We need product teams with designers on them to make sense of AI

Designing digital products over the last decade has mostly involved applying well-understood techniques to novel use cases. We matured as a digital profession, built design systems, agreed on what good looks like, and made some useful things for users who haven’t previously benefitted. I think this is changing, and AI is throwing us back to a world of unknown unknowns, with lots to make sense of.

We don’t know what design patterns for AI products look like, or what new techniques we need to work with it. We are putting ✨ on everything and working out how to make copilots and chatbots that are genuinely useful. This work is really different to what working on the web has been like for the last while, and it’s exciting! There is a lot of new stuff to learn and make, and I think good design is fundamental to how organisations respond to this. Why?

AI teams need designers, because they need to build things. With advanced large language models just an API call away, it’s the products that we build with them that will make all the difference. If this is as groundbreaking as, say, the smartphone was, then it’s going to be fundamental to how businesses deliver products, and it needs to be deeply integrated within them. Empowered product teams with designers on them can do that pretty well.

AI teams need to adopt an experimental approach and to explore new opportunities at pace. Working with AI is complex and ever-changing, and designers are really great at practically navigating landscapes like this. We have tools and frameworks to quickly test assumptions and to try new product propositions. Unless you take an experimental approach, you won’t be able to make informed decisions about which opportunities to pursue, and risk not going beyond the chatbot.

AI teams need to understand their users well in order to build useful products for them. How much trust do users put in your brand? How exactly does a chatbot help them achieve a task? What will they do when receiving a piece of AI-generated advice? How do we increase affordance of generative products? How does a copilot feature fit into their workflow? If you’re not asking these questions from day one, you risk building things that only attract early technology adopters, do not solve genuine problems for users, or get you sued.

Most interesting and fast-growing start-ups of the moment are design-led. Look at Arc or Perplexity or Adept, or at how companies like Notion are integrating AI within their products. User experience, and your ability to design compelling product propositions is the differentiating factor.

That said, to work well on AI teams, designers need to adapt their processes:

• AI products can now have infinite outcomes, and it’s a lot harder for us to cater to every possible user journey. How do we design for this?
• AI products need to evaluated at scale to measure the quality of AI-generated outcomes. How does this fit into the design process, and the scrappy phase of endless pivots?
• Making workable prototypes is simpler now. Matt wrote about this, I made some demos too. How do we use this in our process?
• A lot of design decisions are taken when preparing the underlying infrastructure. How do we work side-by-side with data scientists and developers to help them consider impact on users?
• Data, with its biases and omissions, is the main ingredient of AI products. How do we learn to work with it, and how do we make it work for all of our users?

The early signs are not looking amazing, but embedding user-centred professionals in multidisciplinary AI teams is essential to ensuring new products genuinely help people.