Instinct's AI agent enters group chats, and friends without accounts can join

Instinct, the AI agent startup valued at $10 billion in its most recent funding round, is expanding into group chats. According to TechCrunch, the company announced on Monday that users can add its agent to shared conversations, where it can help with tasks like planning trips, tracking down event tickets, running fantasy leagues, organizing carpools, or dividing up who brings what to a holiday dinner.

The detail that stands out: those group chats work even when the other participants have never created an Instinct account. The agent, rather than a signup flow, becomes the way many people first encounter the product.

How the feature works

The rollout begins with Instinct's early access users and will reach everyone "soon," TechCrunch reports. Anyone who wants earlier access can ask their Instinct agent to add them to the waiting list. Once an agent is in a chat, it works for the entire group, letting people weigh options together and settle on a plan within a single thread.

The privacy model

Founder Noah Shinn explained the privacy and security design in posts on X, as relayed by TechCrunch. A user's personal Instinct agent must ask permission before connecting to the group's agent, and users decide which groups to trust, with the ability to withdraw that trust at any time. The group agent is siloed from personal accounts and cannot access them, and the personal agent always asks before sharing information or taking action. If someone new joins the chat, any pending replies from a personal agent are held back before reaching the group.

"Making plans with friends usually turns into a frustrating back-and-forth over times and places," Shinn wrote. "With Instinct in the group, you can explore options together, agree on a plan and get it done, all in one thread."

Where the race stands

The launch nudges Instinct ahead of Meta's Muse in at least one respect: according to TechCrunch, Muse does not yet offer group chat functionality, although Meta's older Meta AI assistant can already be added to group chats on WhatsApp, Messenger, Instagram and Facebook. The two rivals have been trading features in recent weeks, with both adding the ability to make phone calls. OpenAI moved into adjacent territory last week with ChatGPT Dots, personal agents that can operate inside ChatGPT Spaces, collaborative workspaces where agents and people work side by side.

The hard part is not the model

An analysis post on dev.to argues that the genuinely difficult piece of multi-user agents is deciding how much of a person's private context a shared agent may see, and that Instinct's launch is a live test of exactly that problem. Read literally, the setup ships two distinct components: a personal agent holding everything it knows about a user, and a group-scoped agent that receives only what has been explicitly released into the thread. That boundary, the post argues, is effectively the product itself, and everything downstream of it depends on the boundary holding.

The dev.to piece also cautions that handling consent per user and per action would swamp a five-person thread with approval prompts, and suggests thread-scoped grants, such as giving a chat access to your availability but not your location, with revocation visible to the whole group. In a shared setting, one person's leaked detail damages trust for everyone, the post notes, so a silent revocation reads as either a bug or a betrayal.

Why it matters

Consumer AI agents have so far been largely one-to-one products. Moving them into group chats places them in shared social contexts where the bottleneck is reaching agreement rather than generating text, and where trust decisions are made collectively rather than individually. It is also a distribution strategy: every thread containing an Instinct agent exposes the product to people who never signed up, turning each group chat into a potential funnel. If the permission boundary holds up in practice, it becomes a template for how multi-user agents ship; if it leaks, confidence in the whole category takes the hit.