Holy shit... I just automated sth I thought was impossible with AI: product tutorial videos
The problem at MindPal was pretty simple: we have hundreds of AI templates to share. We know videos of these templates work - some have gotten us tens of thousands of views. But actually making them was a total nightmare.
We tried everything. At one point, we even hired a freelancer, but the feedback loop was exhausting. It actually took longer to give feedback and wait for revisions than it did to just make the video ourselves. It was slow, expensive, and impossible to scale.
When we did it ourselves, it was a massive grind:
→ Record the screen of the behind-the-scene agent builder
→ Record a demo of the agent working
→ Write a script that didn't sound like a robot
→ Record a voiceover or an avatar
→ Spend hours editing everything together
If my co-founder or I were tired or busy, the videos just didn't happen. I assumed this was just the "manual tax" you had to pay for quality.
Last weekend, I got fed up and asked Claude if I could just automate the whole damn thing.
Turns out, I can.
So I spent the weekend cooking something - an internal AI SOP to turn any workflow URL (yes, from just a single URL) into a publish-ready use case video that passes all quality standards in ONE GO.
Here is the new setup:
→ Playwright: Records the screen and even moves the mouse like a human
→ @Claude by Anthropic: Writes the narrative based on our actual product info
→ @HeyGen: Creates the avatar and voiceover
→ @Remotion: Programs the entire edit - syncing everything into a final file
→ @Zernio + @Railway: Automatically publishes the video and saves the assets.
Now, I just give the system a URL and a finished video comes out. I don't even have to click "upload."
I just wrote a post sharing the full behind-the-scenes build, the architecture, and the logic behind of this AI video agent. Check it out here if you think this could be helpful for your company: https://mindpal.space/article/ai-video-agent
P/s: This is what I wake up to every day now 😎✌️
Replies
Asa.team
The freelancer feedback loop problem is so real. Explaining + waiting + reviewing often costs more time than just doing it yourself.
Curious how you handle UI changes between recording and publishing. does the whole pipeline rerun or can you patch specific segments?
The predictability is doing the heavy lifting here. Most video automation fails when flows are too variable — sounds like you found a sweet spot.
This is a great example of AI automation working because the workflow already had a clear quality bar. The hard part is not just stitching Playwright + Claude + HeyGen + Remotion together, it is encoding taste: what makes the script not sound robotic, what counts as a useful demo, and where the edit should pause for clarity.
I’d be curious whether your SOP has an explicit critique/check step before publish, or whether the first generated version is usually good enough now.
This hits close to home. We went through the same grind with internal onboarding content, someone always had to record it, someone else had to edit it, and the moment the product changed the whole thing was outdated. The feedback loop with freelancers is exactly as painful as you described, sometimes it's genuinely faster to just do it yourself.
The Playwright + Claude + Remotion stack is clever. Curious how you're handling cases where the UI changes between when the script is written and when the recording runs, does it just pick up whatever's on screen or do you have some fallback logic for that?