How I actually use AI in a creative workflow

I have a Google AI Essentials certification and I spend my evenings training small models for fun, so people assume I want AI to generate the final design. I don’t. Most AI-generated visuals look like AI-generated visuals, and clients can tell.

What I want from AI is speed in the parts of the job that aren’t design.

Where it earns its place

First drafts of copy. Case study text, alt text, social captions, email subject lines. I write the brief, the model writes six versions, I keep the sentence I would have written on the third try.

Reference and mood. Before I open Illustrator for a logo, I’ll have a model describe ten visual directions in words. Ravens for an AI radio company. Wings and tiger scratches for a track team. Half are obvious, two are bad, one or two are worth sketching. That’s a good ratio for a five-minute exercise.

Grunt work in production. Batch-renaming exports, writing a script to resize 100 images to platform specs, turning a messy list of station formats into a clean table for InDesign. This is where “I can code a little” pays off every single week.

Rubber-ducking. Explaining a layout problem to a model often solves it before it answers. That’s not magic, that’s the same reason talking to a coworker helps.

Where it doesn’t

• The mark itself. Every logo on this site was drawn by hand in vector. Generated logos don’t survive a print vendor, a one-color embroidery, or a favicon.
• Brand decisions. Choosing yellow and pink for a sales brochure that had always been blue was a judgment call about the client’s audience. A model has no stake in whether it works.
• Anything the client will read as “the work.” If it’s the deliverable, a person made it.

The honest version

AI has made me faster at the 70 percent of a project that is logistics, and it hasn’t touched the 30 percent that is why anyone hires a designer. That ratio feels right to me. If it changes, I’ll update this post.