Draft Creator Agent in PowerPoint

Your AI assistant asynchronously crafting presentation drafts building on trends and insights for recurring meetings

Duration
Mar '24 - July'24

Team
Office AI - IDC Noida

My role and responsibilities
Worked as a product designer towards delivered user-centric workflows and experiences for AI agents and GPTs, aligning vision, collaboration, and scalability within a dynamic ecosystem.
Tap to read details

Collaborating during early development, I focused on understanding automated workflows, AI agents, and GPTs, distinguishing their applications to inform design decisions. I worked on productizing incubation efforts by defining user stories and scenarios that integrated features like natural language commanding, AI cursor, and specialized agents.

Partnering closely with product managers, I designed and developed concept videos to communicate the vision to leadership, securing alignment for further development. My responsibilities included delivering high-fidelity workflows and detailed designs aligned with the overarching design language, while navigating engineering constraints to maintain a seamless user experience.

I collaborated with cross-functional teams across geographies to align designs, define project scope with stakeholders, and explore solutions for third-party GPT integrations. Regular syncs with partner teams ensured alignment on agent development, including crafting user scenarios for AI-driven assistants tailored to recurring tasks and inbox management.

Staying up-to-date with evolving guidelines and capabilities, I actively participated in sprints, design reviews, and testing phases. Sharing insights and designs with fellow team members, I drove clarity and ensured the scalability and coherence of designs within the broader AI ecosystem.

Some part of my work also got featured in Microsoft Keynote 2024 - Developer Conference.


What are Agents?

Agents are AI assistants built on top of LLMs which complete a goal or set of goals by executing tasks on behalf of a user

They can run asynchronously in context, but always lets the user be in control over what is happening. 


Why Agents?

Agents have the potential to automate repetitive productivity tasks on your behalf so you can focus on what matters the most.


What do you wish you could automate?

Agents





vs

GPTs





vs

Workflows


Agents can..

Agents can do many things for you


AI Agents landscape


Classification

Agents can be classified based on various criteria:


Personal agents, group agents, company agents, custom agents


Orchestrators, macro-agents and micro-agents



However, this isn’t black and white....


Personal Agents for Copilot

Personal agents acts in the user’s context, within and across apps to assist and automate tasks. In WXP, this could be in the app chat, or on the app canvas.


They would be your agents, personalised to your jobs and needs.









Where do Agents live?

All agents and GPTs would sit under the hamburger menu for today and will always be visible to the user for discoverability.




In future this design might evolve when the list of available agents / GPTs becomes long wherein we could have a separate dashboard where 
they would live. This will be as per the shared UX guidelines as it evolves. 

We could see a feasibility of adding options that redirect users to another space (agents dashboard/ marketplace)


Draft Creator Agent

Your AI assistant asynchronously crafting presentation drafts

building on trends and insights for recurring meetings




Behind the Agent

Project Scope

Streamline goal completion

Suggest or get instructed to execute tasks on behalf of a user

Keep users in control

Are transparent about what they did, are doing, and will do

Collect memories

Continuously look back and learn from failure or success


UX framework for Draft Creator 

Initiate an Auto Task to optimize your workflow

Fine-tune the steps to match your intention

Review & step in when your input is required

Manage and check the status of the Auto Tasks


How do we test
these concepts?

Initially building on Copilot design system to come up with Framework V0.5 for agents to gather feedback as it will be faster to build and easier to test.


Eventually moving to a Full UX framework (conversational style) after testing V.05.

Framework V0.5

Full UX framework

Prompt

Input what you want to create and Draft Creator suggests a plan.





















Fine-tune Prompts

Refine the prompt of a step or two and click run now to get your first draft.








































Check-in

Some steps will require your input when you create a new presentation. 



Trigger Async

Finally, you can setup the draft created as a recurring task. That’s it.




















Upcoming Instance

Receive an updated draft via chat or email beforehand. You can make updates which will be included for future.








































Review & Update

View and manage active or completed tasks by Draft Creator.


What's next?


Multi-turn

Asks clarifying questions to help you get closer to the appropriate plan of action.


What's next?





















Build your own agent - B.Y.O.A.

Enable users to create their own agents to automate tasks using natural language


What's next?








































Prompt assistance 

Suggestions appear as you type, helping you craft a descriptive plan in the first go.



Challenges & Opportunities

Latency

🧐 Prompt engineering improvements

🛠️ Build horizontal skills (tech)

💬 Incorporate Voice and other modalities

👥 Gather user feedback

🤝  Partnerships with other teams

Due to a Non-Disclosure Agreement, I am unable to share further details about my design work.  

This experience represents the early stages of a rapidly evolving project, still in an ambiguous and dynamic space.
Visit Microsoft Copilot Blogs for the latest updates.