How Meta designed Muse: an AI agent people actually want to talk to | Alex Cornell (Design Director) · How I AI summary

Key takeaways

Muse’s design emphasizes messaging paradigms, making the AI feel like a conversational partner rather than a traditional app.
A three-layer status system helps users understand when the agent is working and what it is doing, supporting different levels of technical familiarity.
Generative UI with consistent primitives (avatars, ideas, feed) can be dynamically personalized, increasing engagement and delight.
Muse can handle diverse personal data sources, such as Instagram saves, blogs, and chess game data, turning them into useful and entertaining formats.
Designing AI interfaces requires understanding non-deterministic systems and managing user expectations about inconsistent or variable responses.
Politeness and gentle prompting are effective strategies when the AI acts obstinate or produces undesired outputs.
There is still room to improve how users organize and differentiate between main and side chats within AI apps.
Experiments with visual and playful elements, like avatars and animated status videos, make the AI experience more personal and approachable.

Summary

Designing Muse as a Familiar Messaging Experience

In this episode, Alex Cornell explains how the design of Meta's Muse AI agent centers around making it feel like a messaging app rather than a typical AI interface. He discusses the primitives such as the bot character, status indicators, ideas, and feed, which are familiar to users. The design approach emphasizes conversational patterns over traditional features like threads, helping users interact with the agent as they would with another person. The speed of responses and the ability for the agent to work in the background also contribute to a feeling of a lively, responsive conversation, reducing frustration caused by latency.

Balancing Transparency and Simplicity in User Feedback

Alex highlights the importance of providing sufficient feedback during interactions to keep users informed without overwhelming them. The three-layer status system shows when the agent is working, what it's doing, and offers detailed logs if the user seeks more information. This progressive disclosure caters to both technical and casual users, making the experience transparent for engineers while maintaining simplicity for general audiences. Such design choices were driven by considerations of latency and user patience, ensuring users feel engaged and trust the agent’s actions.

Implementing Generative UI for Personalization and Delight

The episode explores the concept of generative UI, where core interface elements like avatars, status messages, ideas, and the feed are dynamically generated to reflect individual interactions and preferences. Alex discusses how personalization, such as a custom avatar or tailored suggestions, enhances user engagement. He shares examples like turning Instagram saves into a SkyMall-style catalog and visualizing golf performance with data and graphics. These generative elements make the experience more personal and delightful, with Alex emphasizing that the primitives remain consistent, but their content varies per user.

Use Cases Showcasing Muse’s Versatility

Alex demonstrates Muse’s ability to handle diverse tasks, from organizing Instagram saved posts into shopping recommendations, monitoring price changes, consolidating forgotten blog posts into podcasts, to generating nostalgic summaries of old memories. He details how users can leverage the agent for practical goals like tracking a golf game or improving practice sessions, as well as for fun, such as creating customized characters or exploring visual data in Pinterest. This flexibility illustrates Muse’s potential across hobbies, personal memory management, and creative workflows.

Lessons for Designers in AI Interaction and Prompting Strategies

The conversation concludes with Alex offering insights into designing for AI systems. He emphasizes understanding non-deterministic models, managing user expectations about response variability, and the importance of setting realistic design goals. Alex also shares prompting strategies for dealing with obstinate or uncooperative agents, highlighting politeness and gentle persistence. Moreover, he underscores the shift needed in mindset from pixel-perfect interfaces to flexible, adaptive outputs, encouraging designers to embrace the unique challenges and possibilities of AI-enabled products.

Show notes

Alex Cornell is a design director at Meta, where he leads design on Muse, Meta’s consumer AI agent. He came to the role with a long background in software product design, including time as a founder, and has spent the past year building one of the most thoughtful agent interfaces in the consumer AI space. Muse ships with a custom avatar, a three-layer status display, a personalized ideas surface, and a generative feed. Alex has been sharing the reasoning behind each of those design decisions publicly on Threads.

What you’ll learn:

Why Alex built Muse to feel like messaging, not an AI app
How Muse’s three-layer status system accommodates different levels of technical expertise
What generative UI actually means in a shipped consumer product
How Alex turned years of forgotten Instagram saves into a daily shopping catalog
Why his mom apologized to her Muse, and what it means for AI design
How he uses Muse to track his golf game, shot by shot, as he works toward a single-digit handicap
The mindset shift designers need when the output is never pixel-perfect
When to use the main chat versus a side chat, and where the architecture still falls short

—

Brought to you by:

WorkOS—Make your app enterprise-ready, with SSO, SCIM, RBAC, and more

—

In this episode, we cover:

(00:00) Welcome and guest intro

(02:50) Designing Muse

(05:33) Choosing which primitives to surface: characters, ideas, feed

(08:50) Speed of model and response times

(11:01) Generative UI and personalization

(14:25) SkyMall demo: Instagram saves turned into a shopping catalog

(17:16) Shopping, monitoring, and price alerts with Muse