I wanted more than a finished graphic

I’ve been treating AI less like a magic button and more like a professional friend I can learn from.

About a month ago, I started using Claude Design to create motion graphics for videos for one of my clients. Yeah. I was blown away because it was one of those things that made me realize that I would have taken so much longer to design all the assets and then animate them. Now I can do one or the other, but to be able to both in the timeframe I need for a video is so hard.

However, even though I was using this awesome skill and tool to amplify my videos for a client and for myself, I still felt like I was cheating the system. Not because it wasn't good, not because I don't think other people should use it, but because I like being in a professional work environment, being able to know how to manipulate something after the fact, not have to waste more credits than I have to, and really The best way to make something look more original and unique is to have more of me working on the project or orchestrating it in some way.

The node tree gave the magic away

When DaVinci Resolve released version 21.1, they added an MCP server to their client, and that truly unlocked so many things. I'll write my thoughts on Vibe editing another day, but I wanted to share what it was like to experience being able to plug this new professional friend into an ecosystem I work in.

I tried having Claude build a complete Fusion composition. It created the background, text, shapes, and all the little pieces that made the animation work. Watching the finished graphic appear inside Resolve felt a little like magic.

Then I opened the node tree.

It worked; the best part was that I could review and begin to see how it pulled it together. I then could see how I could make this easier for it to build. In the initial build, it built the background plate inside of Fusion, but it could have easily pulled a graphic for the background plate.

A better brief meant fewer moving parts

On the next project, I changed my direction. I started from a mockup, asked Claude to render the pieces that should stay static, and kept the elements that needed movement editable inside Fusion. The composition was simpler, ran more smoothly, and was easier for me to adjust.

Claude still duplicated some assets and broke others into more pieces than I would have chosen. But now I knew what to look for. All in all it took about 15 minutes to build it on Opus 5.5.

Make something that teaches you back

That’s the part of working with AI that interests me most. It can give me a working example before I know how to build every piece myself. Taking that example apart teaches me what the tool is doing. What I learn helps me give it better direction the next time.

I don’t think you need to master every part of a process before you use AI to enter it. But if you stop at the result, you miss part of the opportunity. Pick one piece you’re curious about. Ask why it’s there, how it connects to everything else, and whether there’s a simpler way to do the same job.

AI helped me get an idea onto the screen. Curiosity helped me turn the result into a better workflow. Somewhere between those two things, I’m not only making the work faster. I’m learning how I want the work to be made.

There’s still magic in seeing something come together that quickly. I just don’t want the magic to keep me from opening the node tree.

What’s something you’re doing lately to keep learning or challenging yourself the world of AI?