"These AI models need our help!" ...this was one of my biggest insights from a user test last week.

From the outside, it looked just like a traditional user study, featuring an upbeat researcher (that's me), an unknowing participant, and a series of tasks to be completed using a digital product. This test, however, stood out because we were assessing a novel AI feature.

I encountered a scenario I believe will become increasingly familiar in our field: I found myself analyzing the model's behaviors and perceptions as closely as the participant's.

It felt like two users were in the study, both deserving my careful attention as they tried to complete a task together. Just like any good user study, it revealed many 'aha' moments—apparent opportunities for improvement—but across two dimensions: the human and the model. 

There seems to be a hidden assumption that you can strap an API to any product and drive immediate value, bypassing the need for design. Yet last week's user test illuminated the opposite: we must study the expectations, behaviors, and curiosities of the human user as well as those of these AI models. 

The success of future AI experiences depends on design, and the time to build these experiences is now.

Understanding UX Design from a Different Perspective

Recently, many articles portray a gloomy year for the design discipline. We're witnessing harsh layoffs in favor of anything that will create profit and drive short-term growth. 

In the corporate environment, design is surrounded by an even larger cloud. With the arrival of the next significant era of technology, AI software companies are projected to grow 63% faster in 2024 than non-AI software companies.

Yet, demoralized designers are stepping back instead of leading confidently with our unique skills, waiting for engineers to take the lead. 

If designers are crucial to the success of AI, and AI is essential for the success of businesses, it's time to reverse course our craft.

Like previous eras, designers will enable organizations to overcome the monumental task of reinventing experiences and processes from the ground up.

Reinventing every CX/UX with AI Design Sprints

The reinvention of every CX/UX, which I'm calling AI UX Design (AUX), will also transform the object and output of Design.

To summarize…  

From digital interfaces → AI agents. Just as a product is synonymous today with a mobile app or digital interface, it will soon become synonymous with an AI agent. Instead of users being left to navigate an app or interface solo, they will be continuously supported by an AI partner. The value of every product will be evaluated by the quality of its built-in AI agent.  

From software features/flow → AI "skills." In AI UX Design (AUX), if a company wants to create differentiated value for customers, it will have one option: increase the functionality, or 'skills,' of an AI agent to solve more customer needs. Future product roadmaps will prioritize these 'skills' over traditional features. These new 'skills' will offer more value without complicating navigation or requiring a user to learn new functionality.   

From wireframes/mockups/prototypes → scenario maps. Designers will still need to communicate what a fantastic interaction feels like. However, the nature of the deliverable will change. Moving away from crafting wireframes that outline fixed paths, designers will now develop scenario maps. These maps will articulate criteria for success across a spectrum of open-ended scenarios and adaptable user journeys, fundamentally redefining what it means to define an interaction.  

As we navigate this new paradigm in UX, it's clear that the designer's role will also undergo significant evolution. 

Required capabilities for today's designer

For designers to remain valuable, they must develop three essential capabilities. 

1. AI Strategy

The term "strategy" intimidates many designers, as if it's a complex concept they've never been taught. However, product strategy is just an action plan to solve customer problems.  

This is crucial for companies, especially now, as the AI "arms race" has led them to run frenzied proof of concepts and ship random features that don't solve real customer needs. 

AI is merely a lever to deliver differentiated value to users, and companies need help figuring out which levers they will pull.  

Furthermore, companies should strategically reinvent their UX/CX with AI, not just tackle it as an afterthought. 

Prerequisites:  

• A deep understanding of your customer/user's needs 

• A foundational understanding of GenAI (see Google's free GenAI learning path)  

• An ability to envision the future AI-enhanced experience, bringing to life the power of UX design. (check out Vincent Pirenne's talk here)

2. AI Interaction Design 

In the AI era, interaction design looks entirely different. We're used to designing along linear pathways with predictable outcomes, but now, the introduction of open-ended interactions and flexible paths brings limitless possibilities. 

We've moved from designing "guided tours," which focused on a predetermined route and a uniform experience, to "adventures," which have no clear path and engage every user uniquely.

In this new landscape, like an adventure, our objective is to establish the conditions that will lead to a safe and valuable experience for users. Interaction design will become far more abstract than applying UI components based on standard patterns. It will demand a deep understanding of user expectations as they begin the experience, the ability to pinpoint what success and failure look like, and a theoretical eye for potential risks or vulnerabilities. 

Google's People+AI initiative highlights four critical areas for consideration in designing an AI interaction:  

• acceptable actions 

• unacceptable actions   

• thresholds of uncertainty 

• vulnerabilities  

This framework offers a valuable guide for designing effective AI interactions. 

Google' People+AI Interaction Policy Framework (PAIR)

Prerequisites:  

• An ability to evolve user testing methods for AI functionality   

• An ability to define the criteria surrounding a successful interaction

• An understanding of emerging best practices for interactions (see IBM's GenAI design principles)  

3. Model Design 

Traditionally, there's been a clear division between the land of engineering and design, with the best designers and engineers occasionally crossing over for productive discussions before retreating to their respective kingdoms.  

Now, however, natural language processing allows for direct interaction with large language models, drastically reducing the divide between the two disciplines.  

Instead of writing a line of code, you can simply give the AI instructions with a prompt. What’s interesting about training a model with all the text out there (quite literally) is that it starts to pick up human quirks. For example, even using a smiley face emoji can make it work better.

This presents an opportunity for designers to apply their skills in understanding and empathizing with users directly to AI models.  

In fact, I'm convinced that the average designer's prompt writing ability might outshine that of the average engineer — 🤫 as designers have extensive experience in distilling complex user requirements and clearly communicating needs.  

To do this, designers need to move beyond 'smoke and mirrors' experiments and become comfortable informing the technical execution of an ideal experience.  

Prerequisites:  

• An understanding of Design's value in model development process (see Paz Perez's illustration of the 'model designer')   

• An expertise in writing effective prompts (see Open AI prompt engineering guide)  

• A foundational understanding of how an LLM works (see Stephen Wolfram's summary) 

It's time to dig in 

Let's face it: no one will sit around waiting for designers to become valuable in the AI world. 

But here's the thing—companies need what designers have in their toolkits, especially their knack for solving complex problems without obvious answers. 

When imposter syndrome strikes, remember that nobody has all the answers. This whole AI thing is new for almost everyone, and we're all trying to figure it out together as we go.