Dive Club 🤿 Ed Bayes - Designing 6 months into the future at OpenAI
91 snips
Sep 9, 2026 
Ed Bayes, a design lead at OpenAI working on Codex and ChatGPT Work, explores how teams prototype AI products six months ahead. He discusses turning Codex into an agent-first command center, making coding agents accessible, and designing dynamic interfaces shaped by future models. He also shares how persona-led teams, evaluation loops, live prototypes, and strong research partnerships influence product direction.
AI Snips
Chapters
Transcript
Episode notes
Structure Design Teams Around Personas
- Organize design ownership around personas rather than narrow features so someone protects the user’s entire product lifecycle.
- A developer designer might cover code review, onboarding, desktop, and CLI experiences instead of optimizing one isolated feature.
Prototyping Evolved From Figma To Production Branches
- Ed Bayes moved from Figma prototypes to coded mini-products and finally to experimental branches that feel like the real product.
- Designers can now create a branch, connect APIs, and share an authentic interaction rather than merely presenting polished screens.
Design Language Replaces Pixel-Level Specifications
- Generative interfaces require designers to define skills, tokens, and behavioral guardrails rather than specify every possible pixel.
- Ed’s visual backpropagation prototype used model-generated interactive graphics, then improved through engineering and evaluation.