Four design process principles used by successful AI product teams, from startups to Meta

At Austin Tech Week, companies from startups to Meta shared their design process for AI. These were the four principles they all agreed on.

1. When designing for AI, the problem isn’t generating ideas; it’s selecting the right ones to pursue.

Right now, everyone from employees to customers has ideas for new AI features. This doesn’t mean they’re all winners. Companies need robust strategies to ensure that the features they build provide enough user value to justify the engineering effort. For companies who aren’t developing their own models (which is most of them), this often means gauging the cost of using commodity LLMs, necessitating strong partnerships between research and engineering.

2. Concept testing early-stage AI concepts is hard because nothing comparable exists.

How do you get good user feedback on AI concepts that don’t exist yet? If you had talked to the average consumer in October 2022–before ChatGPT launched–it would have been hard to explain what ChatGPT is and how it differs from Alexa or Siri or any of the corporate chatbots that existed at the time. This explanation problem continues to challenge product designers as AI capabilities evolve.

3. At the same time, there’s more pressure to mock up new feature concepts early using generative AI.

In the words of Meta engineering manager Raghavan Vasudevan, genAI means “there’s no excuse” for coming to the table with just a feature idea, when you could illustrate it with a few well-chosen words in a prompt box. This makes it easier to present feature concepts internally, getting to shared understanding faster.

4. User-facing Gen AI requires preparing for unexpected user experiences.

The non-deterministic nature of genAI is both blessing and curse. Engineering SVP Andrew Mattie recounted the delight that ensued when Realtor.com users discovered they could search for “Harry Potter-esque” homes using a new AI tool (originally designed to support more mundane queries). But that same capability also enabled users to break the UX by entering overly long and detailed prompts. Designing for AI means accepting risk and being prepared to embrace the good and mitigate the bad.

These principles are examples of how UX and product professionals are evolving their design process for the volatile world of AI. The rapid pace of change makes it all the more important to ask for help, and to get involved in forums where we can learn from each other.

My own work involves helping leaders adapt their design process for emerging technologies like AI, smart glasses, and spatial computing, in a few ways:

🔎 Conducting user/market research for complex and early-stage emerging tech concepts

📈 Training their teams on customer research/innovation methods for AI and emerging tech

🎯 Coaching them to add new capabilities and solve problems in their innovation process

If you’re looking for support in your AI design process, I’d love to chat!