The workflow We'd recommend to any founder
Based on this experiment — and on working in design across early-stage products — here's a practical roadmap. Not theoretical. Not defensive about AI. Just what actually makes sense given the tools that exist today.
The key insight in this roadmap: AI is most powerful at the two ends — rapid prototyping at the start, precise development at the end. The middle phase — where the product's specific feeling and visual identity are defined — is where human design judgment matters most. Not instead of AI. Alongside it.
When do you actually need a designer?
Here's a practical guide, not a sales pitch.
What "the feeling of the product" means in practice
Every product creates a feeling when you use it — whether the designer intended it or not. Notion feels calm and organised. Duolingo feels playful and encouraging. A banking app should feel safe and trustworthy. Lull was designed to feel like picking up a snow globe: immediately beautiful, effortless to start, with a sense of something alive happening on screen.
These feelings don't come from one big decision. They come from hundreds of small ones made consistently: the spacing between elements, which information is shown and which is hidden, how a button responds when you press it, what happens in the moment between one screen and the next. AI can execute individual decisions well. What it struggles to do is keep all of these small decisions pointing in the same direction — toward one consistent feeling — especially when that feeling is subtle rather than obvious. That is the work a human designer does.
The cost of skipping the designer in Phase 02 doesn't show up immediately. It shows up when users churn at onboarding. When the product feels slightly off in a way no one can name. When you're three iterations deep in Claude Design and still can't quite get it to feel right. These are expensive problems to fix in Phase 03.
Use AI tools. Use them early and often. Bring in a designer when the work shifts from "does this concept work" to "does this product feel exactly right to the people using it."
The core insight
AI understands your users. It processes your pain points. It can build a working prototype that a real person can interact with in minutes. That's genuinely remarkable, and founders should be using it from day one.
What AI can't do is hold the full weight of a design intention — the accumulated reasoning behind micro-decisions that together define how a product feels to use. That's not a flaw. It's just where the human designer's work begins.
