DaVinci Resolve's New Native AI Agent, Put Through a Real Edit
A video editor runs GPT-6 Astra through DaVinci Resolve 21.1's brand-new native AI connector on his own unedited footage, then teaches it his personal cutting rules mid-project.
Posted
2 days ago
Duration
Format
Demo
educational
Views
4.3K
51 likes
57 · 43
Big Idea
The argument in one line.
DaVinci Resolve's new native AI connector lets a language model operate the timeline directly instead of through a bolted-on plugin, and it's capable enough that an editor's job starts shifting from cutting footage to directing and correcting an agent that cuts it.
Who This Is For
Read if. Skip if.
READ IF YOU ARE…
A video editor or content creator who already uses DaVinci Resolve and wants to know exactly what the new native AI connector can and can't do.
Someone deciding whether DaVinci Resolve Studio is worth paying for versus staying on the free version for AI-assisted editing.
A solo creator curious about handing off rough cuts and motion graphics to an agent instead of cutting everything by hand.
SKIP IF…
You edit in Premiere, Final Cut, or another NLE. This connector is built into DaVinci Resolve Studio only.
You want a step-by-step installation tutorial rather than a demonstration of what the tool does on real footage.
TL;DR
The full version, fast.
DaVinci Resolve 21.1 ships a native connector that lets an AI agent control the timeline directly, replacing a slower third-party MCP setup. Studio only. Tested on the creator's own same-day, unedited footage, the agent produced a full rough cut in about fifteen minutes using an existing personal editing skill, then folded specific timeline-marker corrections into a tighter second render. A later, deliberately vague prompt had it place motion graphics overlays that stayed synced to paused frames in looping screen recordings. The conclusion: the tool's ceiling is set by the editing judgment fed into it, and the likely shift is from creator-editor to creator-director.
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.
The video opens already mid-demo, with the agent assembling a timeline unprompted and catching its own scoring error before the host explains what's happening.
00:45 – 03:34
02 · Two ways to connect, and what the free version gives up
Walks through the third-party MCP versus the new native connector, how to open it (Files > AI Assistants), and that the native path is Studio-only.
03:34 – 05:35
03 · The first rough cut
Prompts the agent to cut a same-day recording using an existing personal 'first pass' skill; the agent assembles a rough cut in about fifteen minutes.
05:35 – 08:08
04 · What's still the human's job
Reviews the AI's first cut and marks dead space and motion-based cut points directly on the timeline, explaining the judgment calls a human still supplies.
08:08 – 08:59
05 · Version two: did teaching it work
The agent reviews the timeline markers, explains the causes behind its misses, and renders a tighter, nearly 11-seconds-shorter version two.
08:59 – 11:32
06 · Pushing it further: an open-ended motion graphics pass
A deliberately vague prompt lets the agent choose overlays and animations for a motion graphics sequence, syncing them to paused frames in looping screen recordings.
11:32 – 15:45
07 · What this means for editors
Frames this as editing's 'version 3.0,' predicts a shift from creator-editor to creator-director, and closes with a free skill giveaway and a research call offer.
Atomic Insights
Lines worth screenshotting.
DaVinci Resolve 21.1's native AI connector only exists in the Studio edition; the free version's AI tool-calling is far less capable.
A same-day, completely unedited recording became a full rough cut in about fifteen minutes once handed to the agent with a plain-language prompt.
The agent applied an existing personal editing 'skill' to unfamiliar footage without being explicitly told to use it.
By the creator's own estimate, an AI first pass lands around 80% of the way to a finished cut, not further.
The concrete rule for tightening a rough cut is to find the first frame with visible motion blur and cut there, landing every edit on motion instead of stillness.
Feedback given as literal timeline markers, not freeform chat instructions, is what let the agent fold specific corrections into a second render.
A revised cut ran nearly 11 seconds shorter purely from applying markers the human had already flagged in the first pass.
A deliberately vague prompt was used as a test of agent judgment, not carelessness, and the agent matched overlay placement to specific paused frames in looping screen recordings.
The same open-ended test run on a different leading model produced visibly weaker results, pointing to a real capability gap between models rather than a connector problem.
The historical framing used is three discrete eras of editing technology: physically cutting film, then software timelines, now an agent operating the timeline directly.
The predicted career shift is from creator-editor, who does every cut by hand, to creator-director, who sets the vision and reviews the agent's output.
Takeaway
The tool's ceiling is your own editing judgment, written down.
WHAT TO LEARN
A native AI connector can cut a real rough cut in minutes, but what it produces is capped by how clearly a person's own editing rules get fed back into it.
01Cold open: it edited itself
The demonstration opens already in progress, with the agent controlling the timeline unprompted, which sets the bar the rest of the video measures against.
The agent caught its own factual error mid-task and flagged it before being asked to double-check anything.
02Two ways to connect, and what the free version gives up
There are two integration paths: a third-party MCP built by an outside developer, and a new connector built into DaVinci Resolve itself.
The native connector only exists in DaVinci Resolve Studio; the free version's AI tool-calling is described as much less capable.
If Studio's cost is the blocker, cheaper motion-graphics alternatives like Hyperframes or Remotion are a reasonable substitute.
03The first rough cut
Handing an agent a same-day, still-unedited recording and a plain-language prompt produced a full rough cut in about fifteen minutes.
The agent used an existing personal editing skill without being told to, and applied it to unfamiliar footage.
A first pass isn't meant to be final: by the creator's own estimate it lands around 80% of the way to a finished cut.
04What's still the human's job
Reviewing an AI cut for small judgment calls still takes real time, even when the first pass is already solid.
The instructable rule for tightening a cut is concrete: find the first frame with visible motion blur and cut there, so every edit lands on motion instead of stillness.
Leaving a small amount of breathing room between cuts is a deliberate choice an agent has to be told explicitly, not something it infers on its own.
05Version two: did teaching it work
Feedback given as timeline markers, not freeform chat text, is what let the agent fold specific corrections into a second, tighter render.
The agent explained its own miss, a timing threshold that let short pauses through, rather than just silently applying the fix.
A revised cut ran nearly 11 seconds shorter purely from applying markers the human had already flagged.
06Pushing it further: an open-ended motion graphics pass
A deliberately vague prompt was used to test how much the agent would decide on its own, not as a shortcut.
The agent matched replay loops and pauses in screen-recorded demos to the commentary about them, keeping an overlay in sync with a specific paused frame.
The same test run on a different leading model produced visibly weaker results, pointing to a real gap in model capability rather than a connector problem.
07What this means for editors
A three-era framework treats this as a discrete shift, not an incremental feature: physically cutting film, then software timelines, now an agent operating the timeline directly.
The likely career shift is from creator-editor to creator-director: setting the vision and reviewing output, rather than doing every cut by hand.
A tool's ceiling is set by the skill or rules fed into it, so a creator's own documented editing judgment becomes the differentiator, not raw access to the tool.
Glossary
Terms worth knowing.
MCP (Model Context Protocol)
A standard connection that lets an AI agent read and control an application directly, instead of a human copying instructions and results back and forth by hand.
Native integration
An AI connector built into the software itself, in this case DaVinci Resolve Studio, rather than added on top by a third-party plugin or script.
First pass
A rough, unpolished assembly of raw footage cut down to something watchable, before any refinement pass. About 80% finished, by the creator's own estimate.
Creator-director
A role where a person sets the creative vision and hands the actual cutting, effects, and assembly work to one or more AI agents.
“Even if you're not a video editor, I get why video editors are so scared.”
cold-open hook that names the anxiety the whole video answers→ TikTok hook↗ Tweet quote
05:07
“I just watched through the first minute roughly and the cuts were very solid.”
credible, understated praise delivered right after the reveal→ IG reel cold open↗ Tweet quote
07:15
“This is the boring part of editing that I've done hundreds of times... I hated it. Absolutely hated it.”
an honest confession that reframes what the AI is actually replacing→ newsletter pull-quote↗ Tweet quote
11:01
“It's like having a very, very smart junior editor now.”
the single line that sums up the whole thesis of the video→ TikTok hook↗ Tweet quote
13:15
“We've hit kind of like the version 3.0 of video editing.”
names the historical framing in one clean line→ IG reel cold open↗ 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.
17px
metaphoranalogy
And even if you're not a video editor, like, I get why video editors are so scared. Okay, it is currently assembling everything. See?
Codex is currently controlling everything now in just a few minutes. Interesting. And it did find an issue here as...
Oh, look at that. Holy... It just put everything onto the timeline.
It was so good at getting the exact point. What you're seeing is an AI agent working inside of DaVinci Resolve. I set it up, I run it on my footage, and I show you what you need, what to do without Studio, where your job stands, and what changes for creators, people making content to get customers, and full -time editors.
If we click here, check for updates, we're going to see that DaVinci Resolve 21 .1 is now officially out. There are two of these, the third party one I've been running, and now the one built into resolve itself. The previous integration, it was by a guy named Samuel Gursky.
There's a whole GitHub repo on it and everything to set up this process. But in order to set this up, very simple. You're going to go into your menu up here, click on files, go all the way down and set up AI assistance here.
It will pop up. The only one that was a little bit different was cloud desktop. It had to install an extension.
Other than that, very easy setup. That native connection is DaVinci Resolve Studio only. If you're on the free plan, you do not have this.
And for the non -studio version, it is much more inefficient at how it does the tool calls and what tools it has access to. It's actually way worse than the third -party integration that was here before. This run specifically used a free version of DaVinci Resolve.
There were no scripting connections open for Codex to use a third -party MCP, so it managed to make it work, but through Lua scripting and a lot of computer use. Now, if the Studio version is too costly for you, maybe my recommendation would be to go use Hyperframes or Remotion. The Studio MCP, though, is on a whole different level.
So let's make something clear first. Fable is part of Claude, and Fable is Claude's highest... thinking model.
Whereas Astra, GPT -6 Astra is ChatGPT's state of the art model. And so these two, it doesn't seem like, let me try it one more time through connecting it on here. So we have to close out.
I don't see it. Oh, these are the different tools that we would get. So definitely seems like way less tool than what we're already using.
So may not even be that much better. But now, since we have this native MCP connector with DaVinci Resolve, this actually changes so much about what we can do. The native integration is a lot better than I expect it to be.
It's so very powerful. You had to create a transition and bake it in. You export it, render, then put it back on the timeline.
If you switch it, you redo that whole entire process. It was very wasteful and time -consuming. Can you figure out and help me set this up for Codex so that we can have a native connection?
Oh, okay. So we didn't have to do anything for it to be installed. Well, that was pretty easy.
That line on screen is the one you try. If it reads your timeline back to you, You're connected.
And real quick, if you want to talk through your own setup, there's a research call linked in the description. Let's do this instead. I don't want to run the skill quite yet.
So I'm going to have Codex or Astra here. Create for me a rough cut of this video. Create for me a rough cut of this video.
Use all the footage inside. This is actually one of the videos that I was working on for this week. And so I've literally just finished it.
yesterday night. See how it differs from what Claude would have created. Okay, so it is using my skill already.
I didn't even tell it to use my skill. We already have a skill in terms of getting a rough cut. It's called first pass.
This is the skill that I use for all of my talking heads. I've refined it and have pretty much got it to the point where I'm fairly happy. It doesn't do 100%.
It does about 80 % of that first cut. So what's the first pass? A first pass, it takes everything you filmed, all the raw footage, and it cuts it down to something you can actually watch through.
Use your best judgment on giving me that rough cut. Let's look through our plan real quick. Let's see what it does.
Okay, it is currently assembling everything through here on a new timeline. Oh, okay, see? Codex is currently controlling everything now.
Interesting, and it did find an issue here as well. In the lower thirds, you say both entries score 3, but the screen records A as 5 and B as 3. That affects the final average.
I'll preserve the recorded material and add red review markers at that round and the tally. Okay, and in just a few minutes, we have finished the first cut. 15 minutes.
to get your first rough cut i just watched through the first minute roughly and the cuts were very solid there wasn't that much dead and open space that i would have taken out maybe like one or two frames at max as smooth as an agentic editor can get right it wasn't perfect i still would have changed a few things but it's so small that i don't you definitely wouldn't have noticed and even if you're not a video editor like this changes the game so much for you like even more so i get why video editors are so scared now especially of like gpt6 astra this is one thing i'm curious about and has been a constant issue is that i typically have to do a polish pass inside of my own videos And you have three frames that have basically nothing of value to offer the video I would Cut this up right here so that it starts on this slight motion This gives it a little bit of what I call breathing room whenever we're watching videos We don't want the cuts to be like way like boom boom boom boom boom.
We want it to be smooth I'm gonna mark this On the timeline, we'll mark it in white. Take note of any empty dead space that doesn't show anything happening on screen and nothing is being said.
We're going to mark it. This is where motion begins on. My general rule for making cuts like this is find the first part where there is some motion blur on screen so that we're cutting on motion.
Cut on motion. cut the head early, cut once the subject is done and there's no more motion. The system was working to a 15 frame threshold.
So that's the feedback I give the agent as we go along. This is the most important part of building a skill. Small notes, questions you ask, that's where it picks up the most.
And this is the boring parts of editing that I've done hundreds of times over like 60 % of editing. It's just making small decisions like this. And it's very boring and nobody likes doing this part.
I mean, if you're crazy, you like doing this part. But me, even as like an editor, I made like my profession in editing. I hated this part.
Absolutely hated it. This is essentially how I build out my skills and kind of how over time I've refined my first pass so that it gets closer and closer every single video. It's learning.
The agent is learning, right? Astra, Fable, any other agent is learning how you edit. So I went through the video and have left you feedback on what I would cut and what I would change.
I want you to go ahead and review those markers. How would you implement these into your rules? We are rendering out the new version, version two.
And we found seven markers, okay. It went through, reviewed everything. And two causes explain most of these misses.
The 15 frame threshold let shorter empty pauses through and preserve the mark version one and make the fixes in version two. The cut is 10 .9 seconds shorter. The first time through, I did have to make quite a few adjustments.
Now this polish pass. It actually ran a lot smoother with fewer errors in general compared to my previous runs on the third party integration. I don't know.
This is the problem with changing too many variables at once. I don't think that this has to do with our integration, our MCP integration that is now native. I think it has more to do with GPT -6 Astra and its ability to understand.
Normally, this is where I go through the timeline for motion graphics and text animations. I'm giving it a vague prompt on purpose because I want to see what Astra will do. given only this amount of context.
Then go ahead and I'll let you decide the rest of the edits of the video. I want you to finish up, make a YouTube edit, turn this into a YouTube video that is engaging, retains people. In the previous edit, the app I record with, Screen Studio, was zoomed into the demo while it played.
That can be a little bit distracting. So it doesn't get fully shown. What I want you to do is overlay those parts that have me watching the actual animations onto our timeline let's see what's been going on since i've let it run it found all six clean renders and you're reacting to without the browser controls it went through our video folder and it worked out where each demo goes okay and this is the part where i find very impressive is that this these demos loop while you comment on them so a single overlay per round would drift out of sync i'm matching the replays and pauses too so a remark about particular frame still lands on that frame here in the full screen version when i was doing the fable version it didn't include any of these loops at all it knew that i uh was still re -watching these animation it like right here it goes a straight line whenever we're doing motion graphics around like planets we do want to give it a natural curve like right in the earlier part of the earth critique also needs a clean overlay at this point we're on the earth now and right here up next we're going to talk about how that line is straight from here all the way to the tokyo version it's just a straight line i mean yeah like that's a straight line it try to like push it in and right here is when it curves it
like it was so good at getting the exact point that i was referring to curve and there's gonna be a lot of people inside the comments too as well i've received it before who are gonna say like this is gonna put editors out of work or that you don't lose your creativity as long as you're not just overseeing it and you you can be the brain to execute and the agent is the hands to do these cuts it's like having a very very smart junior editor now like is not looking like a junior editor to me this is looking like almost the first pass skill i'm putting it out for free the links in the description and pinned in the comments it does expect a folder convention so you might have to adjust it a little bit the tool itself is limited to how far your own skill already extend this thing is very very powerful what changes for creators someone making content online to get customers and for a full -time editor Here's kind of what I see with the landscape.
A lot of creators are going to shift from just creator editors to creator director. You're going to be the one who envisions the whole thing in your mind and be able to delegate it out to different agents. Now, this is a very scary thing if you're a video editor, for sure.
Like if this is your profession, you should be very, very scared. But for somebody like a creator, a business owner. or just even a hobbyist.
This dramatically lowers the barrier of entry for video editors. Like having this tool, this MCP native connector inside of DaVinci Resolve, it will change how editing is done. Like if not already changing it.
I just can't see a future. It's like when the first editing program came out. You know, like Avid was probably one of the first largest editing program.
And that really shifted the industry from how video editing was done before. Like literally you take a roll of footage and you actually cut it, then you tape it back together. That was old school editing.
Like you would have to have very specialized skills for that. But once we moved into technology, more apps, people had to evolve and swap and switch up the way that they were. essentially doing video editing.
And now we've hit kind of like the version 3 .0 of video editing. If cutting like 35 millimeter clips was version one and video editing NLE's nonlinear editing programs were version two, this is version three of video editing. And it's finally arrived, at least with DaVinci Resolve.
Like this will shift the entire market. This is what a lot of competitors have tried to do in the past few months and like even the past year of trying to introduce these nice, pretty MCPs. But this is the one that has actually been done well.
I'm really excited to see more of its capabilities as the models get better, the tools get better. Like, I don't know, Astra has been phenomenal. in what is able to accomplish with the autonomy that it's able to accomplish.
If you saw earlier, I gave it a few instructions for the overlay and it went a distance further in terms of it. And that's just crazy. Fable wasn't able to do that.
Like I've ran these two tests like pretty close together and Fable and Astra just like quite a big difference in its capability and its thinking. So take it what you will from this. Video editors, if that's your only job, you might need to level up your skills real quick, very soon.
If you want to figure out how this fits into your workflow, I do a 45 -minute research call. Link is below. So this is a little bit shorter of a video from the full walkthrough.
The full walkthrough was hours long. And I mean, setting it up, I shortened the free run itself. I definitely shortened that a lot.
the editing rules, what I'm looking for while I'm editing, the skills that I use while I'm editing. All those things are things that I had to take out a lot of information just so that I can keep it under whatever this video is. But this could be like a 30 or 45 minute video if I did a full walkthrough.
Oh, what video do you guys want to see? promises or anything like that but leave it down in comments below if there's like a section that really interests you let me know with that thanks for watching i hope you enjoyed learning about where ai is headed peace
The Hook
The bait, then the rug-pull.
The video opens mid-demo: an AI agent is already assembling a timeline inside DaVinci Resolve, cutting on motion and flagging its own scoring error, before the host has said a word about what's happening. That's the pitch for DaVinci Resolve 21.1's new native AI connector, tested here on the creator's own unedited footage instead of a canned demo reel.
Frameworks
Named ideas worth stealing.
13:15model
Version 1 / 2 / 3 of video editing
Version 1: physically cutting and taping 35mm film
Version 2: nonlinear editing software (NLEs)
Version 3: an AI agent operating the NLE natively
A three-era model for how editing technology has shifted, used to argue that native AI control of an NLE is a discrete leap rather than an incremental feature.
Steal forframing why a new AI capability is a real inflection point rather than just another feature announcement
06:16list
First-pass cutting rules
Mark dead space where nothing is said and nothing happens on screen
Cut where visible motion blur begins, so the edit lands on motion
Cut the head of a clip early
Cut once the subject finishes and the motion stops
Leave a little breathing room instead of hard-cutting every beat
The concrete, repeatable rules fed to the agent as timeline markers so it can imitate a specific personal cutting style.
Steal forwriting a reusable prompt or skill file that has an AI agent imitate a documented personal workflow
CTA Breakdown
How they asked for the click.
VERBAL ASK
14:35link
“I do a 45-minute research call. Link is below.”
Soft, single-line offer near the very end, stacked right after a separate free-skill giveaway (email opt-in) mentioned a few minutes earlier, rather than one hard pitch.
Add Modern Creator as a preferred source and Google shows you more of our breakdowns in Search, Top Stories, and AI Overviews. It only changes what you see, and you can undo it in your Google settings anytime.
Add to Preferred SourcesOpens your Google source preferences with us pre-loaded. Tick the box and you're done.
How to connect ChatGPT to free DaVinci Resolve with an open-source MCP server so it can transcribe, cut, and caption a highlight reel, a feature Resolve normally locks behind its $300 Studio license.
A video editor wires ChatGPT's Codex into DaVinci Resolve through an MCP server, then lets it build a full first-pass cut on its own, mistakes included.
A content director runs 336 unsorted vlog clips through a Claude Code + DaVinci Resolve Studio pipeline that classifies A-roll from B-roll, proposes cutaway placements against four editorial rules, and drops the picks onto a real timeline — then shows exactly where it still needs a human.
A creator wires Claude Code into DaVinci Resolve through an open-source MCP server, then stress-tests whether it can actually follow marker-based, relative instructions the way a real junior editor would.
An editor reverse-engineers 900 of Alex Hormozi's uploads into a rule file, then points the resulting Claude Code system at three unrelated podcasts to see where the rules hold and where they break.
00:20
00:28
00:52
01:06
01:24
01:40
01:56
02:04
02:20
02:32
02:43
02:50
03:07
03:14
03:30
03:48
04:02
04:13
04:23
04:34
04:49
05:01
05:13
05:23
05:33
05:48
06:04
06:24
06:31
06:41
06:55
07:08
07:22
07:39
07:46
07:55
08:22
08:33
08:42
09:09
09:23
09:32
09:44
09:56
10:13
10:31
10:39
10:55
11:07
11:19
11:42
11:54
12:06
12:18
12:30
12:44
12:53
13:02
13:17
13:29
13:40
13:55
14:04
14:16
14:25
14:40
14:51
15:03
15:27
15:39
13:44
17:02
15:15
17:37
11:32
19:21