AI Design: Four Trends Redefining the Industry

A Diverse Vision of AI Design

When we talk about "AI design," the term can evoke very different concepts depending on the person. For some, it means using artificial intelligence to generate variations of components in a design system. For others, it involves designing a chat interface or structuring data so that an AI agent can analyze it. Still, others envision defining the behavior of a language model. These different interpretations are reflected in job postings, LinkedIn publications, and discussions at conferences. One person might declare, "We need to define our AI design strategy," and while everyone nods in agreement, each imagines something completely different. Six months later, frustration sets in because the AI design initiative does not meet the initial expectations.

The conversation around AI and design has diversified. What was once a single, albeit vague, topic has split into at least four distinct orientations. Each of these orientations focuses on a different type of design work, fits into various organizational structures, and uses different definitions of what constitutes "good" design. Most teams are prepared for only one of these orientations, confused about which one they are adopting, or trying to juggle all of them without realizing it.

The Car and Engine Metaphor

To better understand these orientations, a useful metaphor is that of the car and the engine. An AI model is comparable to an engine, a powerful and complex technology. Companies like Anthropic and OpenAI sell these engines directly to developers via their APIs. However, most users experience AI through a "car": the dashboard, the steering wheel, the seatbelt — all the parts that transform raw capability into something drivable. Users care little about the internal workings of the engine, as long as it gets them to their destination.

The four modes of AI design work can be mapped to a car: the product, the user, the model, and the infrastructure. Interestingly, this image is itself an example of design with AI: it required numerous iterations in ChatGPT to be produced, likely taking more time than drawing it by hand, albeit in a simplified version.

The Four Meanings of "AI Design"

Laura Costantino sketched an early version of this framework in a LinkedIn post. I took a different direction by reworking one of the categories and developing the rest.

1. Designing with AI

In this orientation, you are the driver. You are the one behind the wheel. This is the orientation that most people think of first. You use AI in your design process: to generate ideas, create prototypes, write text, build mood boards, conduct competitive research, or critique your own work. AI tools are your vehicles, and you use them to go further, faster.

This is where most designers currently find themselves, and there is nothing wrong with that. AI as a tool transforms how design work is done. Entire steps in workflows that used to take hours are now completed in minutes.

The trap is to consider this type of work as the only facet of "AI design." Management declares, "we are integrating AI into our design practice," and what they mean by that is "our designers are now using Claude Code." While this type of AI design is a good start, it overlooks all the other ways AI is redefining design opportunities within the organization.

2. Designing AI Products

You are the manufacturer. You build the car (or you attach AI to a car already in motion). Here, you design the product or interface that provides AI capabilities to users: the dashboard, the steering wheel, and the seatbelt. Everything that transforms a powerful engine into something that people can drive without understanding combustion.

This type of work can be done in two ways:

• AI-native products where the entire product is the AI experience: Claude, Perplexity, Cursor, ChatGPT, or Midjourney. You design the car from the ground up. There is no existing product to integrate. Every design decision shapes how someone experiences the AI capability itself.

• AI features integrated into existing products represent a different design challenge: Notion AI, Smart Compose in Gmail, or an AI summarizer in your meeting tool. You add a new component to a car already on the road — a car with existing patterns, existing navigation, and existing user expectations. The AI must fit into something.

These types of work involve different design challenges with varied obstacles. For AI-native products, the difficulty lies in the fact that there are no established mental models to rely on: you are teaching new paradigms to the user. For AI features in existing products, the challenge is to connect the AI functionality naturally and meaningfully to the rest of the experience, even if no one asked for it.

3. Designing for AI Agents

You are the civil engineer. You design the highways and create the infrastructure that agents need to operate. This may seem like a departure from user experience. Stay with me.

In this type of work, you design content, data, or interactions that AI agents (not humans) will read, analyze, or act upon. You build the infrastructure that autonomous systems navigate. If AI agents are the self-driving cars, you design the roads, traffic signs, and lane markings. AI agents are your users.

This can mean structuring product data so that a purchasing agent can compare options on behalf of a user, writing instructions that an AI assistant will follow, or optimizing content for AI search and discovery rather than (or in addition to) human search and discovery.

Some of this infrastructure may not even be readable by humans. A traffic sign designed for AI-controlled vehicles might encode information in a way that no human driver can analyze — data transmitted via radio or embedded in parts of the spectrum that are not visible. The "user" of this sign is an agent, and the design constraints are entirely different.

Designing for AI agents is important because it involves design decisions. What data is exposed? How is it structured? What can an agent do versus what requires human confirmation? These elements shape the final user experience just as much as an interface; they are just at a different level of abstraction.

4. Designing AI

You are the mechanic. You tune the engine. You help decide what type of engine it is and what it is built for.

In this type of work, you help shape the model: its behavior, evaluation criteria, and principles. You work alongside engineers to define what the AI does, how it reacts, and where it draws the line.

For example, you might define evaluation criteria to test whether the model's outputs meet quality standards, collaborate with engineers to refine the data, or establish the principles that guide adversarial testing — what the model should refuse, where it should express uncertainty, and how it should weigh competing instructions.

So far, the industry has treated these activities as purely technical work, but they are not just that. They involve design decisions. Designers are not the only ones who can make these decisions (an ethicist, a writer, or a psychologist could each contribute), but designers are trained to integrate them into a cohesive experience and test them against user needs. We are just beginning to see job postings that recognize this strength.

Where the Field Stands Currently

These four types of "AI design" work are not mutually exclusive. A single designer may work on two or three of them in a month. Not all organizations need all four either. A company building on a ready-made LLM via an API may only need the first two (designing with AI and designing AI products). The third and fourth orientations become relevant as an organization's relationship with AI becomes more complex — for example, because it is building agent-oriented infrastructure or an internal LLM.

What makes them distinct is that each involves designing for a fundamentally different type of behavior: human behavior with tools, human behavior with AI interfaces, agent behavior within the infrastructure, and the behavior of the model itself. Different behaviors require different types of expertise.

My point is not to say you must choose a single path. It is that each orientation develops a different type of expertise, and that expertise deepens over time. For example, the more time you spend designing AI products, the better you become at that work.

Understanding what type of AI design you are developing expertise in is important because the market is evolving rapidly. Currently, few designers have deep experience across multiple orientations. However, this window will not remain open for long. In a year, many designers will have significant experience in AI. Those who deepen a specific direction now will have a considerable advantage over those who remain largely "adjacent to AI."

Here is where the field stands currently:

• The distribution of designers across the four types of AI design work. Most designers today are designing with AI, a growing minority are designing AI products, and only a small group are designing for AI agents or designing AI itself. The demand for these last two is growing much faster than the supply of designers capable of delivering them.

• Most designers today use AI as a tool in their workflow. An increasing number are designing AI products and...