Paul Conlin

Principal software engineer at Microsoft. Saint Paul, Minnesota.

I work with design and product to build AI agent experiences in PowerPoint. I lead extensibility for the agent, including skills and plugins, and I build agentic engineering workflows that engineers across PowerPoint now use.

Before that I spent seven years on cameras and video at Flipgrid and Microsoft, where I led the team behind Microsoft’s shared camera.

Email, LinkedIn, GitHub

Selected work

2026–nowPowerPoint, Microsoft

Agentic engineering

I build agentic engineering workflows for the PowerPoint org. A teammate and I started a plugin for coding agents at a hackathon, and it became the org’s default for autonomous development. It takes a task all the way to a draft pull request, validates each change in the running product, and posts the evidence to the pull request.

I also write the guidance our coding agents follow, and I published a plugin that helps people who aren’t engineers write skills for PowerPoint’s agent.

1022025–nowPowerPoint, Microsoft

Skills and plugins for PowerPoint’s AI agent

I lead extensibility for PowerPoint’s AI agent, which includes skills and plugins. A skill teaches the agent how to do a specific task. People can write their own, and subject-matter experts can package what they know for everyone else. I led the skills platform end to end, from picking a skill in the app to how the agent loads it, and we shipped the first version in six weeks.

I work with the Excel, Word, Office AI, and Microsoft 365 Copilot teams so skills work across apps and on both Claude and GPT, with the admin and data-protection controls that companies need.

Tighten every slideBuilt in
Match our brand voiceShared by Marketing
Quarterly business reviewShared by Finance
Check accessibilityBuilt in
Figure 2. Picking a skill from the message box. The skills are made-up examples.

1042025PowerPoint, Microsoft

Clarifying questions

When a request is missing something important, PowerPoint’s agent stops and asks. I designed and built these questions, from the React UI to the tool the agent calls, to replace a template and style picker. The agent can ask a few questions in one card in the chat, one step at a time, and the card folds into a short summary once you answer.

The card became the standard way agents ask questions across the organization, and Excel’s agent team reused the component.

Make a deck about our third-quarter results

Who is this deck for?

Figure 3. One card in the chat, three questions, one step at a time. Answer them to see the card fold up.

1062024–2025Microsoft 365 Copilot

Microsoft 365 Copilot Create

I architected the front end of Create, the foundation hundreds of engineers built on for its May 2025 launch. It launched without performance regressions. I also owned image, poster, and banner generation with GPT-4o.

Figure 4. One design laid out as an image, a poster, and a banner.

1082023–2024Microsoft Designer

Microsoft Designer and Image Creator

I set the architecture for Designer’s AI apps, including Image Creator, Sticker Creator, and Restyle. With design and product, we merged the desktop and mobile builds into one responsive app, so Image Creator launched on both.

I shipped Image Creator’s prompt templates, where you fill in the blanks to make a prompt your own. People used one of them to make 3D avatars of themselves and shared them on X. It was Designer’s first viral moment, and Satya Nadella posted about the templates on LinkedIn.

1102021–2023Video Experiences, Microsoft

OneCamera

I led a 22-person organization across web, iOS, Android, and backend that built OneCamera, Microsoft’s shared camera. It shipped in 10 Microsoft products, including Teams, Xbox, Skype, Outlook, and SwiftKey. We launched Skype Clips, OneCamera’s first consumer web integration, and brought Snap camera lenses to SwiftKey after working through biometric privacy laws with our legal team.

I reorganized the team from platform silos into cross-platform feature crews whose leads also owned platform health. Crashes fell over 95% on web and 60% on mobile.

Pick a lens, then record a clip of up to 15 seconds.

Figure 5. A camera with lenses. In a drawing, a lens changes how the scene is rendered.

1122016–2021Flipgrid and Microsoft

Flipgrid’s camera

I joined Flipgrid as a front-end developer, led React development through Microsoft’s 2018 acquisition, and then led the web team. I built the in-browser camera, and Teams, Stream, and Yammer later embedded our camera SDK.

When schools moved to remote learning in 2020, I proposed and built a screen recorder in two weeks. Its launch became Flipgrid’s most popular social post. I also worked with the Edge, Safari, and Chrome teams on browser video: the first web video-editing API with Edge, MediaRecorder in Safari, and Chromium fixes that landed upstream.

0:14Proposed trim: from the first word to the last.
Figure 6. The idea in my trim patent: find where the speaking starts and stops, propose a trim, and apply it in one click.

About

I work as a product engineer. I like to start with design, prototype until the idea is clear, and then finish the details people notice: loading states, keyboard focus, and motion that shows what changed.

I was an engineering manager for five years and coached six engineers into lead and manager roles. On PowerPoint’s agent team, I served as interim manager for five engineers and led a crew of more than 12 engineers building presentation creation. I still mentor engineers on my team.

Experience

  • 2025–nowPrincipal software engineerPowerPoint, Microsoft
  • 2023–2025Principal software engineerMicrosoft Designer and Microsoft 365 Copilot
  • 2021–2023Principal engineering managerVideo Experiences, Microsoft
  • 2018–2021Senior software engineering managerFlipgrid, Microsoft
  • 2016–2018Front-end developerFlipgrid
  • 2015–2016Front-end developerGravie
  • 2013–2015Front-end developerValtira

Patents

Named inventor on four granted U.S. patents, three as the first-named inventor.

  • US 12,142,303Enhanced Trim Assistance for Video Editing EnvironmentsGranted 2024
  • US 11,804,247Collaborative Media Object Generation and Presentation in Improved Collaborative Workspace (continuation)Granted 2023
  • US 11,594,257Collaborative Media Object Generation and Presentation in Improved Collaborative WorkspaceGranted 2023
  • US 11,451,594Generation and Management of Data Insights to Aid Collaborative Media Object Generation Within a Collaborative WorkspaceGranted 2022

Education

  • 2009–2013B.A., EntrepreneurshipUniversity of St. Thomas, Opus College of Business