Libraries we build on

Microsoft Research and Aether

Microsoft HAX Toolkit

Eighteen Guidelines for Human-AI Interaction from a CHI 2019 paper by Amershi and colleagues, with a design library of examples, a workbook to prioritise guidelines and a playbook.

Best for. Checking a design against the guidelines early, and naming what a pattern is for (G1 to G18).

Evidence. The guidelines distil two decades of research and were refined with design practitioners. The examples are illustrations, not test results.

Google PAIR

People + AI Guidebook

Six chapters: User Needs + Defining Success, Data Collection + Evaluation, Mental Models, Explainability + Trust, Feedback + Control, Errors + Graceful Failure. Each chapter has patterns and worksheets.

Best for. Product thinking before pixels: what people need, how to set expectations, how to fail gracefully.

Evidence. Based on research and product experience at Google. No per-pattern user-test results.

Emily Campbell

Shape of AI

A pattern library in six categories: Wayfinders, Prompt actions, Tuners, Governors, Trust builders and Identifiers, with product examples. Shared under CC BY-NC-SA.

Best for. A shared vocabulary and examples for generative interfaces, especially prompting and control.

Evidence. Curated examples from products. No user-test results are reported.

Aiverse

Aiverse

Forty patterns in five phases: Discovering, Instructing, Observing, Refining and Adapting, backed by a library of several hundred real product interactions and case studies. Commercial, with a paid plan.

Best for. Seeing how well-known products solved a specific moment in an AI flow.

Evidence. Examples from products. No user-test results are reported.

IBM

Carbon for AI

Conventions for AI inside the Carbon design system: how to mark AI-generated content and where explanations live in components.

Best for. Teams that ship a design system and need consistent AI marking across products.

Evidence. Design-system guidance. No test results per pattern.

Apple

Human Interface Guidelines: Generative AI

Platform guidance for generative features on Apple platforms: what to tell people, how to handle mistakes and how to keep people in control.

Best for. Native apps on Apple platforms.

Evidence. Platform guidance. No test results per pattern.

A receipt on the card

Which case, what was compared, how many people, what happened and what we do not know. The label Tested is derived from that card, never typed. How a pattern earns its label.

Flemish practice cases

Real products of companies and organisations in Flanders, tested with their own users in the Odisee UX Lab or on site. The cases.

A compass for your context

Five questions rank the patterns and the metrics for your situation, with the rule that fired named on the result. Find my patterns.

A research side

Why a pattern should work, the properties it targets in the KU Leuven evaluation framework, and how to measure them in ten minutes. Vocabulary bridge.