Instinct's Text-Message Agent Takes On Muse and Dots
Instinct launched its AI agent in August on an invite-only basis. The company started with no marketing and barely a website, offering a text-message assistant as an alternative to app-centric copilots, helpers that live inside one app The Verge.
Instinct was founded by Noah Shinn, an early employee at Sierra before he started Instinct. The startup was valued at $10 billion in late September, after the launch of Muse. There is no app to install and no monthly subscription fee for now. Access requires an invitation from a current user or approval from a waitlist.
You talk to the agent by text over iMessage, WhatsApp, or email. It can connect to Google Workspace, Slack, and Notion. Assistants of this type finish complete digital tasks, such as booking flights, like handing off a to-do list and having it done Forbes.
Muse launched after Instinct, followed not long after by Dots. Muse is Meta's AI agent. Dots is OpenAI's AI agent. Instinct is the startup in that three-way comparison, operating without the distribution of a major platform.
Both Instinct and Meta's Muse added the ability to make calls around the same time. Meta's Muse added the ability to place calls to U.S. businesses, while Instinct's agent added the ability to make calls TechCrunch.
Instinct then extended its agent to group chats. The feature lets friends use its AI agent together, and it works even for friends without an account TechCrunch. The company points to group chats for planning trips and organizing carpools, coordination work that often gets split across threads, calendars, and direct messages.
The broader context here is architectural as much as commercial. Text as the main interface removes install and setup steps, while the hard problems shift to identity, permission scoping for Workspace and Slack connectors, reliability in using tools, and keeping a clear record of actions taken for a user. Voice calling adds another way to act, with its own failure modes. Group chat adds the task of managing shared state across people with different access levels.
In my view, Instinct is testing how little interface an agent needs if the integrations and delegation are sound. For enterprise readers, the pattern will look familiar from messaging-based operations and chatops, where work is run from chat. Users meet the system where they already coordinate, and the agent inherits both the convenience and the ambiguity of that channel. The long-arc reason for optimism is that trip planning and carpool organization are useful tests for joint planning, where an agent has to reconcile preferences rather than simply carry out a single person's command.
