Key Moments
Teaching AI the Language of Design | Deep Dives with a16z
a16zWant to know something specific about what's covered?
We've already dissected every moment. Ask and we will deliver (with timestamps).
Key Moments
AI is poised to automate routine design tasks, but human taste, judgment, and unique vision become more valuable, not less.
Key Insights
Designers using AI tools like Claude achieve better results than engineers due to their distinct vocabulary, focusing on terms like 'vertical rhythm' and 'negative space'.
AI models are trained on the output of human creativity, not the input, meaning they understand 'what' was created but not 'why' a design decision was made.
Impeccable, an open-source project, aims to bridge the gap by providing designers and engineers with a specialized vocabulary and quality layer to improve AI-generated designs.
The 'slop' in AI design, such as default purple gradients or beige backgrounds, is a moving target that evolves as models and users adapt, requiring constant rule updates.
The future of design may shift from User Experience (UX) to Agentic Experience (AX), focusing on designing for AI agents rather than just humans, with an emphasis on non-visual affordances like API and CLI design.
Human trust and accountability are key differentiators for bespoke, high-value designs, making human 'smell' and conviction more critical in a commoditized design landscape.
Human-centric design language elevates AI output
The conversation highlights how designers, despite using the same AI models as engineers, consistently achieve superior results due to their specialized vocabulary. Terms like 'vertical rhythm,' 'negative space,' 'bolder,' or 'quieter' are essential for communicating nuanced design intent to AI, a lexicon often absent in engineering prompts. This distinct language allows designers to steer AI more effectively, distinguishing their output from what engineers might produce. Impeccable, an open-source project developed by Paul Bakaus, specifically addresses this by embedding design-specific language into AI agent harnesses. This not only helps designers but also bridges the gap for engineers and product managers who are increasingly asked to collaborate more closely with design tools and AI.
AI's understanding of design is limited to output, not intent
A critical point raised is that current Large Language Models (LLMs) are trained on the vast output of human creativity but lack understanding of the underlying 'input' or the reasoning behind design decisions. This means AI can replicate styles and patterns it has seen but cannot grasp the 'why' of a design choice. John Maeda likens this to having millions of definitions of taste without understanding the human experience or scarcity that informed it. This distinction is crucial because while AI can automate many aspects of design, it cannot replicate the human intuition, judgment, and unique perspective that drive true innovation. The implication is that the value of human designers will shift towards these less automatable, higher-level cognitive functions.
Automation raises the floor, but human taste defines the ceiling
AI's primary impact on design is seen as 'raising the floor' by automating mechanical and repetitive tasks, making good design more accessible. However, the 'ceiling'—the pinnacle of creative expression, uniqueness, and sophisticated problem-solving—remains firmly in the human domain. John Maeda draws parallels to historical shifts, like the arts and crafts movement responding to industrialization, suggesting that as machines take over drudgery, humans will ascend to higher levels of craft and creativity. Tools like Impeccable aim to empower human creators, moving beyond simple filters to enable nuanced iteration and stylistic control, ensuring that AI serves as a co-pilot rather than a replacement for human ingenuity. This dynamic creates an exciting environment for novel experiences, both visually and in agentic interactions.
The 'slop' of AI design is a moving target
The discussion touches upon the phenomenon of AI 'slop'—predictable, often uninspired design patterns that emerge from AI models. Initially, this manifested as purple gradients and italicized text; now, it might be beige backgrounds or specific font choices. Paul Bakaus explains that AI models, when steered away from one overused element, simply default to the next 'best' option in their latent space. This 'algorithmic Uniqlo' effect means the aesthetic of 'slop' constantly shifts. Tools like Impeccable combat this by developing 'anti-attractors' and introducing deliberate randomness based on user input and context, aiming to foster uniqueness rather than conformity. This is crucial for creating standout designs, especially for landing pages or branding, where differentiation is key.
The future of design: from UX to AX and computational craft
The conversation pivots towards the concept of Agentic Experience (AX), a shift suggested by John Maeda, where design increasingly focuses on interactions with AI agents rather than solely with humans. This involves designing for non-visual affordances such as API design, command-line interfaces, and robot.txt files, areas where AI agents currently struggle. Paul Bakaus and John Maeda agree that while AI can automate 80% of tasks, the remaining 20%—requiring human insight, craft, and conviction—will become more valuable. They also discuss computational craft, where tools like Impeccable, with their well-designed APIs and ability to work seamlessly with agents, enable designers and engineers to build complex, unique experiences that push creative boundaries and cater to smaller, yet compelling, market niches.
Human taste, trust, and conviction remain paramount
In an era of increasing AI-driven automation, human taste, trust, and accountability are identified as increasingly valuable differentiators. Paul Bakaus argues that AI models can approximate taste based on vast datasets but lack the human input and intentionality that define it. He emphasizes amplifying human taste over replicating AI taste. John Maeda adds that taste is culturally shaped by factors like scarcity and maturity, suggesting that in an age of abundant materials, distinctiveness becomes even more crucial. The concept of 'conviction' is highlighted as a leader's ability to bet on unique ideas that aim for 'global maximums' rather than local ones, a trait seen in figures like Steve Jobs. Ultimately, for brands where trust is paramount, the human ability to sense and ensure quality may become more prized than ever.
The challenge of instinct versus deadline and the developer's role
Communicating the value of design instinct against hard deadlines is a persistent challenge. Paul Bakaus acknowledges past failures but suggests strategies involve guiding leaders to a similar future vision, framing design decisions in terms of desired outcomes. John Maeda uses the analogy of Steve Jobs testing employees to gauge their conviction – an employee must double down on their ideas, even under scrutiny, to earn trust. This mutual conviction, from leader and employee, drives innovation. The discussion also touches on the importance of well-designed APIs, citing the Macintosh's QuickDraw graphics library as a foundational unlock for Photoshop's success, demonstrating how computational craft in API design can profoundly impact downstream tools and applications.
Nurturing unique experiences and the future of craft
The overarching sentiment is excitement for a future where AI democratizes design, enabling creators to focus on higher-level conceptual work and unique expression. Paul Bakaus aims to build tools that ensure not everything looks the same, fostering a new era of craft. John Maeda anticipates that while the mass market will be served by automated, above-average work, a smaller segment will always value and pay for bespoke, human-crafted excellence, much like letterpress printing today. The collaboration between Impeccable and GitHub signifies a step towards empowering both designers and engineers to reach new creative potentials, ensuring that the human element—taste, intention, and vision—continues to define the cutting edge of design.
Mentioned in This Episode
●Products
●Software & Apps
●Companies
●Organizations
●People Referenced
Navigating AI in Design: Dos and Don'ts
Practical takeaways from this episode
Do This
Avoid This
Common Questions
Designers often use specific vocabulary related to aesthetics, rhythm, and visual elements (like 'vertical rhythm' or 'negative space') which leads to better results when prompting AI models compared to the more functional language engineers might use.
Topics
Mentioned in this video
An AI model mentioned as being used by designers for better results than engineers, due to their language. Also referenced in the context of AI 'slop' like purple gradients.
Mentioned as a font Muriel Cooper imagined people would use to look at things on a screen, predicting the desktop publishing revolution.
Mentioned as an AI model that Paul's tool Impeccable aims to improve upon, as it previously did a terrible job when asked to make design better without understanding the context.
An open-source agent skill created by Paul for himself to push AI engineering. It brings design vocabulary to the agent harness and aims to stop AI over-fitting.
An application that blew up for Photoshop in the 1980s by enabling algorithmic effects and filters through a plugin architecture, serving as an inspiration for Impeccable's approach.
An application from the 1980s that allowed users to manipulate pictures and photographs. Its plugin architecture, enabled by Kai's Power Tools, is compared to Impeccable's potential.
A critical technology from Adobe that properly encoded primitives for visual graphic design, requiring technical and design-minded individuals. Impeccable is compared to its 'moment'.
A system created by Donald Knuth that combined function and form, serving as an example of how technical innovation can be deeply integrated with design principles.
A CSS framework whose default theme, with purple gradients, is cited as a possible reason for the prevalence of purple gradients in early AI-generated designs.
A JavaScript UI library for which Paul created a theme framework, and whose default theme was orange, leading to him coloring the web orange unintentionally.
An AI model that samples frames from video, doing slightly more for temporal resolution than older models, but still limited.
An example of a tool that started for beginners (like vibe sound designers) but evolved into real music production, mirroring how tools can push the envelope for craftsmen.
Apple's graphics library on the Macintosh, whose API design, based on regions and pixel shapes, enabled optimizations that made Photoshop possible and were difficult to replicate on other platforms like Windows.
A graphics API for Windows that did not have the same shape as the Macintosh's QuickDraw API, making a Windows version of Photoshop difficult to create.
The company where Max, a former design lead at GitHub, now works.
A venture capital firm where John worked and contemplated the importance of design in 2014-2015.
Mentioned as an example of a successful company whose success was linked to design, particularly in the mobile era.
Mentioned as the publisher for whom Muriel Cooper designed book fans and a prominent logo.
The company behind Photoshop, whose engineers are noted for understanding graphic design well, similar to the creators of PostScript.
The company where John recently took over design. They are integrating Impeccable into their products, seeing it as a powerful design skill crucial for the digital harness age.
The company name that became popular after the typeface Meta was designed by Erik Spiekermann.
Erik Spiekermann's design firm in San Francisco, which emphasized the concept of design related to the typeface Meta.
A print graphic designer from the 1980s who envisioned electronic publishing and worked with the MIT AI Lab on automatic design.
Cited as an example of a great editor who understood the importance of saying 'no' and having a viewpoint. Also mentioned for testing employees' conviction.
A mathematician and incredible typographer who created TeX, a miracle in combining function and form, similar to the ideal approached by Impeccable.
A typographer who designed the popular sans-serif typeface Meta and founded MetaDesign, illustrating the concept of design as abstraction of craft.
An engineer who worked on the Macintosh and developed QuickDraw, whose API design influenced Photoshop's development.
Paul's boss, formerly at Meta, who used the word 'conviction' to describe leaders who bet on ideas aimed at the global maximum.
More from a16z Deep Dives
View all 57 summaries
42 minBuilding the Future of Image Generation with Ideogram's CEO
50 minHow Exa is Building the Perfect Search Engine | Deep Dives with a16z
51 minWhy AI Agents Need Context | Deep Dives with a16z
46 minPyramid of Work and The Future of Enterprise Automation | The a16z Show
Ask anything from this episode.
Save it, chat with it, and connect it to Claude or ChatGPT. Get cited answers from the actual content — and build your own knowledge base of every podcast and video you care about.
Get Started Free