The story
As Copilot expanded across Microsoft products, people met different AI
experiences depending on where they entered. Windows had its own paradigms.
M365 Copilot had its own patterns. Emerging capabilities, multimodal
understanding, local context, voice, and future agents, added even more
complexity. Without intervention, users risked seeing several different
versions of Copilot across the ecosystem.
The question became a simple, provocative one:
What should the Windows equivalent of a browser address bar be in the AI
era?
Universal Composer was the answer. Rather than treating Copilot as a
separate application, it aimed to become the primary way people interact
with AI on Windows, supporting quick questions, search, productivity,
local context, multimodal inputs, and eventually agentic workflows.
Why this mattered
Windows interactions have always centred on a few anchors: Start, Search,
the taskbar, the shell. AI was a new category, one that blurs the line
between search, creation, assistance, automation, and productivity.
To succeed, Copilot needed fast access, platform consistency, familiar
patterns, low cognitive overhead, and room to scale beyond chat. Universal
Composer became the strategic layer connecting these experiences, a
reusable foundation for future intelligent experiences rather than another
isolated AI surface.
What we learned
The hardest challenge wasn't visual design. It was alignment. Multiple
teams were building AI experiences at the same time, each with different
requirements, customers, and release priorities. Users, however, only ever
see one Microsoft.
The biggest learning was that platform experiences are won through
coherence. If every Copilot experience looked, behaved, and evolved
differently, people would lose confidence and discoverability would suffer.
The turning point
Early on, the conversation was about screens: composer layout, button
placement, window size, modalities. Over time it shifted. Instead of asking
what the Windows composer should look like, we started asking what every
Copilot interaction should feel like.
That reframing turned Universal Composer from a UI component into an
experience system, one that had to support quick questions, search, local
files, photos, voice, screen context, app interactions, agent workflows,
and long-running tasks, without asking people to learn new behaviours each
time.
Design principles
One Copilot, many surfaces
Whether people enter through Windows, M365 Copilot, or a future experience, interactions should feel connected and familiar.
Lightweight first
Universal Composer should be almost instant to invoke, while remaining powerful enough for future scenarios.
Context is a superpower
Local files, screenshots, drag-and-drop, screen context, and camera input should be first-class citizens in the interaction model.
Design beyond chat
The future moves past prompting into workflows, approvals, handoffs, and agents. The design had to scale accordingly.
Native before novel
The experience should feel native to Windows, not a web app transplanted onto the desktop.
What I designed
Universal invocation
Defined how people launch and reach Universal Composer from across Windows
surfaces, reducing friction and turning AI into an immediately available
capability rather than a destination.
Composer architecture
Worked on the composer's silhouette, control hierarchy, layout, multi-state
behaviour, mini-app experience, and full-app transitions, so a single
interaction model could scale across different levels of engagement.
Multimodal inputs
Contributed to experiences supporting voice, file uploads, screen context,
drag-and-drop, local content, and camera input, letting people work
naturally instead of forcing everything through typed prompts.
Local context
A key Windows differentiator was bringing local context, files,
applications, screens, directly into Copilot, while keeping user control
and transparency intact.
Cross-team alignment
Built alignment frameworks and ran multiple review cycles to keep Windows
AI, M365 Copilot, and future Composer scenarios consistent, preventing
fragmentation across Microsoft's AI products.
Future agent foundation
Even in early phases, the composer was designed with agent workflows in
mind: long-running tasks, approvals, permissions, context persistence, and
artifact refinement, areas that later became central to the agentic vision.
Building to learn
The project ran on collaboration across organisations. The design evolved
through Windows reviews, UX walkthroughs, build presentations, alignment
trackers, prototypes, validation exercises, and end-to-end build audits,
continuously refined against platform feedback rather than designed in
isolation.
Outcomes
Universal Composer became part of a broader shift that repositioned Copilot
as a native Windows experience. Overall chat engagement grew by roughly
121%, premium engagement by around 202%,
and voice usage by about 200%, with voice satisfaction near
70%. Retention held strong across the ecosystem, roughly
80% overall and higher still among premium users.
Beyond the numbers, Universal Composer mattered enough for Windows teams to
formally evaluate its effect on device retention, new-device usage, M365
engagement, search efficiency, and AI adoption across the platform. The
results validated the core bet: AI becomes far more valuable when it's part
of the operating system, not just a standalone app.
Looking back
The most valuable contribution wasn't designing a composer. It was helping
establish a common language for AI interactions across Microsoft. Universal
Composer represented the start of a larger shift: from launching apps to
invoking intelligence, from opening software to expressing intent, from
navigating interfaces to collaborating with agents.
What's next
Universal Composer is evolving from a chat entry point into an agentic
foundation for Windows: persistent context, multi-step workflows, local and
cloud intelligence, agent orchestration, long-running tasks, artifact
generation and refinement, and cross-device continuity.
Make AI feel like part of Windows itself.
