Talk · HCDT, UNIST · 17 November 2026

Learning design in the age of AI

Generalist or specialist? A one-hour talk and discussion on how to learn design while AI takes over more of the craft, with my own path from visual design to spatial computing and research as the case study.

  • Expert

  • Generalist

  • Deep generalist

Across: domains. Up: depth.

The talk’s central idea: the expert goes deep in one domain, the generalist spreads across many, and the deep generalist goes deep in one region of them.

About the talk

Event

Seminar at the Department of Human-Centered Design and Technology (HCDT), UNIST

When

17 November 2026, 13:00, in person

Slides

Two parts and extras · PDF, 3.2 MB · free to share with credit

The question

Design is evolving so fast that it’s really difficult to keep up. Should a student go deep in one field, or learn a little of everything, when AI can already do so much of both? The talk is a personal perspective, with some speculation, rather than fact or advice.

Part one · Conceptual

Two conventional ways to learn design, and what AI does to each

A designer is a modern craftsman: given a specification, they make an end product fitted to it, combining art and function. There have been two ways to become one.

  1. 1

    The expert

    Goes deep in one domain, through a design PhD or specialized training such as texture art. Later in a career, experts either specialize in a parallel domain or broaden their skills.

  2. 2

    The generalist

    Learns several kinds of design in a bachelor’s degree, then builds depth and a distinct taste, a design language, over time. Before AI, the generalist usually came out ahead.

  3. 3

    AI against both

    Fine-tuned, domain-specific models reach into the expert’s niche: ChatGPT can now replicate Hayao Miyazaki’s distinctive style on demand. And general tasks are exactly what AI is best at.

  • Generalist

  • AI model 1image generation

  • AI model 2research and text

Defeat in detail: separate AI models can beat a generalist one domain at a time.

“A single AI model can’t cover all the domains a generalist does, but it can cover certain parts with more precision.”

So is design dead? Not really, but its soul is

Craftsmanship is no longer required. In the Industrial Revolution, craftsmen became factory workers; in the age of AI, the designer’s role becomes increasingly managerial, overseeing how the AI system works and prompting the machines. The way through this new, AI-based industrial revolution is the deep generalist.

A blacksmith at work at his anvil in a workshop, in black and white
Rows of workers at a factory assembly line, in black and white
From craftsman to factory worker.

The deep generalist

Rather than expanding to cover a whole domain, the deep generalist expands within one region, at greater depth. There are two ways to do it: across neighbouring domains, which makes extremely proficient people, or across distinct domains, which makes revolutionary ones.

A diagram of the deep generalist: a person linked to a cluster of connected skills inside a circle of domains

What it means for students. Technology changes exponentially, but organizations change only logarithmically (Martec’s law), and large organizations resist change because they are built for reliability and accountability. Smaller, newer players adapt faster. So don’t depend on coursework alone, and accept smaller defeats for bigger wins.

Good reads

Part two · Practical application

Myself as a case study: from visual design to spatial computing, development and research

A generalist bachelor’s in visual design, then a master’s spent going deeper into neighbouring domains, each step measured against experts I could learn from.

  • Bachelor’s. Five years of video editing, photography and visual 3D, with over 70 freelance projects. Then AI image generation went mainstream in 2023, right where my work was focused.

    See my design work

    A 3D render of a weathered sci-fi pilot helmet with a red visorA 3D render of a fighter jet on an aircraft carrier deck at sunsetA 3D render of a dinosaur fossil dig site in red rock, seen from above
  • Master’s, first year. Unreal Engine, Meta VR and SolidWorks. First place for Bangudae XR and Lemmy at CES and the Gwangju Design Biennale, but all three papers rejected.

    See my research

    Presenting Bangudae XR on stage to an audience at the Korea Heritage Digital Contents ContestThe Lemmy exhibit at the Gwangju Design Biennale, with screens and visitors in a white hallVisitors trying a headset at the Lemmy exhibit at the Gwangju Design Biennale
  • Master’s, second year. Swift, iOS AR and Vision Pro. A thesis built on a Blender add-on and an AR app, and three papers from it at KSDS, IASDR and CHI.

    See my research

    The True RoboAnimator panel in Blender, with settings for wheels and calibrationAn iPhone on a stand showing the AR app, with the Lemmy robot placed in a roomReceiving a presenter certificate at IASDR 2025 in Taipei, in front of a slide on component-based actions for elderly care robots
  1. Design and 3D

    Visual and product design, through the bachelor’s.

  2. Development

    Swift, AR apps and Blender tools.

  3. Research

    Papers at KSDS, IASDR and CHI.

  4. AI

    In progress.

The metamorphosis, from 3D visual and product design to development and research, with AI now in progress.

Reflection

The benchmarks were experts around me: Prof. KwanMyung Kim and Ulugbek Ismatullaev for research, and Prof. Jaeyeon Lee in Computer Science and Engineering. Could I be a peer to them? Could I pass CSE classes and graduate with a CSE committee member? None of the goals were fully achieved, but I’m much closer to them than I would have been without this direction.

Danyal Sarfraz with his thesis committee after the MSc defense at UNIST, in front of the title slide
A flowchart of the generalized strategy: from what you are already good at, through distinct or clustered domains and the closest benchmark, to short-term goals, a unique angle and an end strategy
The generalized strategy behind each step: start from what you are already good at, choose distinct or neighbouring domains, study the closest benchmark, and find your own angle.

Extras

Beyond being proficient, and the future of AI-driven design

How to be more than a worker

A deep generalist across distinct domains connects fields to find a unique solution: the Wright brothers and bicycles, Steve Jobs and calligraphy, Larry Page borrowing PageRank from academic citations. You can’t pick those domains in advance; the link is a natural mutation that later looks revolutionary. The best you can do is collect unusual experiences: learn unrelated skills and tools, go on hikes, talk to people.

That matters because AI is weak at discovery. It remixes statistical patterns rather than originating unfamiliar, effective ideas, and one estimate caps the creativity of systems like ChatGPT at 0.25 on a scale of 0 to 1, between amateur and professional.1

The Wright brothers’ first powered flight at Kitty Hawk in 1903, the plane just off the sand

What might happen next

Designers become more managerial: more productive, at the cost of the actual skill, as has already happened in theoretical fields where learning in depth is the expertise. Frank Herbert’s Dune warns that handing our thinking to machines only lets the people who own the machines rule the rest.2 But there is hope at the end: with design opened up to everyone, a design renaissance is possible. Before AI, being a revolutionary like Steve Jobs was optional; now it is essential, the same shift painting went through after photography, writing after the printing press and horses after the Second World War.

An engraving of an early wooden printing press

The slides

Open the whole talk or one part, then step through with the arrows, arrow keys or a swipe.

  • The title slide of Learning Design in the Age of AI
    17 November 2026 · HCDT, UNIST

  • The divider slide for part one, Conceptual
    17 November 2026 · HCDT, UNIST

  • The divider slide for part two, Practical application
    17 November 2026 · HCDT, UNIST

  • The divider slide for the first extra, Beyond being proficient
    17 November 2026 · HCDT, UNIST

  • The divider slide for the second extra, Future of AI-driven design
    17 November 2026 · HCDT, UNIST

Using these slides. You’re welcome to share the slides or use them in teaching and talks. Please credit me as below and link back to this page. Photos and quotes in the slides that come from other sources belong to their owners.

Sarfraz, D. (2026, November 17). Learning Design in the Age of AI: Generalist or Specialist? [Talk]. Department of Human-Centered Design and Technology, UNIST. https://danyalportfolio.com/research/learning-design/