
Introduced AI Agents into Copilot Notebooks to boost productivity
TIMELINE
2 months
TEAM
2 Designers
1 Product Manager
TOOLS
Figma, Figjam
ROLE
UX Design
AI Interaction Design
OVERVIEW
TL;DR
CHALLENGE
3rd-party companies need ways to interact with AI Agents within existing workflows to improve productivity.
AI Agent conversations are disconnected from Notebooks, making them difficult to discover, access, and carry context across workflows. This creates unnecessary context switching and limits continuity between Notebook work and Agent interactions.
SOLUTION
Incorporated AI Agent interactions directly via Notebooks to help users transition between agents and agent conversations.
As AI Agents become part of everyday workflows, we brought their capabilities closer to where work already happens. Integrating Agent access within Notebooks made relevant AI assistance easier to discover and enabled people to build on their existing work without disrupting their workflow.
IMPACT
15+ interviews
in 4 weeks, synthesizing insights into actionable ideas
5 explorations
and prototyped it for further feedback
100% positive
feedback and approval from leadership
CONTEXT
Understanding The Ecosystem

Copilot
An AI assistant built into Microsoft 365 that helps users create, analyze, summarize, and complete tasks using natural language.

Notebooks
A focused workspace in Copilot where users bring together relevant files, notes, and sources to organize information and work with AI on a specific topic or project.

Create a new notebook.

Add files, Copilot pages, and other types of references.

Get focused answers based only on the references added to notebook.

Ask Copilot questions about the content in your notebook.
PROBLEM STATEMENT
Notebooks are limited to Copilot chats and do not extend to agents.
Microsoft's partner companies, including Cohere, S&P Global, LSEG, and others, have their own agents integrated into Copilot, which is built on Microsoft 365 foundational models. However, these companies experience challenges with the Copilot agent due to limitations in interacting with agents within Notebooks.
Preventing users from bringing agent interactions into their projects and creating gaps in cross-surface continuity, making it impossible to save, reference, or build on past agent interactions alongside their project context.


London Stock Exchange

S&P Global

Autodesk

LexisNexis

Cohere
RESEARCH
Discovering User Insights
1:1 WITH CUSTOMERS
15 Stakeholder Interviews
With any project, I start with questions.
Where do customers see the biggest gaps in the current experience?
What would make them want to bring an agent chat into a Notebook?
What is the difference between Copilot chats and agent chats from a user perspective?
We needed explorations and ideas quick. We had to move fast. I got creative with a limited research budget - let's interview the partner company PMs from next door about their B2B needs?
Internal research > No research.



















PAIN POINTS
Limited Access
Relevant work from other agents can’t be brought into the project.
Context Loss
Work gets interrupted when moving between different agents due to context loss
Manual Effort
Connecting work across agents requires unnecessary copying, switching, and repetition.
HOW MIGHT WE
…enable easy knowledge continuity through a richer, more connected workspace in M365 Copilot Notebooks.?
TECHNICAL CONSTRAINT
Copilot Notebooks are technically implemented as agents, preventing agent nesting.

Agent within agent (Notebook) nesting

Access agents from within a Notebook
Design Decision
Instead of embedding agents within a Notebook, we approached agent linking model where users could add and access specialized agents from the Notebook.
Why? Because this preserved the user's mental model of "my Notebook has the agents I need" while avoiding architectural changes required for agent-to-agent nesting.
Interaction Pattern
The Notebook became the workspace for discovering and navigating between agents, rather than the container that owned those agents.
FEATURE COMPARISON
I also looked at our competitors. They all offer a similar experience to Notebooks, called Projects or Spaces. Across the board, these tools allow users to add files and chats into a project, which is consistent with what we see in Copilot Notebooks.



Claude, ChatGPT, Perplexity
Currently, none of them integrate agent experiences within those spaces.
So this opened up an opportunity for me to define how agents and notebooks could work together.
Let's connet!
c
2026 Aditi Gupta


