Can AI become a design material for adaptive interfaces?

Josh Clark and Veronika Kindred, introducing their book for Big Medium, argue designers should adopt AI as a design material, not just a productivity tool. They define "Sentient Design" as "the practice of creating intelligent interfaces" — experiences "with the awareness and agency to respond to your users in the moment: dashboards that design themselves, intelligent canvases that bloom with data visualizations, AI that participates in apps like any other user, and so much more," stating this "goes way beyond text-based chat and agents." The book's contribution, per the excerpt, is a "framework that navigates 14 experience patterns you can use today" — a taxonomy of interface types for this material, filling what the authors say was a gap with "no map for this new territory, no uniform vocabulary, no framework for choosing what type of intelligent interface to create—or even what those types might be."

Their reasoning rests on a demonstration they call "Sentient Scenes": a weekend project where a glowing square, told to be "a graceful ballet dancer," began to "glide and pirouette across the screen," then told to "rocket to Mars" turned red and "zoomed back and forth," then asked to play "underwater adventurer, inept wizard, cartoon puppy" and "obliged every time." They stress "none of this was explicitly coded... we never taught it to dance or drive a submarine," characterizing the result as "an interface that took any request and adapted its entire form in response," producing "experiences we had neither designed nor anticipated." The implied mechanism: letting the model's generative range drive interface behavior directly, rather than scripting responses to anticipated inputs, is what lets form — not just content — adapt at runtime.

Where How should AI interfaces handle the shift from doing to supervising? proposes a fixed structure that mature intent-based systems should converge on — three layers, each answering a specific failure mode — Sentient Design proposes a different kind of map: a pluralistic taxonomy of 14 interface types a designer chooses between, organized by what kind of experience the material is shaped into, not by failure mode. The two are compatible rather than contradictory — Nielsen's orchestration/direct-manipulation split could sit inside any one of the 14 patterns — but Sentient Design's claim is explicitly broader than chat and agents, while Nielsen frames his layers around conversational, intent-driven systems. How should AI systems be designed across different interaction phases? organizes by interaction phase and behavior rather than interface type, making it a behavioral complement to Sentient Design's form-level taxonomy: one says what kind of interface to build, the other how it should behave once built. What UX principles do workplace users want in AI agents? frames its principles around agentic behavior specifically, narrower in scope than Sentient Design's stated reach beyond chat and agents into dashboards and canvases.

The excerpt is a book introduction written to sell its premise, not to argue it in full: it names "14 experience patterns" but does not define, list, or demonstrate any of them here, and the Sentient Scenes anecdote is a single informal demo, not a study of how often AI-as-material interfaces succeed versus the authors' own admission that "we've seen the sparkles fizzle as often as they dazzle." It does not say what methodology produced the 14-pattern taxonomy — client work, systematic survey of existing products, or post-hoc naming are all left unstated. What the excerpt establishes is a framing and vocabulary claim — that form-adapting, agentive interfaces are a nameable, mappable category distinct from chatbots — not evidence that the specific 14-pattern map holds up under use, or that comparable adaptivity is reliably reproducible outside a demo.