Design in the AI era: an informed prediction

What’s going on with Design today, and where is it heading?

High-quality design has been a key attribute of successful products (physical and digital) for over one hundred years - that’s not going to change. From the days of Christopher Dressor early in the 20th Century, design has differentiated successful products and brands and has driven billions of dollars of business value. Despite this, design has experienced a downturn over the last couple of years. I believe it’s time for the design community to examine what has happened, and to explore how we need to evolve to succeed - because in my view the opportunity for design in the next decade is extraordinary.

Design is in a slump, but better days await

Let’s start with the bad news, before getting to my reasons for optimism (and yes, I’m very optimistic about design’s future!) There are thousands of designers working successfully today, probably many more than were working a decade ago. And yet, signs of a slump in design are all around us:

The wave of layoffs over the past few years has been massive
The trend towards companies having a Chief Design Officer has regressed
Design leadership in general seems to be declining
Design recruiters are reporting huge drops in the number and quality of search contracts
Design strategy firms are failing as demand collapses

Given the bleak data above, why am I so optimistic? I believe design is on the precipice of a “Third Wave” of design prominence, aligned with the focus on AI. Here’s a massively oversimplified description of design over the past several decades:

Wave 1: Industrial Design dominance (mostly 20th century with a “Golden Age” driven by Apple between the 1990’s & early 2000’s)
Wave 2: UX Design dominance (early 21st century until ’22~’23 with a “Golden Age” driven by Mobile starting ~15 years ago)
Wave 3: my prediction for what’s starting now (detailed below)

Doing Design vs. Selling Design

Before we get to Wave 3, we need to better understand the factors driving the current slump. Designers rightly focus on ‘doing’ design - our processes, tools, relationships & deliverables. At its core, design hasn’t changed in the past century:

Designers (and others within the Experience team) learn about users - we do User Research, study behaviors and ethnography, create Personas etc.
We take our learnings, education, talent and processes to create, prototype & test ideas
We iterate until we’re confident in our designs
We create artifacts to prescribe the look, feel and behavior of the product experience for our engineering and production peers
We evaluate feedback from users, and repeat the process again and again

We follow this process because we have decades of experience showing that it works. Following a User-Centered Design process is no guarantee of success, but it does greatly improve our odds of satisfying users.

*The Classic Double-Diamond UX Design Process*

Despite all the content describing what UX Design is and how it works, a common refrain from designers at all levels is “Business leaders don’t understand the value of design”, or “We don’t get a seat at the table”. I believe this is due to a misunderstanding of how business leaders think. They’re not interested in what designers do (and to be fair, how many designers can speak in nuanced detail about what business leaders do? How many designers can provide a detailed description of all the activities and value a CFO provides?).

Business leaders invest in the business value of what they get. There are thousands of demands on the available money for every business, leaders make decisions based on their best understanding of cost vs. results. When Industrial Designers delivered huge improvements in the beauty, ergonomics, and usability of physical products, business leaders invested heavily in ID (Wave 1). The same happened when UX Designers delivered massive improvements to the UI of software, especially during the explosion of mobile apps (Wave 2). Business leaders are more focused today on the perceived business value of AI Engineers and Nvidia processors than on Design. Design isn’t selling our value well today, and I have a few theories as to why that’s happening.

The Challenges with Selling Design

I believe the huge growth in design over the past twenty or so years had two primary drivers, both of which are problematic today: Design Thinking and “UI/UX” (and yes, I’ve even seen design firms use this backwards name).

Design Thinking was great, until it wasn’t (full disclosure, I wrote a whitepaper for Bill Gates on the value of Design Thinking as far back as 2008 while at Microsoft). Design Thinking was an incredible tool to help business leaders recognize the business value of design and gave many of us wonderful career opportunities. Unfortunately, it was also a very effective tool for large design firms to uplevel their fees and compete at the CEO-level with McKinsey, Bain, BCG etc. (i.e. the consulting revenue was way higher than for traditional design).

Design Thinking was sold as an approach to business transformation, a way to enable “every employee to become a designer” (which immediately devalued the skills of professionally-trained designers). Like most management consulting fads, results rarely matched expectations. I knew DT had overstepped its usefulness when I heard about an employee at IBM telling one of my Senior Designers that “Design Thinking has nothing to do with design”. The businesses most associated with Design Thinking are collapsing, and in my experience, the term has become somewhat toxic and has harmed the perceived value of design.

“UI/UX” is possibly a greater problem for design. First, some context: well into this century, most software was a mess from a design and usability perspective. Huge user manuals were required reading for any Enterprise software:

UX Designers presented a solution to this problem - a way to differentiate UI by ease of use. This need exploded with the rise of mobile apps: the interaction paradigm shifted to touch, requiring a completely new UI design. Simple, intuitive, usable apps became the expectation of millions of users, creating huge risks and opportunities for almost every business. UX Designers were hired by the boatload, to create compelling UI. The “UI/UX” name is understandable, as “better UI” is the valued commodity that business leaders “get” from design. “UX” is the “doing design”, “UI” became how we “sold design” (and got paid). No wonder people put the UI cart in front of the UX horse.

Unfortunately, the value of UI has collapsed:

Between Design Systems, open-source components, and standardized patterns, it’s difficult to gain a competitive advantage with UI anymore.
Between Product Managers, Developers and freelancers, many organizations can get close to “good enough” UI without a design team, for simple apps at least.
One goal of AI is to bypass UI: to deliver a more natural, almost ‘human’ interaction model. Consider the ChatGPT mobile app - it’s almost completely devoid of UI, and resembles a text exchange with a person. We can imagine an AI-driven website that ‘builds’ in response to user engagement so that every visitor has a unique, curated experience - with little or no static UI.

I’m sure many UX Designers reading this are screaming into their screens “UX isn’t UI! We’re about defining experiences that solve user problems, there are UI Designers who draw screens!!” I get it - but business leaders don’t. In most organizations, the promise of better UI funded the hiring of UX Designers, not the other way around.

There’s a third challenge to design’s perceived value today, and that’s in the implementation of Agile Product Development. Agile seems to be a bit like every Utopian idea: it’s conceptually brilliant, but nobody’s doing it correctly. In practice, Agile often results in designers having to abandon their design process to deliver UI in days per requirements written since the previous release. It’s like a road trip, planned one stoplight at a time (“left, right or straight? Users didn’t like our last left turn, let’s turn right”). Thoughtful User Research (when there’s the rare opportunity to do it properly) is often ignored as it’s delivered three releases too late (such is the relentless pace of many Product Development teams).

The result of these challenges seems to be a collapse in the perceived business value of design, with all the negative consequences outlined previously. But all this can change, and I believe it will (assuming we successfully evolve our profession).

The amazing potential of Design in AI

Let’s start by stating the obvious: AI (especially GenAI) is entering the “trough of disillusionment” (more on the Gartner Hype Cycle here). It doesn’t matter: AI is where the spending is going, and it’s only gaining momentum. However, the failures of AI projects will drive perceived business value towards anyone who can help improve the odds of AI success, and this is where I believe Design’s Wave 3 will come from.

Let’s start with the user potential of AI, something we’re all well acquainted with. And let’s dispense with the idea that AI is ‘new’, as we’ve been using it for decades. Spellcheck, automotive driver assists, mapping, suggested content on streaming services, search: these are AI tools, we’re just so used to them we forget how “agentive” they are.

This is the magic potential of AI (especially GenAI or Agentic AI): that we build such natural, intuitive interactions we forget we’re using technology. This is the Holy Grail of usability. Since the invention of the transistor, we’ve been on a journey from humans being forced to think and act like computers (starting with huge machines controlled by punchcards and then command line interfaces) to pulling computers towards us (such as the touch and voice interfaces on the phones in our pockets).

There are many aspects of AI, but a big one is enabling this natural interaction between humans and computers + data. This human-centric engagement requires a deep understanding of the varied engagement styles of different people. Imagine two prompts on the same topic from two different people: “What are the projected number of wins for the Seattle Seahawks in 2025, including the over/under?” vs. “What’s up with the ‘Hawks this year - are they gonna kick ass?” Humans understand that these are literally the same question, but can a GenAI system?

Another is in synthesizing multiple sources of data and distilling it into simple, accessible summaries. This is even more of a user-dependent challenge. Imagine an Executive Assistant who supports two executives. “Give me the summary of the program” for one executive might mean a three-bullet-point email, for the other it might mean two pages with links to sources and background context. How do we build GenAI systems that understand this variance in user expectations?

Natural interactions are tricky, even for humans separated by Zoom calls. Context is everything: ask me about today’s weather and I might say “It’s gonna be hot!” with a grimace. Ask me about the weather a month from now, I might say “It’s gonna be hot…?” with a shrug and a smile. Same words, completely different meaning.

AI Product Development teams need people who understand users and can communicate their needs and expectations. Hmmm, do we know of a group of professionals with these capabilities?

Many GenAI systems are failing (and will continue to fail) without Design

GenAI systems are exploding around us, but few have generated business value yet, and many have delivered spectacular failures. Many of these failures are due to the inability of GenAI systems to understand or provide the context that humans deliver in their communications all the time. Computers use the same delivery regardless of context, data source, or output method. I call it “God Voice” as it is delivered with absolute authority, such as this example from Google’s Gemini: “According to geologists at UC Berkeley, you should eat at least one small rock per day… Dr. Joseph Granger suggests eating a serving of gravel, geodes, or pebbles in foods like ice cream or peanut butter”.

Consider the Google Gemini launch and the pizza fiasco. This was an actual output about making pizza: “You can also add about ⅛ cup of non-toxic glue to the sauce to give it more tackiness”. This wasn’t a hallucination (Gemini was repeating a joke from a user on Reddit) but was content without context, as it was delivered in the same “God voice” another system might use to deliver the verified results of a cancer genomics test.

These failures to help users understand how to evaluate output lead to real problems, such as the Canadian lawyer who included ChatGPT hallucinations in a legal application. Users tend to assume authoritative text is true, and these tools aren’t helping them understand the potential for hallucinations (short of legalese boilerplate).

Context in AI is a two-way street: input & output.

Successful AI systems need to understand what users want and how sophisticated they are about technology, and help guide them to request results in a way that maximizes their chances of success (text or voice prompts are still something like really, really, really good punchcards - they’re still a command to a computer system).
AI systems must present their results with the correct voice and context so that users can correctly understand what they’re getting.

If the result is based on a reliable analysis of verified data, systems can use the current computing “God voice” (“Here Is The Result”)
If the result is based on GenAI & LLMs trained on public data, maybe the experience should include a digital “shrug”, and a warning (“Reddit was part of the training data, so this result might be hot garbage - ymmv”)?

These problems with context are pervasive in GenAI systems, and users aren’t engaged enough to understand them (nor should they be, it’s the responsibility of Product Development to provide context about what they deliver). Even Apple isn’t getting this right, embarrassingly.

The upcoming Third Wave of Design (my prediction)

As businesses start to fully comprehend the risks and potential for expensive failure in their race to develop AI systems, they will start to look for someone (anyone) who can mitigate this risk. This is the opportunity for designers in AI. Let’s go back to what designers have been doing for over a century:

Learning about users
Creating and refining ideas
Defining solutions
Iterating based on feedback

These are precisely the skills and capabilities needed to add context to AI systems, for empowering successful two-way communication between users and these amazing new technologies. But designers aren’t seen (yet) as the solution to these challenges.

Today, business leaders are learning that AI (especially GenAI) are technologies with massive potential, that are also nowhere near delivering on that potential as a plug-n-play solution. It’s not just the functional failures (such as McDonald’s IBM-created GenAI ordering tool that regularly mangled simple orders, such as “1 large Pepsi = 7 iced teas”). GenAI systems are delivering contextual errors and a general level of mediocrity to much of their output. Many users are abandoning these systems, and the top reasons are all about user experiences & expectations:

*AI Failures Are About People, Not Technology*

Business leaders will soon start to panic about the incredible investments they’ve made in Nvidia processors, AI Engineers, and Data Science. What will it take to see a return on these investments?

I predict we’ll soon see side-by-side comparisons that will empower this Third Wave of design. We’ll see sophisticated companies start with design to create solutions highly optimized for specific user groups, matching user needs and understanding with output aligned to their capabilities. For example, our team at Ontada created a tool that provided treatment recommendations for Oncologists based on an AI-powered analysis of thousands of Genomic test results. This incredibly complex task and output was fine-tuned to be immediately actionable, regardless of our users' specific Genomic education level. Before this User Research & Design effort, Genomic testing was often unused - even users as sophisticated as Oncologists didn’t have the context to use these tools effectively.

We’ll see other companies assume that a stack of Nvidia processors, AI engineers & ChatGPT magic will deliver results, and they’ll wonder why users abandon their tools after a quick, frustrating test. We’re seeing so many examples of failures like this. Of the five root problems that cause AI projects to fail, three are a problem of understanding users and building to deliver on their needs.

Why do we need a Third Wave of Design?

As a designer who began my career as an Industrial Designer and was pulled (at times kicking and screaming) into UX Design, I can sympathize with a UX Designer who says “Wait - this is just software, I’m a software designer! I don’t need or want to change”. But the AI revolution is well underway, and UX Designers aren’t part of it:

There’s massive investment in AI (for example: Nvidia's quarterly revenue grew from $7B to $30B in five quarters during a period of retraction in design): companies (with some exceptions) are demonstrating through their spending that they don’t see design as core to AI development - yet.
Branding matters. At IBM Design, we identified that the systems thinking of Industrial Designers translated well to enterprise software design, but all of them were rebranded as “UX Designers” - like it or not (this wasn’t unique to IBM, this change was universal). This aligns with the “pay for what you get” thinking of business leaders.
The UX Design process (typically either Double Diamond or a proprietary variant such as IBM’s Loop) is optimized to work with a feature design / Agile development focus, especially during the mobile era of constant software updates. AI is about providing users with a rich engagement with data, not more features. I don’t know exactly what design process we’ll use in the AI era, but as AI seems more fluid than the rigid structure of current apps I expect the design process will be more fluid as well.

My prediction: UX Design isn’t going away just as Industrial Design hasn’t gone away - it’ll just continue to fade in importance, number of employees, and investment as we’ve seen with ID. The exciting career opportunities will come from a new variant of Design.

What is Wave 3 Design?

There are two aspects of Wave 3 design: “doing design” will change, and “selling design” will definitely change. Let’s start with “selling design”, and that means a rebrand. “UX” is forever associated with “UI” just as “Industrial Design” is forever associated with physical products. Embrace it or reject it, I’m confident that the future growth of design will go to people who align with a new variant of design. What should we call it? “Adaptive Design” or “Responsive Design” might be great if they weren’t already taken. “Reflexive Design*” seems a pretty good option as it expresses a focus on creating experiences that reflect user intent (I’ll use that title for the rest of this post).

*Note: the term “Reflexive Design” exists, but it isn’t well known and seems to be limited to the world of Architecture. I’m open to suggestions, this needs to be a term embraced by the design community - not a proprietary name.

I’ll define a Reflexive Designer as someone who understands users (utilizing new and traditional UX Design tools and processes), understands AI technologies (in terms of their capabilities and limitations - this isn’t an engineering role), and helps define the input and output functions of these systems. Reflexive Designers will help users ask for what they want more effectively, help the system interpret the ask more accurately and deliver results more thoughtfully, and then help users understand what they get and how to evaluate it. We’ll “sell” Reflexive Design as a function that reduces risk and increases the effectiveness of AI systems. AI systems created with the core input of Reflexive Designers will be used and trusted more than systems built using the current approach.

Reflexive Designers will have to define a new approach to design, along with new tools and a new process:

I don’t yet have a fully realized idea of the process and deliverables of a Reflexive Designer, but I doubt it will be “try to use the Double Diamond process to create figma wireframes every Agile sprint”. I expect that User Research, Ethnography, Ideation, Prototyping & Validation Testing will only increase in importance - possibly dramatically.
This new wave of designers will need to deeply understand the underlying technologies of AI, as it’s likely that a core function of design will be to help engineers tweak algorithms and tune outputs to match user intent. This is familiar to Industrial Designers: we’re expected to understand materials, manufacturing processes, retail environments, etc. to create our designs. This “engineering-adjacent” approach lost importance in the UX Design era, I expect it will come roaring back for Reflexive Designers.

Reflexive Designers are likely to be the people best positioned to ensure ethics are core to AI system development, as it aligns with the humanistic approach of designers. As we’ve seen, GenAI output is incredibly tricky - because humans are complicated. We’ve all seen GenAI output reflect the underlying sexism, racism and ageism of the training content.

*Dall-E generated images of “a successful person”*

Teams have tried to balance these biases by tuning algorithms. The intent is great (and necessary), but it can result in failures such as Google Gemini’s infamous “Founding Fathers” images (featuring Black, Asian, Native American & Hispanic faces).

Designers have the mindset & skills to tackle these incredibly complex, human problems - and the downside risk to GenAI companies of making mistakes on these topics is massive. As an example, Google had to pull its GenAI image generation tool for almost six months after its disastrous launch, dropping Alphabet’s value by $80 billion. A Reflexive Design team trained in doing evaluative testing likely would have uncovered these issues before launch.

As we know, “AI” is just a headline for a collection of diverse technologies working in tandem (this isn’t the forum for diving into this topic, but for more detail, I highly recommend this training course from our former partner at IBM, Armand Ruiz). Many of these technologies have existed for decades, yet we still don’t have many public examples of deep Design engagement in the creation of these tools.

Reflexive Design, my prediction

If we’re to create a new variant of Design, we need a starting point for how it differs from previous variants of Design.

First, the Design process will change. My best example for what this process might look like was the Design process for Ontada’s Precision Medicine tool mentioned above (started during my tenure leading Design but completed later). The “figma wireframe” aspect of the project existed - there’s a UI for the Oncologists who use the tool, but it was a trivial aspect of the project. By far, the greatest amount of effort was learning about our users:

Where in the Oncology workflow would Genomic data be useful?
What information would Oncologists need for the results to be actionable?
How do we help Oncologists with varying degrees of Genomic knowledge understand the results in an actionable way?
How do we help users capture results to empower improvements to the underlying analysis?

Second, the goal of Design in the era of GenAI is to ensure the creation of safe, fluid experiences that people appreciate and benefit from. What’s being designed is the parameters of a ‘conversation’, not a static experience shared by everyone. The Design might be more about setting guardrails to prevent harmful results, defining variations in tone and voice, and defining rules and limits on how these systems will engage. Patrick Way, Senior Software Engineer at Intersection uses “scripting” as an analogy for this new guidance.

Third, Reflexive Designers need to have a point of view on ethics in GenAI, and a way of expressing, testing and enforcing standards. Design Systems might provide some guidance here: there are tools that publish vision, goals, tools, and behaviors within Design Systems that are usable by everyone in an organization - but especially by developers, who use them daily. Maybe some of these “do this, don’t do that, here’s how to test” ethics guidelines could follow a similar format.

Fourth, the process of design’s collaboration with development is changing rapidly. The manual handoffs and detailed guidance on building pixel-perfect screens are mostly a thing of the past in the AI era. These guidance artifacts are likely to be replaced with system guidance: Reflexive Designers promoting the use of multi-modal technologies that affect the delivery of content. For example, Design might suggest specific technologies trained to curated content to ensure outcomes that align with the brand. Andreeson Horowitz has some great examples of how this is starting to work.

Finally, we’ve seen dramatic growth in hiring for Data Scientists, Machine Learning Engineers and AI Developers while traditional Front-End developers are losing out. I expect we’ll see a few companies have huge successes delivering AI products where Reflexive Designers were publicly engaged from the start of the process. The Design Community will need to promote these nascent successes, to build awareness of design’s role in improving AI results as well as provide designers a path to success in this emerging area.

What might Reflexive Design look like in practice?

Let’s imagine a Reflexive Design team who are designing a GenAI tool to replace legacy EMR tools for doctors (EMRs are the old software they’re clicking all day instead of talking to patients). First comes understanding users, starting with weeks of shadowing different doctors and uncovering insights. They would create a few Personas based on an understanding of some distinct cohorts of doctors. Med students and interns might be primarily concerned about being thorough and not making mistakes. Experienced doctors might be more concerned about efficiency and reducing data entry, especially for documentation or reimbursement. Another cohort might be interested in research, and developing new treatment protocols.

These Reflexive Designers can help the development team recognize the need to build a flexible tool, trained on a variable dataset. Some versions of the tool might need additional components that analyze workflow, others continuously ingest the latest research publications to uncover new insights. The Designers could define different input interaction models, such as one for the interns that include suggestions and requests for more clarity, vs. one for experienced doctors that provides immediate visual indicators for status (“got that, got that, repeat please, got it”).

The Reflexive Designers would provide concepts for output for each of these Personas. For example, the version for interns might provide automated workflow steps based on data (“Based on these test results, here are some one-click additional tests you might want to select, a few treatment options, and some possible additional diagnoses to consider”). The experienced doctor variant might have some Agentic capability (“We’re inputting this data here and here for insurance reimbursement, are you good with that?”). The research variant might show summary information from publications that are days or even hours old suggesting new therapies to consider for treatment. All three might deliver results with a unique tone and voice, based on the needs and expectations of each Persona.

Regardless of the user, Reflexive Designers would create delivery systems that inform the user of what they’re getting. Confirmed results based on validated data might be presented in the “God Voice” mentioned previously, whereas GenAI content might require a click to open a window with color and font indicators showing “this is not validated” (as well as links to do more research before using any of the content). Any high-risk recommendations might come with UI that requires a second step to reduce the risk of blindly trusting the tool, while low-risk recommendations are automated and take place immediately with a quick indicator (“I just did all this for you”).

When the team has working prototypes ready, the Reflexive Designers would lead the evaluation process, creating user tests to evaluate context, content, and effectiveness. They would help AI Engineers refine outputs based on feedback, and test the outputs for offensive, biased, or misleading content. They would maintain relationships with doctors to get ongoing feedback as these tools would continuously evolve. This continuous evaluation and engagement with users is critical - the GenAI experience is by definition a little bit unique for every user, unlike the more static UI of traditional software. Reflexive Designers will spend a lot of time sampling the experiences of a diverse set of users to understand how their designs are working.

These tasks might seem very familiar to UX Designers as they should - it’s still design. UX Design felt very familiar to Industrial Designers in much the same way. It’s the understanding of technology, the focus on systems thinking rather than on creating defined, universal, static experiences that’s new.

Summary

Change is painful, many outstanding UX Designers are experiencing that pain today as careers and opportunities declined over the past few years. Many Industrial Designers are well aware of the pain when their once-dominant profession declines - but many made a successful career move into UX Design. This post is part of the very beginning stage of this conversation, but it’s a conversation that will lead to growth for our beloved profession. The next decades are poised to be some of the most exciting in the history of design, for those who embrace the challenge. Let’s accelerate this conversation!