rent a model. the archive is mine.