Copilot Design System
Microsoft's Copilot Design System is useful because it treats AI UX as a system of presence, memory, attention, and shared awareness, not a chat widget or isolated feature. Source: Microsoft Design, "A simplified system", 2026-06
Core Reading
The article argues that design systems in AI products need to codify model behavior and new mental models, not only visual components. Copilot is framed as a thought partner woven across Microsoft 365 surfaces; the design system's job is to make that partner feel continuous, intentional, and humane. Source: Microsoft Design, 2026
Routing Summary
Use this page as an AI-native design-system reference for context continuity, agent presence, memory, focus handoff, and user-control patterns. It informs agent UI design decisions; it is not a component library or implementation dependency. Source: Microsoft Design, 2026
The four useful lenses from Qian Jiang's thread are:
| Lens | Meaning for agent UI |
|---|---|
| Presence | The assistant should be available in context without feeling like a modal interruption. |
| Memory | Chat is not only a transcript; it is the continuity layer that preserves prior work. |
| Attention | The agent should meet the user at the right level of focus: broad exploration, local execution, or suggested next action. |
| Shared awareness | The system should make handoffs visible so the user understands where the agent is focused. |
Source: X/@qianjiang___ self-thread, 2026-06-16
Four Surface Elements
Microsoft describes four architectural elements across Office apps:
| Element | Role |
|---|---|
| Dynamic Action Button | Contextual in-app entry point to Copilot and Chat. |
| Chat | The broad memory and exploration surface. |
| On-Canvas | Local execution surface attached to selected content. |
| Suggested User Actions | Contextual nudges that adapt across entry point, chat, and canvas. |
The useful pattern for Kevin's products is not to copy Microsoft's UI. The useful pattern is to separate global assistant memory, local in-workspace execution, and contextual suggestions as different surfaces with different permissions and density. Source: Microsoft Design, 2026
Throw & Catch
Throw & Catch is the handoff pattern between Copilot surfaces. The agent should move between entry points without making the user feel that context was dropped or that a separate feature stole attention. The interaction is successful when active focus is visible, the inactive surface recedes, and the user still feels in control. Source: Microsoft Design, 2026
For AI-Native Design Patterns, this adds a missing design-system invariant: agent presence needs a focus-transfer protocol. Tool cards, chat, inline edits, canvas suggestions, and command palettes should not compete for ownership of the run. Each surface needs a visible "who has focus now?" state. Source: X bookmark artifact review, 2026-06-30
Kevin-Stack Implication
Use this page when designing Agent Machines, Loop, wiki UI, or any agent console with multiple work surfaces. The concrete checklist:
- one persistent memory/transcript layer;
- one local execution layer attached to the current artifact;
- one contextual entry point that is always nearby but quiet;
- visible handoff when focus moves;
- no interruption without value;
- no hidden state that forces the user to re-explain context.
Source: Agent Product Surface; Microsoft Design, 2026
Timeline
- 2026-06-16 | Qian Jiang amplified Microsoft's article with the four-lens model: presence, memory, attention, and shared awareness. Source: X/@qianjiang___, 2026-06-16
- 2026-06-30 | Page created during one-by-one X bookmark artifact review and linked into the AI-native design-system capsule. Source: raw/x-bookmarks/enriched/2066929153782546792.json