
Designing with the image model first

Building With Agents Today 🤖 — with Charlie Guo
Topic 14 of 16
Published Aug 10, 2026 · Clip 46:01–52:31 (6:30)
Charlie praises the new GPT image model and shares his workflow: design first with the image model as his designer, iterating with visual feedback and even mouse drawings, then hand the resulting image to the coding model as his engineer. This is faster and cheaper than arguing about button placement in code, and captures intent that plan-mode prompts miss — a picture is worth a thousand words. He uses it beyond web apps for motion graphics and animated launch videos, noting that as a non-Figma power user it's a real capability unlock, though OpenAI still uses Figma internally.