My Next Guest with Daniel Ro
Two months of podcast episodes. Zero minutes of editing.
An agent cuts the multicam, builds the reels, writes the description and posts it to YouTube and Spotify. Here is what is actually happening behind the timeline.
Watch the episode
Two months of podcast episodes sitting on my YouTube channel. Not one of them edited by me.
Everything you see was cut by an AI agent. The multicam, the reels, the posting to YouTube and Spotify, the shorts that go to Instagram, TikTok, LinkedIn, X and YouTube shorts.
The question I get most is how the agent actually does the editing. So here it is. This was recorded in October 2026, so the tools are Claude Code and ChatGPT.
Editing was the bottleneck
Video editing is the thing I always hated. Since the wedding studio I started with a friend 10 to 15 years ago. If you shoot, you know the wall is always the same one, and it is always the editing.
1:50
The channel is a passion project. I do not care about the likes, I do not care about the views, I do not care if it ever monetizes. It exists because I always wanted to do it. It only exists because the editing stopped being my job.
There is no timeline
A 33 minute two camera podcast with my friend Madison, cut without opening Premiere, Final Cut or DaVinci Resolve. What those programs show you is boxes. Camera one, camera two, the cuts, the frames.
3:00
Those boxes are a visual representation of a list of things that need to be done. The agent builds the list. Premiere just draws it for you.
The timeline is a picture of the list. Nothing more.
Everything starts with the filming. Two 4K cameras, a Rodecaster, and a 65 inch TV because I like it. Then the footage goes on a hard drive. A machine at my place stays on with an agent watching that drive. New footage lands and it tells me there is a new job for this client, do I want it to start the edits.
Transcription comes first
The first step in every multicam is the transcription. You need it to find the best parts. I use Whisper, a piece of code that lives on your Mac. It takes a little while because it is accurate.
Nvidia has one called Parakeet that is very fast and not as accurate. For an hour or two hours of podcast I want the accuracy, so Whisper it is. A one hour recording runs about seven to ten minutes, depending on your machine. If you already have Premiere, drop both cameras in and export the transcript from there. Fast and accurate. I just do not want to touch anything.
Take this with you
Choose your transcription tool by the length of the recording, not the speed. On an hour of audio, accuracy costs you ten minutes and saves you the whole edit.
What the agent does with the list
Whoever is talking gets the shot. That is the camera cut. Because we shoot in 4K it can also punch in, so I ask it to find the best spots for close-ups.
9:03
Then it checks the plan and comes back to me. Daniel, it is safe to render, do you want me to do it. I get a preview first. I watch it, I approve, it runs the full render.
The render itself is ffmpeg. A video toolbox that has been around for ages. Color grading, audio corrections, sharpening, punch-ins, graphics, cards, titles, music. Anything you can think of in video editing goes on the list and ffmpeg puts it together. If something fails it fixes it on the fly.
Once the multicam is done it hands me a list of reels. On that episode it found 20. I can ask for more or fewer. Then it cuts them and makes 4x5 versions too, because those play better on LinkedIn and X.
The layout is mine. Two cameras stacked, a news card headline at the top with the name of the podcast. I defined it, the agent builds it. Yours can look like anything you want.
The part nobody selling you a course will say
No subscriptions needed. Not if you are new to video editing and AI video agents. Any influencer telling you that you need this app, this description, this other tool, do not believe it. Claude Code can walk you through the whole thing. You do not need skills. Go ask it, that is exactly what I did.
7:51
You don't need anybody else's course.
Publishing is part of the job
After the multicam and the reels, the agent uploads to YouTube. Title, description, chapters, thumbnail, hashtags, and the schedule for when it goes live.
The descriptions are written differently on purpose. Chapters on the timeline, the topics covered, a Q and A section I always include. ChatGPT, Gemini and Perplexity are getting better at reading video, but they can already read the description. Be detailed about what you actually talk about and the video gets in front of the right audience.
Spotify gets the same treatment with a shorter description, because that is the limit there. The agent adjusts.
Take this with you
Write your description for the models reading it, not just the viewer. Chapters, the topics, and the questions your episode answers.
Teaching it your style
Five years of Premiere and CapCut projects are sitting on my drives from client work. Those are the reference. Claude Code reads those project files and learns how I cut. Nothing complicated about it.
When something comes out wrong I do not open an editor. I tell the agent to fix this and fix that, because I want it to learn and do better next time. Every couple of weeks I ask it what it has picked up from my edits. Did you save that to memory. Is it part of our skill set.
A mistake means it's learning.
It can export a project file for Premiere or CapCut if you want to adjust things by hand. I do not do that. I just talk to the agent.
Run one recording through Whisper on your Mac, or export the transcript out of Premiere, and time how long an hour takes.
Point the agent at your old Premiere or CapCut project files so it can read how you already cut.
Ask it what it learned from your last two weeks of edits, and whether it saved that to memory.
Four hours a day, gone
Four hours minimum of editing every single day. That was the old number. Now I do not edit at all and we run over 20 clients, with one human employee. Shout out to Aaron. We go shoot, we come home, we let the agents work.
18:39
Four hours a day of editing. Now none.
The AI video editors are not perfect. Some projects will still need a human review. I am not afraid of the agent making mistakes, and I do not believe in perfectionism. Take the time to train one on the kind of work you actually do and it will surprise you.
If you only take three things
Find the one task capping your output. For anyone shooting video, it is the editing.
Keep the approval step. Preview, approve, then render, so nothing goes out unseen.
Correct the agent instead of fixing the file yourself, so the next edit starts better than this one.
Your next read
Tell me what to slow down on
The follow-up to this is a full live edit on YouTube, start to finish, nothing cut out. If there is a part of the setup you want me to go deeper on, send it over.
Send me a text