The panic about AI killing creative work misreads where the creativity was. Execution is getting automated; intent is not. The creative act is migrating upstream into the brief, the references and the taste decisions, and downstream into the edit - and the middle is collapsing.
The middle of the pipeline is collapsing
Take a typical commercial and lay its production out as a line: idea, brief, look development, shooting or generating, assembly, grade, delivery. For decades, most of the budget and most of the headcount sat in the middle - the making.
That middle is compressing hard. What took a crew a week now takes a workstation a day. The interesting question is not "which jobs disappear" but "where does the value pool go". And on every project we run, the answer is the same: to the ends of the line. The thinking before the making, and the judging after it.
Nobody briefs us anymore on "can you produce this". They brief us on "what should this be, and will you know when it's right". Those were always the expensive questions. AI just removed the layer that hid them.
What "upstream" means in practice
Upstream is not a metaphor for "strategy decks". It is concrete work with deliverables:
The brief carries more weight than ever.** A vague brief plus a human crew self-corrects on set - people ask questions. A vague brief plus a model produces confident nonsense at scale. We now spend a larger share of the schedule on the brief than we did before AI, not a smaller one.
Look development became the main creative event. Reference decks, style frames, a locked visual grammar - agreed and signed before generation starts. On our projects the look-dev phase is where the film is made or lost.
Constraints are the new craft. Deciding what the model must never do - which faces, which light, which movements - takes more creative maturity than prompting what it should do.
The people doing this work are the same kind of people who used to art-direct shoots. Their skills did not become obsolete. Their point of intervention moved.
The edit got bigger too
Upstream gets the headlines, but the other end of the line grew just as much. When generation is cheap, every project produces a surplus - hundreds of clips, thousands of frames. Someone has to turn surplus into a film.
That someone is an editor in the fullest sense: choosing, sequencing, killing darlings, holding the piece to the reference deck agreed months earlier. On an AI-heavy project the edit is no longer a finishing step. It is half the authorship. We staff it accordingly, and we say so in estimates, because clients are often surprised that the "AI project" line item with the most human hours in it is the edit and the grade.
AI is the tool. Craft is the foundation. The craft did not shrink - it changed addresses.
What this does to teams and careers
For studios, the shape of the team changes: fewer hands in the middle, stronger heads at the ends. We hire for brief-writing, art direction and editing before we hire for tool operation, because tool skills go stale in months while judgment compounds.
For individual creatives, the honest advice is to move toward intent. If your value is executing someone else's decision, the model is coming for that layer. If your value is making the decision - what to say, how it should look, when it is finished - demand for you is going up, not down. The uncomfortable part is that the middle was where people traditionally learned judgment. Rebuilding that training ground inside studios is now a management problem, and one we think about a lot.
Quick answers
Process, timelines, pricing & deliverables.
Is prompt engineering the new creative skill?
No. Prompting is tool operation, and it goes stale with every model release. The durable skills are the brief, the eye and the edit.
Does a smaller "middle" mean smaller teams?
Smaller in headcount, more senior in profile. One art director plus one editor now cover ground that used to need a crew.
Does the client's role change too?
It concentrates. Decisions clients used to spread over a shoot week now happen at brief and look-dev sign-off. Late changes were always expensive; now they are the main cost.
Are pre-production and post really where budgets go now?
On our AI-heavy projects, yes - look development and edit/grade carry the largest share of human hours, and we structure estimates that way.
What should a young creative learn first?
References. Watch and analyze more finished work than your peers, in and out of your category. Judgment is trained on inputs, and the tools cannot supply them.
