ExpertsFM“Claude cuts my TikTok”: How this creator is scaling video with AI | Brooke Lacey
How I AI · October 19, 2026
5 Selected Lessons:
heuristic · 2 framework · 2 case · 1
Edit-as-text-file for multi-format repurposing
Source
How I AI — “Claude cuts my TikTok”: How this creator is scaling video with AI | Brooke Lacey (October 19, 2026) · at 10:27 ▸ · at 13:41 ▸
When this applies
when you need to produce multiple output formats/variants from one underlying piece of work without redoing the core creative decisions
The takeaway
Separate the decision/logic layer of a creative or operational process from its output-rendering layer by storing decisions in an inspectable, machine-readable format.
Why it works
Representing the edit decisions (cut points, shot sizes, labels) as a structured text file decouples the creative decision layer from the rendering layer, so the same decisions can be replayed against different render parameters (aspect ratio, duration) cheaply.
One level up
Turning tacit or ephemeral decisions into persistent, structured artifacts (a file, a spec) is what makes a process both reproducible and reusable across contexts.
Where it stops holding
This stops working when the output format change requires fundamentally different content decisions (e.g., a 60-second cut needs new material, not just re-timing of the same cuts).
Detail
procedure:
- Store every cut as a row: start time, end time, shot size, label
- Treat this file as the single source of truth for 'the edit'
- Re-run the same command/file against different render targets (duration, aspect ratio) to produce variant outputs
rationale: 'Because the whole edit is a text file, the same command makes the thirty and sixty and the landscape versions from the same take.'
In their words
And because the whole edit is a text file, the same command makes the thirty and sixty and the landscape versions from the same take.
Personal 'operating system' with memory lanes and status files for agent correction
Source
How I AI — “Claude cuts my TikTok”: How this creator is scaling video with AI | Brooke Lacey (October 19, 2026) · at 30:59 ▸
When this applies
when managing a complex agentic system across multiple projects/products and needing a reliable way to correct and propagate fixes
The takeaway
Build a layered memory architecture (local session state + global shared state) so that corrections made in one instance automatically inform the whole system going forward.
Why it works
By maintaining persistent status/memory files per project plus a shared global 'spine,' errors can be diagnosed locally (why didn't you read that instruction) and then the fix is pushed back up to the global memory so it propagates across all related sessions/products, preventing the same mistake from recurring elsewhere.
One level up
Durable automation systems need explicit, inspectable memory structures, not just one-off prompts, so that learning persists and propagates rather than resetting each session.
Where it stops holding
This stops working when the system lacks the discipline to maintain and reference status files consistently, or when the business is simple enough that a shared global memory layer is unnecessary overhead.
Detail
procedure:
- Maintain status MD files / project-level memory that sessions should reference
- Maintain a global 'OS lane' memory shared across products
- When the agent errs, prompt it: 'go back to the way that we started, understand those instructions, and then do it again'
- Ask it to diagnose why it didn't read the correct instruction (e.g., a session hook issue)
- Have it report the fix back to the main operating system 'spine' so the correction is applied globally
rationale: 'So the way that I prompt when it screws up is like, go back to the way that we started, understand those instructions, and then do it again... go back and tell the main operating system spine that you messed up so that it can make the change globally.'
In their words
go back and tell the main operating system spine that you messed up so that it can make the change globally
Encode your own tacit expertise before delegating to an agent
Source
How I AI — “Claude cuts my TikTok”: How this creator is scaling video with AI | Brooke Lacey (October 19, 2026) · at 2:30 ▸ · at 17:16 ▸ · at 22:52 ▸
When this applies
when trying to get an AI to replicate a skill you have but cannot fully articulate — writing, editing, sales pitches, design taste
The takeaway
You can only successfully delegate a skill to an AI to the degree you can first make that skill explicit to yourself.
Why it works
An agent can only reproduce a style if the person doing the prompting already understands, at an explicit level, what makes their own output good (what the hook sounds like, where cuts land, what cadence means) — tacit skill has to become explicit instruction before it can be delegated.
One level up
Automation amplifies existing expertise; it does not substitute for the absence of it.
Where it stops holding
This stops working when the skill is so embodied (facial expressions, live chemistry between two people) that it resists being reduced to explicit rules at all, even by the expert.
Detail
rule: Before asking an AI to replicate a personal creative style, first identify and name the specific mechanics of that style (cadence, hook placement, pause length, cut timing).
threshold: If you cannot explain why something is funny/effective, don't expect an editor — human or AI — to nail it either.
rationale: She says she 'was able to orchestrate Claude the way that I would actually do it' only because she already understood video editing basics from years of practice, unlike hiring editors who 'never worked the way I needed it to.'
In their words
So I was able to orchestrate Claude the way that I would actually do it, if that makes sense.
A strong back-catalog reduces required posting frequency
Source
How I AI — “Claude cuts my TikTok”: How this creator is scaling video with AI | Brooke Lacey (October 19, 2026) · at 25:18 ▸
When this applies
when deciding how much new output cadence is needed to sustain an audience/customer base that has accumulated years of prior work
The takeaway
Years of compounding quality work can substitute for high-frequency new output — distribution systems (algorithms, word of mouth, reuse) continue to surface past work.
Why it works
Once a large engaged audience exists and platforms/algorithms continue recirculating older high-performing content, the marginal value of constant new output drops, because distribution is partly sustained by the existing library rather than solely by fresh posts.
One level up
Sustainable systems let earlier investment keep paying out, reducing the need for constant fresh input.
Where it stops holding
This stops working when the audience/platform rewards recency heavily, when the back-catalog is thin, or when the creator hasn't yet built sufficient scale/engagement for recirculation to matter.
Detail
rule: After building a substantial, engaged following over years, posting frequency can drop significantly (she posts about once a week) while maintaining reach, because past content keeps getting 're-gurgitated.'
threshold: roughly once per week at 200,000+ followers, per her account
rationale: 'I don't have to post that often... I think once you have done the work over a number of years and you do get to the hundreds of thousands of followers, your stuff just kind of keeps getting regurgitated.'
In their words
I think once you have done the work over a number of years and you do get to the hundreds of thousands of followers, your stuff just kind of keeps getting regurgitated.
Human editors couldn't match personal comedic timing
Source
How I AI — “Claude cuts my TikTok”: How this creator is scaling video with AI | Brooke Lacey (October 19, 2026) · at 2:30 ▸
When this applies
when outsourcing a nuanced creative/judgment task to another human produces consistently unsatisfying results despite instruction
The takeaway
When a task depends on deeply personal, millisecond-level judgment, delegation to another human without a way to transfer that judgment precisely will underperform — look for tools that can capture the judgment mechanically instead.
Why it works
The timing required (cutting exactly after a word's sound ends, matching her real speaking cadence) was too precise and personal for someone else to intuit from instructions alone, so she kept doing it herself for years at high time cost.
Where it stops holding
This stops working when the creative task doesn't hinge on exact personal timing/voice and can tolerate generic execution.
Detail
situation: Brooke needed precise comedic jump cuts matching her natural speaking cadence; it took hours per video to edit herself.
decision: She tried teaching video editors how to replicate her exact cut timing and facial-expression-based cadence.
outcome: It 'never worked the way I needed it to,' so she kept doing all editing herself for years until finding Claude Code could do it instead.
lesson: Personal, millisecond-precise creative judgment is hard to transfer to another human but may be transferable to an AI agent given the right explicit signals (word timing, silence detection).
In their words
I tried teaching video editors like how to do it and how I wanted it, and it just never, it never worked the way I needed it to.