What Happens to Design After AI?
a16z PodcastFull Title
What Happens to Design After AI?
Summary
This episode discusses how AI is impacting the field of design, exploring whether it will commoditize design or elevate it by automating routine tasks.
The conversation highlights the evolving relationship between design and technology, focusing on the rise of AI-assisted design tools and the future of human creativity in this landscape.
Key Points
- The historical evolution of design shows a shift from scarcity-driven distinction to an era where abundant materials make traditional notions of taste less relevant.
- AI models trained on human output rather than input lack understanding of the "why" behind design decisions, creating a gap in their design capabilities.
- The increasing ease of generating software with AI prompts raises questions about the future value and role of human designers.
- While some predict design will be commoditized by AI, others believe it will become more valuable as AI handles mundane tasks, freeing humans for higher-level creative thinking.
- The development of "auto-design" tools has been a long-held ambition, now becoming a reality with current AI advancements.
- The interplay between design and technology is seen as a convergence, with AI tools supercharging design capabilities through algorithmic creation.
- There's a perceived tension between craft and automation, emphasizing the continued need for human judgment, intent, and viewpoint in design.
- AI models, while powerful, have limitations in design expression, often lacking the nuance and "taste" that human designers bring.
- Tools like Impeccable aim to bridge the gap by incorporating design vocabulary into AI prompts, leading to better results for both designers and engineers.
- The concept of "agentic experience" (AX) is emerging, shifting design focus from visual affordances to designing for interactions with AI agents.
- The value of human taste and judgment is seen as crucial for creating unique and compelling experiences, especially as AI automates more common design patterns.
- The future of design may involve a rise in smaller, opinionated products with niche markets, enabled by more accessible and powerful tools.
- Human trust and accountability are becoming increasingly valuable differentiators in a world of AI-generated content, potentially increasing the worth of human-centric design.
- Leaders with "conviction" – a blend of design, business, and technical sense – are essential for guiding teams towards global maximums rather than local ones.
- The development of well-designed APIs, like those that enabled Photoshop, is critical for AI agents to interact effectively with design systems.
- Effectively communicating the value of design instincts requires leaders to understand and share the "dream" of a future vision.
- The distinction between "cognitive delegation" to AI and "cognitive surrender" is important for maintaining human intent and control in the creative process.
- The "taste" of AI models is complex, representing a multitude of interpretations rather than a singular, human-like judgment.
- Amplifying human taste, rather than replicating it in AI, is presented as a more valuable and interesting problem to solve.
- Taste is viewed as cultural and often emerges from scarcity, leading to more precious and well-designed outcomes.
- The shift to AI-driven design necessitates a focus on human-centric elements that create uniqueness and differentiate products.
Conclusion
AI is poised to automate routine design tasks, making higher-level creative thinking and unique human insight even more valuable.
Tools like Impeccable are emerging to bridge the gap between AI capabilities and human design intent, fostering better collaboration and understanding.
The future of design involves a focus on "agentic experience" (AX) and the amplification of human taste, leading to more specialized and opinionated digital products.
Discussion Topics
- How will the increasing sophistication of AI design tools change the definition of "craft" in design?
- What ethical considerations arise when AI models generate "tasteful" designs without understanding human intent or cultural context?
- As AI handles more of the design process, what new skills and mindsets will designers need to cultivate to stay relevant and innovative?
Key Terms
- LLMs (Large Language Models)
- AI models trained on vast amounts of text data, capable of understanding and generating human-like text.
- UX (User Experience)
- The overall experience a person has when interacting with a product, system, or service.
- AX (Agentic Experience)
- A new paradigm for design focused on interactions with AI agents rather than solely human users.
- Latent Space
- In machine learning, a compressed representation of data, often used to generate new variations of the data.
- API (Application Programming Interface)
- A set of rules and protocols that allows different software applications to communicate with each other.
- CLI (Command Line Interface)
- A text-based interface used to operate software and operating systems.
- Cognitive Delegation
- The act of entrusting tasks or decisions to an AI or system.
- Cognitive Surrender
- The act of passively accepting AI-generated outcomes without critical evaluation or active participation.
- TAM (Total Addressable Market)
- The total market demand for a product or service.
Timeline
The historical context of design originating from royalty and scarcity of materials.
The impact of mobile technology on the importance of design.
The realization of auto-design capabilities, a concept envisioned since the 90s.
The historical pursuit of auto-design, referencing Muriel Cooper's work at MIT.
The historical exploration of automatic design by MIT's AI lab.
The current state of design and technology convergence.
The reality of easier tool building for algorithmic design creation.
The concept of "raising the floor" in mechanical aspects of design through automation.
The tension between craft and automation in design.
The enduring need for human judgment and intent in design.
The example of Steve Jobs as a great editor and decision-maker.
The machine as a tool that works for humans, similar to a camera.
The limitations of current cloud code and agents in design expression.
The expansion of automation to design, requiring restraint.
The engineering perspective versus design's need for restraint.
The architect's role in designing for human experience.
The combination of AI models and feedback loops for design.
The experimentation with GitHub Copilot app and Impeccable.
Raising the floor and ceiling of human design capabilities with AI.
The use of coding agents for drudgery and human animators for craft.
Introduction to Impeccable, its intention, and current state.
Impeccable's origin as an open-source project for personal need.
The development of Impeccable to address AI's lack of design context.
Designers achieving better results with AI due to their specialized vocabulary.
Bringing design vocabulary to AI harnesses as the first iteration of Impeccable.
The unexpected appeal of Impeccable to both designers and engineers.
Code as a substrate for agents and Impeccable as a way to access it.
Impeccable's target audience and its development process.
Impeccable as an agent skill with subcommands and a visual iteration mode.
Impeccable's quality layer preventing "overfitting" of AI models to common styles.
The importance of uniqueness in design and Impeccable's role in achieving it.
The analogy of Impeccable to Kai's Power Tools and PostScript.
The significance of Kai's Power Tools in expanding Photoshop's capabilities.
The open plugin architecture of Adobe Photoshop enabling third-party tools.
Impeccable's similarity to PostScript's impact on design tooling.
The technical and design-mindedness of PostScript's creators.
The need for technical and design-minded individuals in tool development.
Tech's miracle in combining function and form.
Impeccable as a "PostScript moment" for design automation.
The importance of subtraction and multi-dimensional thinking in design.
The collaboration between John and Paul, and the GitHub Copilot app.
The launch of the GitHub Copilot app and its difference from past products.
The discovery of Impeccable and its integration into GitHub.
The growing collaboration between Impeccable and GitHub.
Impeccable's integration as a powerful design skill in GitHub.
Impeccable's availability to mobile and entrepreneur users through GitHub.
Future directions for Impeccable within the GitHub ecosystem.
The GitHub Copilot app's "pick and polish" mode and Impeccable's implementation.
The opt-in nature and clunkiness of current integrations.
Improving integrations through first-party harness integrations like GitHub.
The future of Impeccable and GitHub integration.
The significance of GitHub as a platform for Impeccable's rollout.
The potential impact of designs generated with Impeccable on future models.
The concept of "slop" as a moving target in AI design trends.
The evolution of AI design tells beyond purple gradients.
Impeccable's role in steering design away from overused elements.
The mechanism behind AI design trends and "slop."
Impeccable's function as an "anti-attractor" to prevent homogenization.
Impeccable's use of random seeds and user interviews for uniqueness.
The use of randomness to steer AI into different latent spaces.
The historical origin of purple gradients in AI design (Tailwind's default theme).
The power of default settings in shaping design trends.
The importance of uniqueness in design and its role in capturing attention.
The commoditization and automation of design, leading to a new form of design.
The shift from UX to AX (Agentic Experience).
Designing for agentic affordances rather than just visual ones.
The force multiplication expected from retiring parts of UX.
The role of automated design solutions in freeing humans for higher-level work.
The role of human taste and judgment in AX.
The distinction between the application of craft and the abstraction of craft.
The evolution of design craft from physical to digital primitives.
Designers becoming closer to programmers of interaction.
Paul's work with Impeccable as an example of computational craft.
The future importance of designing for agents (AX) alongside human-centric design.
The focus on high-level work and the last 20% of uniqueness in design.
The absence of a "next Hayao Miyazaki" from AI creative startups.
The focus on "raising the floor" versus "raising the ceiling" in AI creative tools.
The excitement for tools that enable both ends of the creative spectrum.
The shift towards higher-end experiences and audience demands.
The role of craft in differentiating products when AI raises the baseline.
(23:28) The importance of unique and particular craft in design.
(23:56) The emergence of craftsmen who can add a unique spin.
(24:10) The significance of AX and the next level of human experience.
(24:17) The question of whether agents will value aesthetics.
(24:38) The analog for agents valuing aesthetics: speed, clarity, error messages.
(24:56) The concept of "wicking away generic good quality design" to enable ceiling-raising.
(25:08) The echo of the Arts and Crafts movement in the age of machines.
(25:24) The enabling role of tools like Impeccable for craft.
(25:32) The idea that visual design thinkers can crack AX faster due to complex problems.
(25:46) The role of human-shaped minds in designing interfaces for agents.
(25:49) API and CLI design as examples of interfaces agents struggle with.
(26:13) The irony of agents not being good at designing interfaces for themselves.
(26:16) Information architecture and navigation design as areas for skilled designers.
(26:28) Impeccable's design as an example of computational craft for agent and code manipulation.
(26:49) Instinct versus tooling in raising the design ceiling.
(26:55) The combination of instinct and tools in design advancement.
(27:02) The limitations of current AI tooling for specific aspects like motion design.
(27:18) The barren landscape for tools that push the creative envelope.
(27:37) Zuno as an example of a tool evolving from beginner to professional music production.
(28:10) The potential for building tools to accelerate advanced design.
(28:19) The current "barren landscape" for tools that push the envelope in design.
(29:07) The current state of design tooling being "mental" rather than fully realized.
(29:13) The question of whether design instincts or tooling are more critical.
(29:23) Vision models' inability to see and think like human designers due to training data limitations.
(29:33) The latent space of a visual mind versus functional AI tasks.
(29:44) The automation of work that frees those who want to stay in the craft.
(30:04) AI making it easier to produce average work, but unique creativity remaining valuable.
(30:17) The market mechanics of niche products and specialized crafts.
(30:36) The emergence of products with smaller, compelling markets due to previous infeasibility.
(30:53) The concept of a "digital main street" supporting unique, opinionated products.
(31:03) The value of human trust and accountability in bespoke, higher-priced products.
(31:32) A prediction from 2010 about software returning to a cottage industry.
(31:49) The current movement towards more unique and opinionated applications.
(32:03) Learning from non-computer-based designers and artists about handmade craft.
(32:17) The rise of computer programming for artists and designers, now more accessible.
(32:45) The current exciting but confusing time in design and technology.
(32:49) Impeccable's unique approach in providing high quality and thought-based design.
(33:01) The opportunity for Impeccable to shape both design and engineering minds.
(33:10) Impeccable being used as a default tool, fostering better communication between designers and engineers.
(33:39) The teaching moment provided by Impeccable in developing design and engineering language.
(34:00) The concept of cognitive delegation versus cognitive surrender.
(34:44) Cognitive delegation, using Google Maps as an example.
(35:06) Cognitive surrender to AI, where the AI dictates choices.
(35:38) The problem of students surrendering to LLM-generated plans.
(35:58) The importance of cognitive delegation while preserving human intent and viewpoint.
(36:15) Impeccable as a tool for active collaboration between humans and AI.
(36:17) The current focus on automation, goals, loops, and multi-agent orchestration.
(36:24) The unasked question: where does this lead, and what is the role of human intent and taste?
(36:37) The complexity of AI taste, with models having millions of definitions.
(36:49) LLMs trained on output, not input, lacking understanding of design decision rationale.
(37:19) AI taste as an approximation for a specific audience and time.
(37:26) The question of whose taste AI models are approximating.
(37:35) Taste as rare, human, and difficult to replicate.
(37:48) AI designing tasteful things, but human viewpoint extending beyond that.
(38:03) Amplifying human taste as a more interesting problem than replicating it.
(38:15) Substituting "humanity" for "taste" and its powerful implication.
(38:23) Whether taste is innate or cultivated.
(38:28) Taste as cultural, with older cultures often exhibiting "higher taste."
(38:43) Examples of cultural taste differences (Denmark vs. US furniture).
(39:03) Scarcity of material driving preciousness and good design.
(39:18) Design's origin in royalty and the desire for distinction through scarce materials.
(39:36) The irrelevance of taste when materials are abundant.
(39:50) The quote: "taste is a whisper and velocity is a megaphone" in decision-making rooms.
(40:06) Engineering incentives, not engineering itself, rewarding the measurable over the meaningful.
(40:16) The leader's role in having and recognizing an idea, and betting on it.
(40:31) Impeccable's API design and its similarity to Macintosh Quickdraw's impact on Photoshop.
(40:44) The technical aspect of Quickdraw's API that enabled Photoshop.
(41:15) The "unlock" in Quickdraw's API surface: regions based on pixel shapes and optimized routines.
(41:45) Great APIs encompassing shape and performance.
(41:55) Advice for design engineers on communicating instinct vs. deadlines.
(42:11) The difficulty of advocating for design instincts against deadlines.
(42:22) The importance of aligning leaders with the envisioned future through emotional connection.
(43:09) Bringing leaders along to share the same emotional journey and vision.
(43:36) The concept of "conviction" in leadership, combining design, business, and technical sense.
(43:44) Conviction aimed at the global maximum, not the local maximum.
(44:08) Steve Jobs's method of testing individuals' conviction.
(44:39) The two-way nature of conviction generation and feedback (LLM as judge).
(45:00) The need for both generation and feedback of conviction.
(45:15) The future impact of products created with GitHub and Impeccable.
(45:52) The potential for AI-assisted design to be more thoughtful and in the upper 10%.
(46:14) John's realization of how many computational designers/engineers grew up at GitHub Design.
(46:34) Paul's work representing a design history-infused approach for engineers and designers.
(46:58) Paul's excitement for the future of craft and the experiences that will be built.
(47:09) The contrast between a future of new craft and a bleak future of sameness.
(47:30) Cheers to the new era of craft.
(47:39) A reminder about the podcast's content being for informational purposes.
Episode Details
- Podcast
- a16z Podcast
- Episode
- What Happens to Design After AI?
- Official Link
- https://a16z.com/podcasts/a16z-podcast/
- Published
- June 24, 2026