
Meta's AI assistant product, which Mona worked on for about two years.
Meta AI's chats were mostly single turn, so it gave Muse's team no production data for multi-turn evals, and it is moving toward large canvas answers while Muse kept familiar chat bubbles despite internal controversy. Muse's team used agent swarms to generate multi-turn data and confirm a suspected problem needed an owner. Lesson: build your own eval data when production data doesn't fit, and favor familiar interfaces.
Controversial inside Meta, but the Muse team stuck with it
From Mona on Meta AI and Muse
Meta AI's mostly single-turn chats gave Muse no data for evals. Meta AI conversations were mostly single turn, while Muse is long-running and multi-turn. So the Muse team had no production data to build multi-turn evals from.
The Meta AI team was zero basing it when Muse began. The Muse team filled the gap by swarming agents over conversations and generating synthetic data.
Meta AI is moving toward large canvas answers, not chat bubbles. Meta AI is moving toward large canvas, essay-style answers, the direction most AI products have taken. Muse chose chat bubbles instead, so it feels like messaging a person.
The chat bubble choice for Muse was controversial inside Meta and the team stuck with it. In their words: "this should feel like a messaging service."
The team was zero basing it when Muse began. Its chats were mostly single turn, so it gave no production data for Muse's multi-turn evals, and it is moving toward large canvas answers while Muse chose chat bubbles.
Meta AI's mostly single-turn chats gave Muse no data for evals. Meta AI conversations were mostly single turn, while Muse is long-running and multi-turn. So the Muse team had no production data to build multi-turn evals from.
The Meta AI team was zero basing it when Muse began. The Muse team filled the gap by swarming agents over conversations and generating synthetic data.
Meta AI is moving toward large canvas answers, not chat bubbles. Meta AI is moving toward large canvas, essay-style answers, the direction most AI products have taken. Muse chose chat bubbles instead, so it feels like messaging a person.
The chat bubble choice for Muse was controversial inside Meta and the team stuck with it. In their words: "this should feel like a messaging service."
AI assistantsAn AI research and product company behind ChatGPT, Codex and the Dots agent.
AI assistantsOpenAI's AI assistant app used for chat, plugins and agent work.
AI assistantsAn AI company that builds the Claude models and products such as Claude chat and Cowork.
AI assistantsAnthropic's AI assistant and model family, used for chat, coding and agentic work.
AI assistantsGoogle's family of frontier AI models and the assistant app built on them.
AI assistantsA voice AI company with text-to-speech, dubbing, speech-to-text and music models, plus consumer and creator apps.
AI assistantsGoogle's AI notebook for working with your own documents and sources.
AI assistantsA voice-first AI companion app with an animated alien character that remembers your conversations.