Where AI Belongs in a Studio Workflow

The useful question is not whether to use AI. It is which specific tasks in your week should never have required a human in the first place.

Here is how I divide the work.

Between a shoot and a delivery there is a stack of mechanical work — transcription, captions, reformatting for four aspect ratios, a first cut-down, file naming, folder structure. None of it requires taste. All of it consumes evenings.

The same applies between an enquiry and a quote: classifying what came in, pulling the relevant past project, drafting the first version of a reply. The draft is checked; the gathering is not repeated by hand.

Outlines, alternative headlines, a list of angles you have not considered, a first attempt at a structure. The value is in getting past zero, not in what is produced — most of it gets discarded, and that is fine, because discarding is faster than inventing.

This is assistance, not output. If the first draft ships, something has gone wrong.

Testimonials. Case study outcomes. Anything describing what a client experienced. Anything involving a child, a beneficiary or somebody's personal story.

This is not a technical limitation — a model would happily produce all of it. It is that a claim about a real person has to have been verified by a person, and a story that belongs to someone else should not be paraphrased by a machine for convenience.

Ask what happens if the output is subtly wrong and nobody notices. If the answer is 'a file is named oddly', automate it. If the answer is 'we published something untrue about a client', it needs a person, permanently.

That single question sorts most of a workflow in about ten minutes.

The mistake is using AI manually, forever — opening a window, pasting, copying back. That is the assistant phase, and it caps out quickly.

The return comes from wiring the task into the workflow so it happens without being asked. It takes an afternoon to build and it pays every week afterwards, which is the entire argument.