Why agentic design still needs the human sketch

The core idea. Maggie Appleton, a design engineer at GitHub Next, argues that the most valuable design work happens before any code is written—in sketches, notebooks, and live prototypes. Agents can speed up implementation, but they cannot replace the human process of understanding materials, constraints, and context. The real bottleneck is not model capability but workflow design and team alignment.

Why this matters now. As agents become competent at generating UI code, engineers are tempted to bypass designers entirely. Appleton warns that agent-generated design is easy to spot and lacks cultural nuance. Design is part fashion, part function; agents trained on current aesthetics will produce forgettable, generic interfaces. Teams that rely on agents for design risk losing differentiation.

The tension. Appleton shows a live demo of a "jig"—a set of sliders and knobs built by an agent that lets her tweak animation variables in real time. This is faster than static Figma mockups and closer to the principle of "live programming." But she admits the agent cannot yet match her taste. She still ends up manually adjusting padding and colors, suggesting the craft gap is real and may persist.

The missing piece. The biggest gap in current agent tools is multiplayer collaboration. Every agent session is private and local. Teams cannot align on decisions before handing off to agents. Appleton prototypes a decision-centered interface where each choice gets its own card, assigned to a human, with diagrams and context. This is the unsolved problem: how to make agents work in teams, not just for individuals.

What most will miss. The talk surfaces a subtle trade-off: as agents make implementation cheap, the upfront alignment work becomes more expensive. Teams must agree on design and architecture before agents start coding, because rapid iteration no longer exists. This shifts the skill set needed from "writing code fast" to "specifying intent precisely." Engineering leaders should invest in decision-making tools and shared context, not just more agentic harnesses.

The bottom line. Appleton’s experience suggests that agents are a productivity amplifier, not a replacement for design thinking. The hardest problems—team coordination, taste, and user understanding—remain human. The hype around agents replacing designers is overblown; the real opportunity is to use agents as coaches and rapid prototypers while keeping humans in the loop for judgment.