Muse brings computer-use permissions to macOS as Meta races rival consumer agents.
The AMW Read
Platform expansion of an already-launched consumer agent with opt-in computer-use permissions, notable for segment-wide competitive dynamics (Instinct, Poke/Cognition) rather than a new capability paradigm.
Muse brings computer-use permissions to macOS as Meta races rival consumer agents.
Meta rolled out a Mac version of Muse, its AI assistant app, giving the agent the ability to act directly on files, messages, calendar, notes, and mail inside their native applications. The desktop release follows Muse's earlier launch on mobile and web earlier this month, when it climbed to the top of the U.S. App Store charts. Access is opt-in per data source, and Meta says Muse will always ask for approval before performing a sensitive action. CEO Mark Zuckerberg and Meta Superintelligence Labs chief Alexandr Wang both promoted the launch on social media, framing it as proof of rapid shipping.
The move sharpens an intensifying contest among consumer-facing computer-use agents. Muse is positioned against Instinct, an AI assistant reported to be raising new funding at a $10 billion valuation, and against Poke, an early breakout text-message agent that was acquired by Cognition in July. Muse and Instinct both shipped voice-calling features in the same week, suggesting the current battleground is feature parity and platform coverage rather than underlying model capability. Per the AI Market Watch index, coverage of Meta's Muse rose from zero items in the prior 90 days to six in the last 90 (coverage limited to name-matched, pipeline-ingested sources), consistent with the product's rapid platform expansion since launch.
For builders, the opt-in, approval-gated permission model Meta is standardizing may become the baseline UX that third-party agent makers must match to win desktop trust. For investors, the pairing of fast cross-platform rollout with consolidation activity like Poke's exit signals that distribution reach, not proprietary model IP, is the near-term differentiator in consumer agents — a harder bar for standalone agent startups lacking hyperscaler-level reach.
