Taylor Hack and I recorded a new episode of The Leads Are Sh*t yesterday. A big part of the conversation was about the work that comes after you hit record.
Editing the video. Adding the branding. Writing captions. Pulling out clips. Creating a thumbnail. Writing the article, email and social posts that give someone a reason to watch it.
Recording the conversation is often the easy part. Turning that recording into something useful is an entirely different job.
During the episode, I described the workflow I wanted to build. The recording would land in a folder, AI would prepare the complete content package, and everything would be waiting for me when I came back to work.
The important part of the idea was how I finished the request: “I want it all delivered to me to review.”
We finished recording. Then I decided to see if ChatGPT could actually do it.
I gave it the recording
After the show, the raw Zoom recording was sitting in a folder on my computer. Normally, that is when the second shift starts.
Instead, I opened ChatGPT and gave it a fairly direct instruction:
“The recording is in this file. Edit the video, create a thumbnail for YouTube, write the REM article, create clips for Instagram Reels and build the rest of the content I normally produce.”
Then I gave it examples of past episodes, articles and branding so it could see what the finished work was supposed to resemble.
That part matters. I didn’t ask it to invent a show, a visual identity or a point of view. Taylor and I had already had the conversation. REM already had a brand. We already had examples of the content we publish.
The machine received the raw material and the standards. Its job was to turn them into finished production assets.
Before and after
The original recording had two Zoom windows surrounded by a lot of black space. The finished version had a complete REM broadcast layout, properly framed host footage, names, titles, episode branding and a consistent visual structure.
It also removed the opening false start, cut the private portion we didn’t want included and cleaned up the end. The actual conversation and camera footage remained intact.
What came back
In about 25 minutes, ChatGPT had produced:
- An edited and branded version of the full episode
- A transcript and draft captions
- Three short-form video clips
- A YouTube thumbnail
- A YouTube title, description, chapters and pinned comment
- A draft article for REM
- Newsletter, email and social promotion copy
- Review notes explaining what still needed human attention
I did have to buy additional usage credits because it burned through my included limit quickly. AI may be fast, but apparently editing an hour-long video and building an entire content stack does not qualify as light computer use.
Still, I did not have to leave the chat, open editing software, resize the video, remove backgrounds, design the thumbnail or start separate documents for every piece of content.
I asked for the outcome. It coordinated the work.
It didn’t replace the important part
This is where these AI experiments tend to get exaggerated.
ChatGPT did not create the original conversation. It didn’t build the relationships, experience or opinions behind what Taylor and I discussed. It needed examples to understand the REM brand, and it still needed someone to review the captions, article, visual choices and factual claims before anything was published.
One of the assets it produced was specifically a list of things a human should check. That is not a failure. That is the workflow working properly.
The goal was never to remove people from the process. It was to remove as much production work as possible before the point where human judgment becomes valuable.
That is the distinction I keep coming back to: AI-powered, not AI-centred.
The real opportunity is sitting in a folder
A lot of people are experimenting with AI by asking it to write a social post or brainstorm video ideas. Those things can be useful, but I think the larger opportunity is in all the unfinished work sitting between an idea and its final destination.
The webinar recording nobody edited. The video sitting on a phone. The podcast that never became clips. The presentation that could have become an article, email and client resource but didn’t.
Agents do not necessarily need AI to give them more things to create. They need help getting more value from the things they have already created.
That requires more than one clever prompt. You need to define what finished looks like, supply examples, identify what should never be included and decide where human approval belongs.
Once you do that, the conversation changes. You are no longer asking AI for a draft. You are giving it a job.
The episode became the experiment
The funny part is that the episode itself is about building this exact workflow.
Taylor and I spent an hour discussing whether AI could take a recording and handle the production work that follows. Then, as soon as the recording ended, I gave it to ChatGPT and asked it to find out.
About 25 minutes later, the episode had become the proof.
Watch the episode: Hit record. Then what? Building an AI content workflow for real estate