Give Claude the folder, not the instruction

The phone, the coffee mug, and three folders

Picture a small business owner who just filmed an ad on a phone propped against a coffee mug on the kitchen counter. One take, one script, no crew, no editor booked for next week. The kind of raw footage that usually sits in a folder for months, because hiring someone to cut it feels like more effort than the ad is worth.

Instead of writing a paragraph of instructions, they hand Claude Code three folders. The raw footage. A handful of the brand's own ads that had already performed well. The brand's logo files, fonts, and color specs. That's the entire brief.

What comes back is a full edit: motion graphics timed to each line of the script, sound design, background music, and several different opening hooks built to test against each other. Nobody described what a good ad looks like. The old ads already showed it.

Why the instruction almost disappeared

Here's the part worth sitting with. The instruction, if you could even call it that, was short, something closer to "cut this the way our best ads are cut" than a shot-by-shot brief. The heavy lifting came from the materials, not the wording.

This is the same shift I keep making people notice in training, just wearing a different department's clothes. A lawyer who hands Claude a sample NDA gets a sample-shaped answer. A lawyer who hands Claude the client's actual contracts folder gets redlines that sound like the firm's own voice, because that voice was sitting right there to copy. A controller who hands over a demo spreadsheet gets a demo-shaped variance note. Hand over the real GL, and the note reads like something the controller would have written at eleven at night before the board meeting.

Video editing is just the loudest version of the same mechanism. The folder is the instruction. The words on top of it are almost a formality.

Where I'd slow down

I want to be honest about what this story can't tell you. The most quotable part of any workflow like this is usually a number nobody can check, how many hours a human editor would supposedly have needed for the same cut. I don't know that number. Nobody reposting a clip like this knows it either, so it's not going in this post.

What is checkable is narrower, and more useful anyway: the workflow itself is real and repeatable. Feed a coding agent your raw footage plus a small set of working examples, and it will assemble a first pass. Whether that first pass is good enough to publish is a judgment call, and it belongs to someone who knows the brand, not to the tool that produced the options.

That detail is the whole job description for whoever runs this kind of workflow. The agent can generate five opening hooks in the time it takes to make coffee. Someone still has to know which one sounds like the brand and which one just sounds confident.

What to try tomorrow, even without a video

You don't need a video use case to take something from this. Before your next long prompt, stop and ask a narrower question: is there a folder you could hand over instead of a description?

A weekly newsletter written from three past issues that actually landed, instead of a paragraph explaining what the brand's tone is supposed to sound like. A sales follow-up drafted from the five emails that turned into meetings, instead of a list of dos and don'ts. A slide deck built from the client's actual data room, instead of a template with placeholder numbers.

The teams who get the most out of Claude in my sessions are rarely the ones with the cleverest prompt. They're the ones who stopped typing a description into a blank box and started handing over the real document, examples that actually worked, and constraints they didn't have to guess at. The model fills in the middle. The part that decides what ships still belongs to you.