AI is not short on copy. It is short on someone to decide.

In year one we assumed the bottleneck was output: copy we could not finish writing, videos we could not finish cutting, calendars we could not finish filling.

A year in, output stopped being the bottleneck. Three other things were doing the blocking.

Assumption one: more versions is better

Wrong. Give someone three versions and they read them carefully. Give them thirty and they close the tab.

Choosing is itself a cost. The video director now offers four scripts and no more.

Assumption two: more automation is better

Also wrong. Nobody dares use fully automated output — not because the quality is bad, but because nobody owns it.

So the flow keeps at least one point where a person decides, and it sits at the step with the highest density of judgment. For video that is the script; for paid ads the budget; for publishing the channels and accounts. Everything else runs itself.

Assumption three: the AI's output is the finish line

The most wrong of the three. The plan is not the finish line. The meeting is.

A plan earns its keep by making the meeting argue about "should we go after the commute use case" instead of "let us polish this sentence". Which is why we spent so much effort on export: the schedule has to go straight to the agency, the budget sheet straight into finance, the deck straight onto the screen.

Output that does not enter someone else's workflow is just nice-looking text.