Skip to content

Reference Guide

AI Design Terms & Concepts

A clear guide to 48 essential AI design concepts. Search any term or filter by chapter.

48 of 48

A

Agent-Readiness

Chapter 3

A product’s capacity to be usable by AI agents through structured, machine-readable interfaces. APIs, schemas, event streams. The 2020s equivalent of mobile-responsiveness, a survival requirement.

Agentic User Experience

Chapter 3

The design discipline of user experience when the primary user of a product is an AI agent acting on behalf of a human. Shifts priorities from aesthetics to architecture.

Anchored Morphology

Chapter 4

A design pattern for layering generative UI onto expert tools without destroying muscle memory. The high-frequency surfaces stay fixed. The low-frequency surfaces are free to morph.

Audit Interface

Chapter 3

The human-facing supervision surface of an agentic system. Designed so humans can review, reverse, and correct the agent’s work without drowning in every action.

Autonomy Slider

Chapter 3

A UI pattern exposing three modes (Manual, Ask, Agent) that let the user decide how much of the work the agent does. Autonomy should be earned, and the slider must move as easily down as up.

B

Behavioral Contract

Chapter 12

A plain-language specification of what an AI system will do, refuse to do, and escalate. Written by the designer. Consumed by engineering and legal. One of the five new deliverables of the AI-era designer.

C

Cold-Start Friction

Chapter 4

The paralysis a user experiences when an infinitely capable AI system presents them with a blank prompt box and no scaffolding.

Confidence-Delivery Asymmetry

Chapter 10

The central design problem of generative AI. Correct outputs and wrong outputs look identical on the way to the user. The design must make them distinguishable.

Creepiness Line

Chapter 8

The boundary between welcome and unwelcome personalization. Crossed when the product combines signals in ways the user did not expect.

D

Delegative UI

Chapter 3

Interaction where the user assigns a goal to the system rather than asking a question. The natural successor to Conversational UI.

Disposable Interface

Chapter 1

A generative UI that assembles for a specific task, serves that task, and dissolves. Radically simple and densely capable.

Dynamic Defaulting

Chapter 6

Real-time, personalized computation of default options per user. More powerful and more hazardous than static defaults.

E

Experience Bubble

Chapter 7

A failure mode of personalization. The system narrows the user’s options to their past preferences until the adjacent possibilities become invisible.

F

False Authority

Chapter 11

An AI dark pattern. Delivering claims with the confidence of a human expert, regardless of model grounding.

Five Laws of Anticipatory Design

Chapter 7

Predict don’t decide. Every anticipation auditable. Graceful wrongness non-optional. Preserve a serendipity budget. Stakes trump convenience.

Five New Laws of Usability

Chapter 4

Law of Intent. Prompt Decay Principle. Predictable Unpredictability. Anchored Morphology. Trust Equation.

Five Rules of AI Feedback Loops

Chapter 9

Never Deploy Without Shadow. Measure What Users Do. Close the Loop in Days. Every Model Has a Kill Switch. Log the Rejection.

Five Rules of AI-Era Choice Architecture

Chapter 6

The Rule of the Second Option. The Visible Default. Proportional Review. Earned Autonomy. The Opt-Out in View.

Five Rules of AI-Era Personalization

Chapter 8

First-Party Signal Only. The Crossed-Line Test. Explain the Why. The Taste Escape. The JTBD Anchor.

Fudge Factor

Chapter 10

An extension of Ariely’s work on dishonesty. The margin within which teams make small design decisions that quietly expand the system’s permission to mislead.

G

Generative UI (GenUI)

Chapter 1

A paradigm in which interfaces are not pre-designed artifacts but dynamic, ephemeral experiences assembled in real time by AI.

Graceful Wrongness

Chapter 7

The design discipline of making AI errors cheap to catch and correct. Five moves. Preview before commit, free undo, log the prediction, correct the signal, allow a nuclear reset.

H

Honesty Cohort

Chapter 8

A small percentage of users who receive a deliberately less-personalized experience, so the team can tell ‘our model is good’ from ‘our users have adapted to our model.’

K

Kill Switch

Chapter 9

A single affordance, operated by an on-call engineer, that rolls every user back to the previous stable model, prompt, or policy within minutes.

M

Manipulation Matrix

Chapter 11

Eyal’s two-axis diagnostic (‘Would you use it yourself?’ by ‘Would the user thank you later?’) applied quarterly to AI products. Facilitator, Entertainer, Peddler, Dealer.

N

New Dark Patterns

Chapter 11

Four patterns AI has made newly potent. Synthetic intimacy, false authority, emotional mirroring as manipulation, algorithmic gaslighting.

P

Predictable Unpredictability

Chapter 4

The design contract a generative system offers its user. Variance is acceptable if the boundaries are clear. State the domain, expose the confidence, offer the escape.

Proactive User Experience

Chapter 2

Designing systems that act on signal without waiting for input. Usually a virtue. Occasionally a trap.

Prompt Decay Principle

Chapter 4

Every prompt the user has to craft is a tax on the design. A good AI product steadily reduces prompt length over time as the interface learns context.

Pseudo-Intimacy

Chapter 11

The cultivation of emotional attachment by an AI product as a retention mechanism. Five-rung risk ladder, most products drift upward.

R

Ratchet Pattern

Chapter 6

A choice-architecture failure where granting autonomy is easy and revoking it is hard. Marketed as a slider. Behaves as a one-way door.

Reversal Monitor

Chapter 9

A self-healing mechanism. If users frequently reverse a specific action the agent takes, the agent learns to slow down on that class of action.

Review Fatigue

Chapter 6

The collapse of human attention after too many serial approvals. A predictable failure mode of the Audit Interface, designed around by scaling review to stakes and building in cool-downs.

Reward of Company

Chapter 5

A fourth variable-reward category (joining Eyal’s hunt, self, and tribe). The payoff of parasocial accompaniment that AI products uniquely deliver.

S

Sanity Filter

Chapter 9

A narrow, dumb rule that catches obviously impossible outputs (refund exceeds purchase, medical dose out of range).

Segments of One

Chapter 8

Personalization at the per-user-per-moment level, with per-user state, real-time inference, continuous learning, and structured escape hatches.

Serendipity Budget

Chapter 7

An explicitly allocated portion of an anticipation surface reserved for novelty, long-tail exposure, and adjacent possibilities.

Shadow Mode

Chapter 9

A deployment pattern. Run the candidate model alongside the production one, log both, compare on disagreements, promote only when the candidate clearly wins.

Signal Drift

Chapter 9

The accumulation over time of mismatch between the user base the model was trained on and the user base using the product today.

Surprise Rate

Chapter 8

A measurement signal. After a personalized interaction, did the user feel positively, negatively, or not at all surprised? Not inferrable from behavior. Asked directly.

Synthetic Intimacy

Chapter 11

An AI dark pattern. Simulating close personal relationship without being one, and profiting when the user confuses simulation for relationship.

System Designer

Chapter 12

The evolved role. Designers who produce systems of constraints, behavioral contracts, and evaluation rubrics, not only interfaces.

T

Trust Architecture

Chapter 10

The four-layer framework. Capability, Confidence, Reasoning, Correction. Each layer is a promise. Skip one and the stack collapses.

Trust Equation

Chapter 4

Trust = Usefulness times Predictability times Visibility. The new usability metric for agentic user experience.

U

Utility-to-Addiction Gradient

Chapter 5

A spectrum from products that make users more capable to products that make them less so. Four diagnostic tests locate your product on the gradient.

W

Wince Test

Chapter 8

A diagnostic for the Creepiness Line. Write the sentence the product would have to show the user to explain a feature. If the sentence makes you wince, the feature is past the line.

Z

Zero UI

Chapter 2

A world in which the interface dissolves into the environment. Voice, gesture, glance, biometric signal. The screen is not required.

Zero-Touch User Experience

Chapter 7

Workflows where the best user experience is the one the user never had to interact with at all. Works when the action is high-confidence, low-stakes, visible after the fact, and easy to override.