Zoom

Zoom for ChatGPT

The partnership and the opportunity

When OpenAI opened ChatGPT to third-party apps, Zoom was selected as one of four productivity partners featured at the app directory launch. I led design end-to-end as Lead Product Designer, owning Zoom's entry into a surface where someone could ask ChatGPT about what a colleague said in last week's leadership sync and get an answer grounded in their actual meetings, then schedule a follow-up or open a recording without leaving the thread. Internally the app was framed as a strategic entry point: meeting users where they already work, showcasing AI Companion 3.0, and creating a clean handoff between the two products. The business goals were ecosystem positioning as an open platform, reach into ChatGPT users who weren't active Zoom customers, retention through daily presence in AI workflows, and differentiation from integrations built on external automation layers. I treated those goals as the design brief itself, on the reasoning that an integration that felt bolted-on would have undercut every one of them.

Framing the problem before the feature

The work followed a double-diamond process to make sure the team understood the problem, agreed on how success would be measured, and aligned on scope before committing design and engineering effort. Discovery began with a PM sync to pull shared objectives into one doc, followed by an initial jobs-to-be-done list built ahead of the scope conversation, with AI tools used for first drafts to surface edge cases and perspectives beyond my own. Granular success measures included the share of meetings successfully scheduled, the share that generated an agenda, adoption of the picture-in-picture player, and the accuracy of LLM-generated summaries. Docs, Calendar, and Meetings were prioritized as the hero use cases, with Whiteboard, Team Chat, Tasks, and Phone deferred to P2. Naming what would not be built for v1 was treated as critical to allocating meaningful effort to what would actually ship.

Personas, journeys, and the handoff insight

The central challenge was not what the app could do but who would move between ChatGPT, the embedded Zoom app, and the full Zoom client, and why. Designing the three surfaces in isolation would have produced a disjointed experience, so I built personas spanning an individual consumer, an SMB owner, an enterprise PM buried in meetings, an educator, a sales rep, and a support agent, then mapped their end-to-end journeys. Existing persona research was fed to AI as raw material for a first pass, which I then reviewed and rewrote, cutting verbosity and correcting anything that drifted from what the research actually found; the AI accelerated synthesis while judgment about what was true stayed my responsibility. For each persona I asked when the person uses only one surface, what the journey looks like as they toggle, and what unique value each surface does and does not provide. The pattern that emerged became the backbone of the design: users start where the task lives, conversational and analytical work in ChatGPT, real-time and human work in the Zoom client, with the valuable moments being the handoffs between thinking and doing.

The systems decision and the craft

The most consequential work was also the least visible in any single mockup: a clear division of labor across surfaces. The ChatGPT thread handles conversation and retrieval, returning grounded answers with citations back to the exact moment. The embedded Zoom app handles lightweight in-context actions such as booking a time, generating a doc, spinning up a whiteboard, or reviewing a recording, with just enough interface to complete the task and no more. The Zoom client remains home for live meetings, deep editing, advanced settings, and complex collaboration. The discipline was restraint: rather than porting as much of Zoom into ChatGPT as possible, I designed the embedded app to contain only what was necessary and built clear “Open in Zoom” entry points for the moment users hit the ceiling of the lightweight surface. On the craft side, I designed the scheduling flow so a multi-step transaction feels like a single conversational gesture, an in-thread post-meeting recap with a picture-in-picture player queued to the moment a topic was discussed, and a simplified calendar that surfaces surrounding commitments so users can reason about buffer and stakes rather than grabbing the first open slot. I also built a v0 tool generating evolving aurora backgrounds for loading states; when my front-end engineer flagged the WebGL output as roughly 15 to 20 times the memory and battery cost of a CSS/SVG approach, I re-authored the effect in Canvas2D and SVG filters, landing on a version that matched the visual intent at about 1MB of memory and 0.1ms of JS per frame. Other teams adopted both the generator and my Docs and Whiteboard icon animations. Prototypes started in Figma Make and moved to Claude Code for higher-fidelity builds on the design system, giving five product teams a concrete artifact to react to so a cross-product vision stayed coherent.

Constraints, shipping, and impact

Because the partnership was proprietary and top-secret, the external testing I would normally run before shipping was off the table. Rather than treat that as a blocker, I made a deliberate call to test with internal users for maximum pre-launch insight and plan consciously for reactive iteration once real usage arrived, accepting more post-launch work in exchange for hitting the partnership timeline and designing the feedback loop in advance. Interactions were honed with content and engineering through to a final prototype, handed off with mockups and flows in Figma, which engineering still prefers as a primary artifact at this level of complexity. Discoverable organically for about a month before the official announcement, the app reached 8,400 users in three weeks and 18,000 within five, with no marketing behind it, then grew past 400,000 enterprise employees paying for both Zoom Workplace and ChatGPT after the announcement. My closing takeaway is that on platform-level work in the AI era, the highest-leverage decisions come from returning to UX fundamentals: defining system boundaries, mapping true end-to-end journeys across contexts, and having the discipline to decide what not to build.

Additional projects

Add a CSAT survey to the Build flow

Although I'd created multiple feedback loops for users — chat channels, open feedback office hours, and speaking regularly with our customer support team who are in frequent contact with users and intimately know their pain points — I realized we didn't have a method for collecting feedback from users as the roadblocks happen. I wrote a ticket and PRD for adding a feedback entry point in our build flow. I iterated using pen.dev and Claude Code, adding my own iterations to explore the right pattern for our specific use case. We settled on a left-nav, bottom-aligned entry point giving users a persistent “Give feedback” CTA that was easily discoverable but did not compete with anything else in the visual hierarchy. We opted to ask open-ended questions that fed right into our Qualtrics dashboards, giving us a clear view of the most popular and most painful issues developers experienced. I built a spreadsheet of UX bugs that we prioritize against the roadmap primarily by level of effort and impact, and I made the case for dedicating 10% of resources each release to UX debt so we could continually improve the experience while still hitting the feature goals that earn new deals.

AI-first build flow

More than 90% of apps in the Zoom Marketplace are internal tools people built for themselves and their teams, yet our entire build experience was designed for professional developers looking to monetize. Building an app required roughly a 6/10 in technical skill, which excluded almost everyone. In my own time, I initiated and led a vision to change that: an AI-first build flow that lets any user describe a problem in plain language and get a working Zoom App, backed by an IDE that adapts to their skill level. I ran discovery, executed user research across three personas, designed and tested three prototypes with real users, and presented the vision to leadership. The result was a funded three-phase roadmap the organization is now executing, starting with MCP so builders get value immediately while we prepare the platform for server-less apps at scale.

Other key projects

Platform Studio: as our product area continues to evolve, we now have a new SKU for Build Platform, letting developers build apps with our SDKs, APIs, and Real-Time Media Stream capabilities. I was tasked with designing the new end-to-end surface for users to manage these features in a single place, versus the existing experience which separated each feature set across four areas of the product.

Zoom Build Hub: Zoom has become a platform where users can build all kinds of things to enhance their workflows — apps, workflows, AI agents, internal tools, and more. With many teams building end-to-end surfaces around each of these solutions, it was hard for users to know what they should build for which solution and where to go to build it, and many users didn't know they could build most of these things in our platform at all. To solve both problems, I wrote a PRD for Zoom Build Hub, transforming our existing developer documentation site into a surface that houses a super-agent that prompts users to describe the problem they're trying to solve. The super-agent tells the user the simplest way to solve it (an app, a workflow, an agent, and so on), creates a first draft of the solution, and sends the user to the surface in the Zoom platform where they can complete it. This solved the “what do I build” problem and increased discoverability for every Zoom product users can build from within our platform.

Monetization: I built monetization for app developers to earn from their apps by creating plans users can subscribe to, building an ecosystem around developers who build apps specifically for Zoom users to use from within products like Meetings or Contact Center.

App Actions: developers and users alike wanted a way to access actions for specific apps from various Zoom products — for example, creating a JIRA task from a team-chat message. I collaborated with each product team to identify the locations and actions that would be contextually relevant, providing value to the user who wanted the action and expanding the surface areas from which developers' apps could be meaningfully used.

Connectors: I built the foundations for developers to connect their third-party app data to their Zoom app via APIs with the Connector framework. After release, these features were adopted by more than 700 small businesses and unlocked a 6% increase in Monthly Active Accounts, our north star metric.

In addition to this work, I supported the team by presenting on the foundations of motion design and how to incorporate it, and created animations for icons and interactions across our product suite.