xAI Team Bots brings shared context to Grok Bots at work

When a team keeps searching in different places for the latest customer call, product decision or brand guide, giving everyone another chat window may not solve the problem. Team Bots, announced by xAI on September 28, 2026, lets a team share a Grok Bot built around common files, instructions, skills and connected apps. In the company’s examples, it does more than answer questions: it prepares a morning Slack brief or routes defined work to other agents.

According to the announcement, Team Bots is in public beta on Teams and Enterprise plans. This is not a new Grok foundation model or a separate free chat app for everyone. The capabilities and outcomes below come from xAI’s account; we have not independently tested the product.

A shared Bot with separate conversations

A Team Bot is built around a role or workflow shared by a team. xAI says people can draw on the same expertise while their individual conversations with the Bot stay private. It says user-specific context and memories are kept separate. The Bot can also join a Slack channel where the team can see its responses. That distinction matters: messages in a shared channel are a team workflow, while one-to-one Bot conversations are private according to the company.

The announcement also says corrections can improve the Bot’s later answers. It does not explain in detail how information moves between personal and team memory, how an administrator audits that boundary, or how long the data remains. For an enterprise pilot, those questions and the read/write scope of connected apps deserve direct inspection.

What work does xAI show?

In xAI’s sales example, a Bot for a major account reviews Gong calls, Notion documents, Slack threads and company news, then posts a morning brief and role-specific drafts to the team’s Slack channel. The aim is to preserve account context when people rotate. In a customer example, Harper says it used a Team Bot to identify lapsed policies and help customers reinstate them, claiming more than $120,000 in customer savings. That is a customer statement in the launch post, not an independently verified product metric.

For engineering, the Bot connects Slack, Notion, Linear, Hex, Datadog and Cursor. It files bugs, sends defined fixes to Cursor Cloud Agents and reports progress. xAI says a five-person team used this setup while building Team Bots and shipped more than 100 PRs a day. That describes one internal workflow; it does not establish how much productivity the Bot alone added.

The marketing Bot compares drafts with brand guidelines and creates preview links for approved website changes. The data Bot, xAI says, runs read-only queries against approved Databricks tables. That is a useful boundary for analytics questions, though answer quality still depends on the table definitions and the query. Together, the examples show a product oriented around specific roles and access scopes rather than general chat.

Three boundaries to test first

First is access: when a Bot connects to Salesforce, GitHub or a warehouse, determine whose permissions it uses and whether it needs write access. Second is shared memory: check how the product separates private conversations from knowledge retained for the team. Third is final approval: customer emails, website updates and code changes need a clear human sign-off and rollback path. The announcement offers examples but does not publish a comprehensive technical assurance or independent assessment for these boundaries.

The news in Team Bots is the attempt to combine shared context, app permissions and persistent memory in one team workflow. Whether that works in practice is best judged through a real team task with its access boundaries visible during the beta.