Big Idea

The argument in one line.

Before you can build an AI editing workflow in Claude Code, you need enough hands-on editing experience to judge a good cut, because an agent can only automate a process you already understand well enough to correct.

Who This Is For

Read if. Skip if.

READ IF YOU ARE…

You edit your own videos and want to hand routine cutting work to an AI agent like Claude Code.
You've started building a Claude Code plus DaVinci Resolve editing workflow and want to see how a good prompt turns into a reusable skill.
You're comfortable with basic AI agent concepts (plan mode, models, skills) and want a real example of how one editor structures the handoff.

SKIP IF…

You've never edited a video by hand — the video's core warning is that you need that experience before automating it.
You want a copy-paste prompt template — this is a philosophy and process overview, not a script.

TL;DR

The full version, fast.

The core argument: AI can shortcut editing, but only for someone who already knows how to edit, because a workflow built without that reference point has nothing to judge good cuts against. His actual process, organize and ingest footage, listen through it, then a rough first pass that strips dead air and retakes, is now a Claude Code skill. The 'easy way' to build agent workflows: turn on plan mode with a capable model (Opus or Fable), scope the task with a brainstorm skill, accept the plan, then switch to a cheaper model for execution while giving feedback as questions so Claude learns the reasoning, not just the fix. His hard rule: never automate a skill, like motion graphics, you're still personally learning.

Free for members

Chat with this breakdown — free.

Sign in and you get 23 free chat messages on us — ask for the hook, quote a framework, find the exact transcript moment, generate a markdown action plan. Bring your own key when you want unlimited.

Create a free account →

Chapters

Where the time goes.

00:00 – 00:28

01 · Cold open

Unscripted intro promising five tips for building an AI editing workflow in Claude Code and DaVinci Resolve.

00:28 – 01:53

02 · Tip: you need editing experience first

The walking analogy: you can't teach a workflow you've never done yourself, or you'll miss the mistakes an experienced editor would catch.

01:53 – 02:52

03 · His own editing workflow, turned into a skill

Organize clips, ingest, listen through everything, then a rough first pass removing pauses and retakes, now run as a Claude Code skill.

02:52 – 03:50

04 · Aside: lost mic audio mid-shoot

The wireless mic disconnected from the screen recording partway through, so parts of the video may cut abruptly.

03:50 – 04:48

05 · Tip: two ways to walk your agent through execution

An easy way and a harder way to teach an agent your editing style; the harder way gets closer to your personal style but takes more setup.

04:48 – 06:13

06 · The easy way: plan mode, model choice, the brainstorm skill

Turn on plan mode, pick Opus or Fable depending on plan tier, scope the task with a brainstorm skill, then describe the raw footage and what to remove.

06:13 – 07:28

07 · How Claude plans, executes, and lands cuts in DaVinci Resolve

After accepting the plan, switch to a cheaper model for execution; Claude transcribes, gets timestamps, makes cuts, and lands the timeline in DaVinci Resolve.

07:28 – 08:27

08 · The feedback loop that turns a process into a skill

Give feedback as a question, ask if the result is close, and repeat until a converged process is worth saving as a skill.

08:27 – 09:23

09 · Training an agent takes longer at first, and that's fine

Expect the first attempts at training a process into Claude to feel slower than just doing it yourself.

09:23 – 10:47

10 · What not to automate yet

Never automate a skill you're still personally learning, like motion graphics, using the long-division analogy to explain why.

10:47 – 11:32

11 · Wrap-up and next episode tease

Wraps the freestyled video and previews a follow-up episode showing the full agent edit of this footage.

Atomic Insights

Lines worth screenshotting.

If you've never edited by hand, you can't build a good AI editing workflow because you have no reference for what a correct cut looks like.
A rough first pass just means stripping dead air, pauses, and retakes, nothing more sophisticated than that.
On a $20 Claude plan, use Opus for planning; on the $100-$200 plans, switch to the Fable model instead.
The 'brainstorm' skill, credited to a developer named Obra, is used to scope an editing task before Claude executes it.
Once Claude finishes planning, switching to a cheaper model for the execution phase conserves tokens without hurting the actual cutting work.
Feedback works best framed as a question: asking 'is this close to your result?' teaches Claude the reasoning behind a correction, not just the fix.
A process only becomes a reusable skill after several rounds of feedback converge on a result the editor actually accepts.
Never automate a skill you're still personally learning, you won't be able to judge whether the agent's output is actually good.
Automating a skill you haven't learned is like using a calculator before you understand division: you'll get an answer, but you won't catch it when the answer is wrong.
A full automated edit pipeline stacks in sequence: first pass, polish, VFX pass, text animation, sound effects, then a final manual pass.
The agent isn't always correct in video editing specifically, sometimes it makes noticeably strange choices that still need a human check.

Takeaway

Know your craft before you automate it

AI EDITING RULES

Handing editing off to an agent only works once you understand the process well enough to judge its output, which is why experience has to come before automation.

02Tip: you need editing experience first

A workflow built by someone who has never edited by hand has no reference point for what a correct cut actually looks like, the same way you can't teach someone to walk if you've never walked.
Treat early automation attempts like learning to walk: expect early agent workflows to miss obvious mistakes an experienced editor would catch instantly.

03His own editing workflow, turned into a skill

A repeatable editing process, organize clips, ingest, listen through everything, then cut out pauses and retakes for a rough first pass, is what an agent can reliably learn as a skill.
Once a manual process is consistent enough to describe as steps, it's ready to be handed to an agent instead of done by hand every time.

05Tip: two ways to walk your agent through execution

There's an easy way and a harder way to teach an agent your editing style; the harder way gets results closer to your personal style but takes more setup.

06The easy way: plan mode, model choice, the brainstorm skill

Turning on plan mode and using a strong model (Opus on lower-tier plans, Fable on higher ones) before executing anything lets the agent scope the task instead of guessing.
A dedicated brainstorm skill can scope a task in plain language, describing the raw footage, what to remove, and what to flag for review, before the agent touches anything.

07How Claude plans, executes, and lands cuts in DaVinci Resolve

After accepting a plan, switching to a cheaper model for execution conserves tokens without hurting the quality of the actual cutting work.
The agent transcribes footage, gets timestamps, makes the cuts, and lands the resulting timeline directly in the video editor for a human polish pass.

08The feedback loop that turns a process into a skill

Feedback lands better framed as a question: asking 'is this close to your result?' pushes the agent to reason about why a cut was wrong instead of just accepting a fix.
A process only becomes a reusable skill after several rounds of feedback converge on a result you're willing to accept.

09Training an agent takes longer at first, and that's fine

Expect the first few times you train an agent on a process to feel slow and awkward, that's the cost of building something you can eventually run in your sleep.

10What not to automate yet

Never automate a skill you're still personally learning, like motion graphics, because you won't have the judgment to tell good output from bad.
Skipping a skill you haven't learned by hand is like using a calculator before you understand division: you'll get an answer, but you won't catch it when the agent's answer is wrong.
The agent isn't always correct in video editing specifically, so treat its output as a draft to be checked, not a finished decision.

Glossary

Terms worth knowing.

First pass: The initial automated cut of raw footage that strips dead air, long pauses, and bad retakes down to a rough, watchable timeline.
Plan mode: A Claude Code setting where the agent scopes and explains its intended approach to a task before executing any changes.
Brainstorm skill: A Claude Code skill, credited to a developer named Obra, used to scope out a task in plain language before the agent starts executing it.
Fable: A Claude model available on higher-tier plans that the creator prefers for planning-phase work over Opus.

Resources

Things they pointed at.

04:19toolClaude Code

04:41toolFable (Claude model)

04:49toolBrainstorm skill (by Obra)

07:07toolDaVinci Resolve

Quotables

Lines you could clip.

00:29

“If you don't know how to edit or you've never edited before, creating a workflow from scratch is really hard because you don't have a reference to what works.”

states the video's whole thesis in one line→ TikTok hook↗ Tweet quote

07:29

“There was a research online that says Claude actually learns better if you ask it questions and have it figure things out on its own because it understands the why.”

counterintuitive claim about how to prompt an agent→ IG reel cold open↗ Tweet quote

08:54

“Don't automate anything that you're still trying to currently learn about.”

tight, quotable rule→ newsletter pull-quote↗ Tweet quote

10:18

“I very highly discourage you just accepting whatever the agent tells you because the agent, in terms of at least video editing, I've found isn't always correct.”

warning payoff line after the automation pitch→ TikTok hook↗ Tweet quote

The Script

Word for word.

Read-along

Don't just watch it. Burn it in.

See every word as it's spoken — crank it to 2× and still catch all of it. The same dual-channel trick behind Amazon's Kindle + Audible.

analogy

This entire video is unscripted. I am not looking at anything. I usually script.

Five tips I would give to someone who's currently trying to build an AI assistant editor using Cloud and DaVinci Resolve. Okay. After this video is done, I'm going to walk you through my full AI editor workflow.

You're going to see the whole entire thing. So tip number one in learning how to edit with AI workflows. This is the number one thing that.

I have to mention is that yes, you can shortcut the way that you edit, but the problem here is that if you don't know how to edit or you've never edited before creating a workflow from scratch is really hard because you don't have a reference to what works. Meaning what's that mean? For example, if I try to teach you how to walk, but I've never walked in my own life, it's going to be very weird and strange.

It's like eventually you can learn how to walk. But if I really know all the mistakes that typically come with someone learning how to walk or walk in general, it's like, hey, when you're walking down the street, watch out for any curbs. Whenever you're stepping over a curb, make sure that you lift your feet up.

If you see a crack on the road, make sure that you go over it. If you see ice, make sure that you walk slowly. Don't run.

So. That is kind of the manner that we're looking at. And how this applies to prompting is that if whenever you're going through a cut, right?

Like essentially in terms of editing, there is a series of workflows that you have to do, a series of steps that you always have to do. So in terms of editing workflow, what do you usually have? For me, this is how I look at my own editing workflow, okay?

I first organize my clips, ingest into my video editing. I listen to all the, all the videos and I start cutting it. I start cleaning up, take out any mistakes that's currently in there already.

Like any pauses, long pauses, any retakes. Those are the things that I always remove on my first pass. I'll clean it up to the point where it's what we call a rough cut.

So with this rough cut, I'm at this point where. I can move clips around. I can choose to remove certain sections, especially if it's too long.

Yeah. So, I mean, that's kind of my first skill. This is what I call a first pass.

And I've built this process into a skill so that whenever I, I don't personally manually do this anymore. I run it through my agent, which in this case is Claude and Claude will go through. So I'm not really sure how long this full video will be.

Just that if you stay long enough, you're going to see me go through this whole entire process. And I guess in some follow -up videos, I'll do how I build the skill for each one so that you can build it yourself as well. I'm currently doing an AI research edit call where I learn about how people edit, the different workflows, because what works for me won't necessarily work for you.

All right, so big issue. We lost some of our audio. the way that i have my mics connected we have a mic connected to a shotgun mic like a dji wireless mic and so somewhere throughout this process i noticed that the audio from my screen recording over here stopped moving and so i don't know where that's going to be at we might end up just cutting out that section so if it jumps randomly means that we cut out that section this is either number four or number five I'm going to say both.

And we'll just move from there. Okay, so tip number four is walking through and doing the actual execution with your agents. For example, let's say that you want to teach your agent how to cut your videos.

However, you typically do it. So there are roughly two ways. I'll go over the easiest way and I'll go over the harder way.

But the harder way. is how you get the agents to cut more similar to your own style. Okay, so number one, the easy way.

The easy way is whenever you want to do something, you want to go into Claude, you want to turn on planning mode and use, I would say if you're on the $20 plan, you can use Opus, but if you're on the $100 or $200 plan even, then use Fable, okay? Turn on fable, turn on plan mode and go into brainstorm. Brainstorm is a superpower skill that you can find online is his name is by or Obra.

I think, I think that's his name, Obra. You can find that in the, as a plugin as an, what do we call it? GitHub repo.

However, there's a lot of ways that you can get these two, uh, the superpower skills. go in to Claude and tell it I have a bunch of raw footage that I want you to cut up, take out all the retakes, take out all the dead spaces, and then I want you to mark out sections that have gone on too long. You can flag those for me to review, use a brainstorm.

That's literally the prompt that I would give it if I were going through this process. Now I just type in run first pass on this video. or actually i just i literally say run full edit on this video and it goes through all my different skills it runs through first pass then i go through the polish i do the polish phase myself after which it will go through the vfx pass the text animation pass sound effects and i'll do my final pass on it to make sure everything looks good i don't want to add anything else but typically i'll add more things and that will extend the amount of time I spend on the video.

Once that happens, cloud will go through all of its planning. It will do the thinking and it will figure out all the technical aspects. Well, during that phase, it's going to do all the planning and it will ask you a series of questions to walk you through.

After it's done and you answer the questions, you can press accept and go. like accept the plan and put on auto mode. From there, it will run everything through.

At this moment, you want to switch to a lower performing model, especially if you're trying to conserve your tokens. If tokens isn't a concern for you, then you can just let it run. But like me, I like to do all of my planning on the smartest model, then flip the switch and go to one model cheaper.

As it does that, it will go... do its magic, transcribe the footage if it needs to, get timestamps, and start making those cuts. After it starts looking at the timestamps, makes all the cuts, marks it on its side, it will start putting our timeline into DaVinci Resolve where we can make any polish or we can start giving it feedback.

If we give it feedback, We have to make sure that it understands why. Or we ask it questions.

There was a research online that says that clot actually learns better if you ask it questions and you have it figure things out on its own because it understands the why. It's kind of, in this regard specifically, it's very similar to a human, which I find really interesting. And it's just a process of going back and forth.

Is this close to your result? No. Okay, let's give it feedback and end with the question.

Is this close to your result? Yes? Okay.

Let's turn it into a skill. And that is how I build skills within DaVinci Resolve, or sorry, and that's how I build skills within Cloud. It is a very simple process after you've done it a bunch of times, but at least the first few times that you use it, it will be very weird, very awkward, just how it is in starting anything new.

It might take you, it might, it probably will take you longer depending on how fast you prompt and how fast you iterate on your problems to do this process rather than just cutting it yourself early on. And that's okay. The idea here is that you're training Claude, you're training somebody to help you automate a process that you could do in your sleep because you understand it so well.

Here's where you don't want to automate. Don't automate anything that you're still trying to currently learn about. You don't understand the full process.

Let's say you're creating motion graphics. Well, you don't understand the full process and you need to understand what makes good motion graphics, what style fits for you. I wouldn't necessarily automate that part.

You will have to do a lot of work. I've done a pretty decent chunk of work. in refining the motion graphics that goes on this channel here to the point that you see on screen.

So anything that you've never done before, at least manually, it's how I would always compare it is it's similar to doing long division or long multiplication. Yes, you can pull out a calculator. You can just punch in the numbers and press equals.

You'll get your answer. But if you don't understand the mechanics of how to multiply or long division by hand, then you'll actually create more trouble for yourself because you don't understand what that process looks like.

And then when it comes to automating it, it will feel really weird because now the agent is telling you how to do it rather than you telling the agent how to do it, which again, I very highly discourage you just accepting whatever the agent tells you because the agent in terms of at least video editing, I've found isn't always correct.

Sometimes it does very weird things, like very weird things. I might do something weird in this video, honestly, that might be a welcome thing for it to do, or it may not. Well, this video is, I see it's getting to 30 minutes real quick now, so let's wrap up here.

This is a little bit different format than usual, but you know. I guess I just want to show you guys what a freestyled video would look like with the agent editing it. Let's see how long it can take this video.

Our goal is to take this video from currently 30 minutes all the way down to about 12 to 15 minutes about. All right, thank you for watching. In the next episode, I'm going to go into how I take all this footage and edit it through my agent.

You'll see the behind the scenes of me doing that. And after that... if there's any particular skills that you're interested in we'll go deeper in how i built those skills the thinking behind it how it looks like running in cloud code desktop all those good things all right thanks for watching peace

The Hook

The bait, then the rug-pull.

The creator opens by admitting the video is unscripted, then promises five tips for building an AI editing workflow in Claude Code and DaVinci Resolve. He only cleanly delivers three before losing track of the count himself, folding two more in as 'four or five' after a mid-video audio dropout.

Frameworks