I let Hermes/OpenClaw agent run our video making system for 10 days. This is what happened...
Video has always been one of our biggest customer acquisition channels at my company. But for 2 years, our video making process was a mess. Not the recording - the everything else. Finding editors who understood our specific AI context was hard, so we ended up doing it ourselves. It was draining. It was inconsistent. It was a bottleneck.
Over the last 10 days, I built an autonomous AI video agent (powered by Hermes - or OpenClaw - both work) to kill the drudgery for good.
The system now handles the entire pipeline end-to-end:
→ Sourcing ideas from news and trends
→ Recording product footage programmatically (yes, including human-like cursor movements) with Playwright
→ Sourcing visual assets and generating visualizations via code
→ Editing the final video using @remotion
→ Auto-publishing to our social channels hands-free with @Zernio
The impact was immediate. Our views spiked and our posting cadence is finally reliable. It’s no longer a burst of content followed by silence.
The most important thing I learned? Context beats control.
I stopped trying to micromanage every individual frame and started focusing on the foundational environment. If you provide the AI with deep product context and clear brand guidelines, it can make aesthetic judgments and creative decisions better than you might expect.
We moved from a manual mess to an autonomous system that actually compounds.
I’ve shared the full behind-the-scenes build, the architecture, and the logic behind how we automated our product demos and short-form content.
If you’re running a business where video is a key growth channel, this might be helpful.
Full breakdown here: https://mindpal.space/article/ai-video-agent
Have you tried asking Hermes/OpenClaw agents to make videos for you? How did it go?
Sneak peek: This is what I wake up to every day now 👇
Replies
ProdShort
I was very curious about the quality of the results, as I tried to do it using OpenClaw and Remotion at some point, and I got average results.
Are the results generated on youtube shorts, one shot, or there is multiple that go to trash ? I find Broll well great.
I can still see a huge difference between the one generated with Marwan, and the AI generated ones.
@bengeekly I’ve been struggling with video workflows too, and this really resonates.
@bengeekly @deangelo_hinkle Tried automating parts of my video pipeline, but never end-to-end like this. The idea of focusing on context over control really stands out to me
@bengeekly @deangelo_hinkle @lakeesha_weatherwax I’m wondering how you structured your brand guidelines so the AI could actually make good creative decision consistently
MindPal
@bengeekly @deangelo_hinkle @lakeesha_weatherwax Glad to hear that you find it helpful.
MindPal
@bengeekly I think there's still a lot we can do to further improve the quality of each video, but my main goal for this project, when I first started it, is to make the learning loop shorter so I can iterate on this system faster. Early results look promising to me, esp. on my personal channel: https://www.youtube.com/@sylviangth/shorts - I'm not a professional YouTuber, so having AI handle the execution part while I think more on what to say and the content is really helpful for me.
Regarding the results for the kinds of videos that we make consistently and there is a clear playbook, the AI pretty much can one-shot that. I just add a human-in-the-loop step for me to review the videos created every morning before they go live, but it rarely goes off track. I just added that human checkpoint in case I want some improvements on the editing or delivery of the video.