← Notes

AI as a Second Pair of Eyes, Not the Artist

Most teams point AI at the start of the creative process. I have had more luck pointing it at the end. Generation is cheap. Review is where the leverage is.

Experiment 3 min read

The default assumption with AI in design is: use it to generate. Moodboards, variants, layouts, copy. I tried that for a few months and what came out was either average or aggressively on-brief in the wrong way.

The flip that worked: use AI to review.

After I have the direction and a few rounds of variants, I send the work back to a model with one of three prompts:

  • What is the strongest reading of this design? What is the weakest?
  • Where would a layperson get lost? Where would a peer designer roll their eyes?
  • If you had to remove one element to make this stronger, which one and why?

The model is wrong often, but it is wrong in useful ways. It surfaces a reading I had not considered. It catches the dead bit of negative space I had stopped seeing. It forces me to defend a choice or undo it.

Generation collapses options to one mediocre middle. Review opens options back up. For senior work the bottleneck has always been judgement, not output. AI is finally a tool for that side of the work.