It's so fast we do it with users in the room. Or in workshops.

You can tell the prompt to change the design style, switch to dark mode, add buttons and images. Anything really.

But now - We can scribble a very loose sketch - literally on a post-it or napkin and tell Claude to 'Build it'.

Currently we use UX engineers or designers to prototype out our designs.

We use clickable Figma files or front-end devs / UX engineers build quick and dirty prototypes in React.

🦸🏻‍♀️ Give people agency: People want to feel in control. AI needs to communicate how people's data is being used and allow them to make changes.

💡 Help people be successful: People want to do the right thing but might not understand how the tools work. AI needs to collaborate with users to give them the right level of assistance to achieve their goals.

🧶 Remember past interactions: Nobody wants to start from scratch every time. AI needs to build on past conversations.

🧪 Design for being wrong: We all know things don't always work as expected. AI needs to let people know when it's not sure about something.

🦄 Adapt to different styles: People communicate in all sorts of ways. AI needs to be flexible and adjust to you.

5. Assistance: What is the utility of the AI?

Be clear on its task. When building custom LLMs for customers they often have a specific expertise unlike the broad utility of ChatGPT, Claude etc. Design towards this.