A field guide to AI interfaces
How the best AI products actually behave.
The interaction patterns of the AI era — named, categorized, and shown in real products. Every entry explains why it works, what to steal, and what to skip. Failure states included.
Six categories
Organized by what the user is doing
Invoke
Getting started and steering: how products turn a blank box into a beginning.
Observe
Watching AI work: reasoning, progress, live workspaces, and system honesty under load.
Verify
Trust and grounding: how claims get sourced, checked, and made inspectable.
Control
Agent oversight: plans, permissions, interruption, and the cost of autonomy.
Remember
Memory and personalization: what the AI keeps, and who's in charge of it.
Transact
AI meets commerce and services: when the conversation becomes the storefront.
From the library
Recently documented
Action Plan
The agent proposes its plan and waits for approval before touching anything.
Stream of Thought
Show the model's reasoning as it happens, so waiting becomes watching.
In-chat Commerce
The conversation becomes the storefront — decide, order, and track without leaving the thread.
Live Viewport
A window into the agent's workspace — watch it browse, click, and build in real time.
Citations
Every claim carries a pointer back to where it came from.
Scarcity States
Honest UX for the moment the AI runs out — of capacity, quota, or compute.
Method
Tracked by machine, judged by a human
01
Watch
A pipeline monitors 48 AI products across app stores and changelogs, every week.
02
Triage
Version bumps and diffs get filtered for one thing: did an interaction pattern change?
03
Capture
Real flows on real devices — including the failure states nobody screenshots.
04
Document
Analysis over eye-candy: why it works, what to steal, what to skip.
Newsletter
One AI product, taken apart, every other week.
New patterns and product teardowns from across the AI landscape — what to steal, what to skip, from someone who actually uses them.