I wasn’t even planning to record anything, but this experiment worked so well that it felt worth sharing. What you’re seeing in the video is ChatGPT Atlas — OpenAI’s new browser-based agent mode. Instead of popping out into a separate window, Atlas works directly inside your browser. In this case, it’s running inside my Minnesota Dunker YouTube channel.
For context, the Minnesota Dunk Squad is something I started a while back. It’s a group of guys across Minnesota who love dunking, and we get together for sessions, events, and content creation. I batch-recorded a recent session and uploaded individual videos for each dunker. With Atlas, I didn’t have to manually title or write descriptions for any of them. While I sat back, Atlas went through every single upload and applied custom titles and descriptions on its own.
One moment really stood out. In one of the descriptions, it placed the “Book us for your event | Follow us” line in the middle by mistake. I figured I’d need to go back and fix it manually. But a few seconds later, it recognized the error and corrected itself across the videos. That was the moment I decided this was worth documenting.
Giving Atlas the Right Context
Before sending anything into Atlas, I prepped a regular ChatGPT thread first. I pasted in recent examples of my YouTube titles and asked it for more in a similar style. Then I dictated short notes for each dunker — things like their jumping style, strengths, or standout moments.
For Isaac, I emphasized that he’s a strong one-foot jumper and asked for four or five title ideas. What came back actually surprised me. The titles matched the style I already use and fit the vibe of the channel. Once I gave it additional context — the date of the session, the gym we filmed at (H4 Sports Academy), and the rim height — it generated full descriptions for each video. They weren’t perfect, but they were very solid and matched the overall tone of my YouTube content.
I repeated that process for every dunker: Rich (a six-foot jumper with big power), King James (a top high school prospect), and the rest of the crew. After I collected all the titles and descriptions for everyone, I dropped the entire thread into a new Atlas window and let it run.
Letting the Agent Work
While I was talking through the process, Atlas was already halfway through updating the uploads. It’s not flawless — sometimes it stalls, clicks the wrong thing, or gets confused about what step comes next. But even with the hiccups, it handled the boring part of the workflow for me: the repetitive manual editing YouTube usually forces you to do one video at a time.
This is the whole point of using agent-style tools. I could’ve gone off to shower or started another task in another tab while Atlas continued working. It handled titles, descriptions, and adjustments without me touching anything.
It does have limits. For example, we’re working on the Google Knowledge Panel book right now, and Atlas can’t directly edit Google Docs the same way. It has to use “suggest edits,” which slows things down. But in environments built around forms or repeated workflows — like YouTube — it’s incredibly useful.
Why I’m Sharing This
I wanted people to see a practical example of how Atlas can actually help creators, not just in theory but inside the actual tools we already use. YouTube is an obvious place to start, but this approach works across many platforms.
The agent gets better the more context you feed it: your style, your content, your goals, who your audience is, and what you want to automate. Once you build that foundation, it becomes a real assistant. It frees up your time so you can focus on the creative and performance parts of what you do, not the admin work behind it.
If you’ve been experimenting with Atlas, I’d love to hear what you’ve figured out or what workflows you’ve built. This is still early, but the potential is already huge.
About The Author
