Since late last year, I have been working on an “AI agent” project. What that term actually means, I am still figuring out. During this process, I am discovering many things which are causing me to recalibrate what it means to design such a tool.

For the last several years, I have played a role that would usually be described as a User Experience Designer. The term “user experience” was coined in 1993 by Don Norman for his group at Apple Computer where he was working at the time. He describes user experience as, “It’s everything that touches upon your experience with the product, and it may not even be near the product. It may be when you are telling someone else about the product… It’s a system, it’s everything.” I truly believe this is important, and have worked hard to incorporate this into my own work. Working with artificial intelligence however, adds many new dimensions to what it means to design something.

Before computers were on the scene, designers could usually feel somewhat confident that the thing they were designing wouldn’t change too much once it left the design studio and entered the owner’s possession - whether it was furniture, a coffee mug, or a car. Although mechanical objects could get into various states through the controls offered, most of the time the variety of these states were finite and knowable, allowing the designer to craft and test the experience of each of these states. When designing an electric teapot, the designer can design how to make the teapot easy to handle when the water is hot, because this is an easily predictable use case. It would be reasonable for the designer to assume that the teapot won’t turn into a toaster after the user turns it on.

With the advent of software, the range of possible states that products could get into expanded dramatically. Most software was still largely deterministic. The designer could map out all of the possible states the application could get into, and craft good experiences for the end user before they encounter them.

Recently, we have passed into the age of true artificial intelligence. I say “true” because I think AI has finally reached the point that matches what most people have traditionally considered AI to be: that is, a computer that can talk like a human, sound like a human, look like a human, and think like a human. What people imagined AI to be in the 20th century does now seem to be a reality.

With these systems come all kinds of new considerations for the designer. As I mentioned, designers traditionally worked with systems that were largely deterministic. You could map out all of the possible states and design for them. With AI, this becomes much more difficult. It is much harder to predict what will come out of an AI system. In fact, you don’t want to be able to predict this, because if you could, you wouldn’t need the AI in the first place!

So, if we take Don Norman’s edict to heart - that designers need to be deeply interested in the entire end-to-end experience of the user, how do we design for systems that are so unpredictable?

Well, it turns out there are a lot of things we can do, and many choices can be made. But, they are much different things than what UX designers are used to doing for the last 30 years. Instead of deciding whether to develop the product as a Web app or an iOS app, we might be selecting different models from OpenAI or Google. Instead of designing where and when to put a button, we might be thinking about prompt engineering. Instead of considering the efficacy of a modal interface, we could be fine-tuning the model. Instead of designing a caching mechanism for the UI to make it more responsive, we might be designing a Retrieval Augmented Generation (RAG) system to give the system more effective knowledge.

This is tough stuff, and there is a whole lot to learn, but knowing how to work with these tools, and how to make these decisions are essential for designing an effective AI experience.

There is a lot of talk lately about how AI is likely to put a lot of people out of work. Yes, it will take on a lot of the work that designers and software developers have done traditionally. But, as always, new technologies will create new jobs too. One of these jobs could justifiably be termed the Intelligence Designer. The person tasked with this job would have the responsibility of crafting the AI to meet the needs of the end user as effectively as possible. Although some Intelligence Designers will be involved with the fundamental architecture and training of the AI models, the vast majority of people in the field will be working with the output of the big AI labs. Their job will be to take the raw capabilities of these models, and shape them into a wide variety of use cases, with different knowledge, personalities, and agentic behaviors.

So, if you are a software designer worried about jobs going away - fear not! If you are interested in designing AI systems, there is even more work for you. But, be ready to expand your ideas about what it means to be a designer. If you want to be an Intelligence Designer, you will need to think deeply about how these systems work, and be willing to let go of many old assumptions. Instead of designing GUIs, you will be designing something much more ambiguous, and probably a lot more exciting.