Back to Home






AI Action Management Suite  



Company: Meta
Team: Enterprise Products
Timeline: H2 ‘25
Role: Design & Strategy

As the design owner of both surfaces Intern Home and Mobile Home — and acting PM in the absence of one — I led their transformation into a connected AI productivity suite. I designed Meta's first AI voice assistant on an enterprise surface, then used UXR findings to reshape it into a daily podcast and prompt-driven platform — connecting mobile and web into one configurable system.

Impact:
  • Web app: 37.9K monthly active users, 193K+ page views
  • Podcast: 1,235 WAU, 42.53 jobs-to-be-done per 1K users — exceeding targets by 3.5×

Context:
The Product Ecosystem

I owned two surfaces in Meta's employee productivity ecosystem — at the time, neither AI-powered, both operating independently.

Intern Home (Desktop Web App):
A static employee landing page — widgets showing meetings, to-dos, quick links to internal tools. A traditional intranet homepage.



Mobile Home (Mobile iOS & Android App):
A utility app for employees on the go — calendar, people search, campus info (cafes, shuttles), tasks, notifications, and mini-apps for specific roles.




Problem:
The Engagement Gap

Employees at Meta manage their work across a fragmented landscape — chat, email, calendar, tasks, code reviews, docs — constantly switching tools to figure out what needs attention now. On mobile, this was even harder:



People opened the app. They just didn't engage meaningfully with it.



Research Insight:
The “Information Trifecta”

A foundational UXR study reframed the problem. Action management isn't a surface problem — it's an information management problem spanning three domains:




The critical finding:
"Home' is not a surface — it is a process to recenter, which involves switching between three types of information to manage actions across time."

This became the north star: one centralized configurable hub that consolidates fragmented work information and delivers them in whatever format fits the moment — not three features built in isolation.


Landscape:
The Shift Toward Voice AI



Consumer apps like Gemini Live, ChatGPT and Grok were normalizing voice-first AI on mobile — not just typing, but talking to AI on the go. Internally, teams were shifting toward AI-first experiences, and the team felt voice would be the primary gateway for on-the-go productivity.

We had a north star (consolidate the trifecta) and an emerging paradigm (voice AI on mobile). The question: can voice be the right format for consuming work information hands-free?


Phase 1:
Voice Assistant - Launch

Branding and Motion

Identity, animated states, latency-masking transitions




Entry Point and Onboarding

Discovery, first-use flow, dynamic contextual assistant




Conversation Design

End-to-end flow of initiating, sustaining, and ending a voice session


Results at 5% rollout:



3.5% adoption among exposed users. People tried it — but didn't return.

  • 🐌 Speed was the #1 complaint — "it's just unbelievably slow"
  • 🗣️ Speech recognition struggled with internal jargon
  • ✋ Users couldn't interrupt mid-response, breaking the hands-free promise


Reframing Voice

We reframed voice into two distinct modes:


  1. Voice as output — a daily podcast where the system speaks to you, delivering personalized work updates in audio form
  2. Voice as input — a "thought dumping" exercise where users speak to capture ideas and notes hands-free

We already knew from the V1 build that voice-to-text had reliability issues — speech recognition struggled with internal jargon and complex prompts. Voice-as-output didn't have this constraint: the system generates the audio, so quality is controlled. 

We tested podcast first.



Phase 1.5:
Voice Assistant UXR Research


Unmoderated diary entry study of concepts


Key findings


  1. Voice vs keyboard is situational — not a preference
    Users reserve voice for truly hands-free contexts (commuting, walking). In all other scenarios, typing wins:
    "It's definitely a socially weird thing to do, to talk to it in public... You really feel like you're from a different dimension“
  2. The "Daily Podcast" concept was the clear standout
    Users were enthusiastic about passive audio briefings — but the concept tested execution was too verbose. What users wanted was a podcast that is:
      
    • Concise: Dense, executive-level summaries — no filler
    • Customizable: Let me choose my information sources and priority
    • Interactive: Let me ask follow-up questions if something interests me
    • Multi-modal: Clean transcript alongside audio (users expected Spotify-level polish)


Design Decision on System Loop


Two insights converged: 
  • ‘Information Trifecta’ finding pointed to one centralized, configurable hub that consolidates fragmented work information. 
  • Podcast UXR confirmed users wanted to choose their own sources and priorities.

Together, this helped me define a system:

Let users configure prompts on the web app — pulling from notifications, notes, and time — and deliver them as podcast episodes on mobile. 

Phase 2:
Prompt-Driven Desktop App – Launch



AI-Generated Home

Personalized, prompt-driven updates — your trifecta at a glance



Adding Prompts

Define what matters — choose sources, set frequency, preview output



Prompt Library

Discover and reuse prompts shared by other individuals




Results at GA rollout:

Phase 3:
Podcast – Launch

Branding, Entry Point and Onboarding

Podcast identity and discovery — from first-use prompt to persistent home widget

 


Podcast Player UX

Prompt library and playback — choose a topic, listen to a personalized briefing with transcript




Playback Controls & Sources

Playback speed control, voice preferences, and linked sources for every briefing




Results at GA rollout:



* A "jobs-to-be-done per 1K users" is a measure of how many meaningful work tasks users complete through the feature. For a brand-new product with no historical baseline, this was the team's way of measuring utility rather than just adoption.

© Lulian AhnTop