The editor who isn't a person
On the How I AI podcast this month, Claire Vo's guest explained how her videos get made. Brooke Lacey has been making videos since before Facebook had video; she built a TikTok audience of more than 290,000 by learning to cut short-form herself. Today she runs a weekly short-form operation with no editor. The chain is Claude Code orchestrating FFmpeg, Whisper, and Python on her own machine: raw phone footage goes in, captioned, cut shorts come out in about 15 seconds.
Fifteen seconds is the part everyone talks about. It's the wrong part. Before the machine could do that, she had to teach it her jump-cut cadence, her hooks, and what her idea of funny sounds like. The edit speed is the result. The teaching session is the system.
The skill that got delegated isn't editing
Translated for a business that isn't a creator brand: the mechanical half of content production — transcribe, find the sentences, cut the dead air, caption — is now a task you hand to software. The part that stays yours is the part that was always yours: taste. What your hook sounds like, when you speed up, when you hold a beat, what your customers find funny.
Most owners get this backwards. They assume the AI's job is to make the content good. It isn't. Its job is to make the content fast; good is set by what you teach it. The actual work in this setup isn't learning editing software — it's articulating your style to a machine, once, in writing, in a way it can execute every week.
The three moves
1. Record in one take, delegate the cut
One take, no re-record. The raw clip is the input; the software's job is the mechanical pass — transcription, the cuts, the captions. If the tool you already pay for has a "long video into clips" button, start there. The pattern doesn't require a particular stack; it requires that recording is one step and cutting is a different step that isn't yours.
2. Teach by exemplar, not adjectives
"Make it punchy" teaches nothing. What works, based on how Lacey describes her setup:
- Show finished pieces you already like — yours, not someone else's. Lacey taught the machine with her own published videos, not with descriptions of them.
- Point at moments instead of naming qualities. Her real instructions were things like "notice the cadence and the little laugh that makes the hook work" — the pause before the punchline, the cut that lands right after the word ends, not a beat later. "Make it punchy" teaches nothing; a named moment does.
- Judge the first outputs by pointing at what's wrong, not by quietly rewriting them. Each correction becomes a rule; keep every rule in one file.
That file is the asset. Tools churn every quarter; your written style rules move to the next tool in five minutes.
3. Mine the comments for next week's brief
Lacey's editorial calendar is her comment section: the questions, the objections, the "you should cover this" replies become next week's script. It's the cheapest research loop in marketing and it works identically for a business — the questions customers ask after a video are the topics for the next video. The audience writes the brief. Your job is to pick one and answer it in a single take.
Honest boundaries
- The named stack is a programmer's stack. Claude Code driving FFmpeg and Python isn't a small-business requirement — it's proof of pattern. The auto-captioning editors sold to business owners run the same pipeline with a monthly fee. Buy boring; the style file travels.
- Once a week beat five a week. She grew the audience posting weekly, because one take plus a delegated edit is sustainable. Delegation doesn't obligate you to flood the feed; it makes the one-post commitment survivable.
- The machine copies your style; it doesn't create one. If the raw take has no hook and no point, a faster edit publishes the same nothing sooner.
Tonight's version
One hour: record one take answering the question a customer asked you this week. Hand it to whatever editor you have — AI or otherwise — and while the first cut plays, write down five rules you'd enforce if a new hire had made it. Then fix the next cut using those rules. You now have the start of the file. From here on, recording is the only part that needs you.
Want practical AI automation templates for small businesses? Check out the Boring Automation Pack — no hype, just tools that work.
Top comments (0)