What Happens to Design After AI?

Episode Summary

Executive Summary: The conversation argues that AI is rapidly commoditizing routine design work while increasing the value of human taste, restraint, and conviction. John Maeda and Paul Backus describe “auto-design” and agentic design tools as ways to raise the floor for builders, but emphasize that the future belongs to designers who can raise the ceiling, steer models with precise vocabulary, and shape new agent-facing experiences.

Main Topics: AI as a design multiplier, not a replacement (Priority: 5/5): The speakers frame AI as automating repetitive work while enabling designers and engineers to focus on higher-level creative and strategic decisions. They see this as a shift toward better tools, not the end of design craft. Taste, vocabulary, and the limits of models (Priority: 5/5): Paul argues that designers get better results from Claude because they use richer design language, while John notes that models are trained on humanity’s output, not the underlying decision process that produces taste. Impeccable and computational craft (Priority: 5/5): Paul explains Impeccable as an open-source agent skill that adds design vocabulary, a quality layer, and visual iteration to reduce slop and produce more distinctive outputs. Uniqueness and the moving target of “AI slop” (Priority: 4/5): The discussion emphasizes that default model aesthetics evolve over time, so anti-patterns and design defaults shift; tools must constantly steer away from overused styles and toward uniqueness. From UX to AX (agentic experience) (Priority: 5/5): John argues that design will increasingly move from visual user experience to agent-oriented interfaces like help text, error messages, CLI/API design, and machine-readable affordances. Human judgment, conviction, and cognitive delegation (Priority: 4/5): Both speakers stress that tools can assist, but humans must still drive decisions. Paul warns against surrendering to the model, and both emphasize leader conviction when choosing between speed and taste. Design’s future market structure (Priority: 4/5): They foresee more bespoke, smaller-market software products that become viable because AI lowers the cost of creation, while also increasing trust and accountability as differentiators.

Key Arguments: Designers often get better LLM outputs than engineers because they use more precise visual and aesthetic vocabulary, which changes the quality of the prompt. LLMs can approximate taste, but they are trained on the output of humanity rather than the decision-making process that created it, so they lack true design intent. AI will commoditize routine design tasks, but that makes exceptional taste and high-end craft more valuable, not less. Impeccable improves AI-generated design by injecting design language, quality controls, and iterative visual steering into the agent workflow. “AI slop” is a moving target: once the market learns to avoid one aesthetic pattern, models and users drift to another default style. The next frontier is AX—designing for agents, not just humans—through APIs, CLIs, error states, help text, and information architecture. Human trust and accountability remain key reasons people pay for bespoke products, especially when outputs become increasingly easy to generate. Leaders must help teams choose global maxima over local maxima by communicating conviction, not just deadlines or metrics.

Data Points: Design importance surge: 2014-2015 - John Maida links the rise of design’s importance to the mobile era and Airbnb’s success. Impeccable usage split: 50/50 - Paul says the target audience is now roughly half engineers and half designers, plus PMs, founders, and solopreneurs. Default aesthetic shifts: 2022-2023 - Paul says purple-gradient “AI slop” was associated with models in 2022-2023, but defaults have shifted since then. Design automation forecast: 2020 - John cites his earlier prediction that software would return to a cottage industry with bespoke apps, noting he was about five years early. Design market segment: 80% / 20% - John describes automated design solutions covering 80% of cases so humans can focus on the last 20%.

Pivotal Quotes: "LLMs have been trained on the output of humanity, not on the input." — Paul Backus: Explaining why models can mimic taste but don’t understand the underlying design decisions that produce it. "Taste is a whisper and velocity is a megaphone." — John Maeda: Describing why design nuance often loses to deadlines and measurable outcomes in decision-making rooms. "We’re in this era where design will be further commoditized and automated so that a new kind of design emerges." — John Maeda: Summarizing his view of the shift from routine UI work to agentic experience and higher-level craft.

Implications: AI will make average design easier and more common, but human taste, conviction, and trust will matter more at the top end. Builders should learn to steer models with design language and prepare for agent-facing design as a major new discipline.