xAI has opened Team Bots, a shared mode for its Grok Bot agents, to public beta on Teams and Enterprise plans. The pitch is simple: one configured AI teammate per team, working in Slack and the apps you already use, while every person's private chats stay private.
What xAI just launched with Grok Team Bots
xAI rolled out Team Bots on September 28, 2026, an extension of its Grok Bot agents that turns a single-user assistant into shared team infrastructure. Grok Bot first arrived in August as an always-on agent with its own cloud computer, able to sign into apps and finish multi-step jobs without supervision. Team Bots keeps that foundation and adds one thing that matters for companies: a shared setup everyone can work from.
The idea is straightforward. You build a Bot around a role or a recurring workflow, give it the files, apps, and know-how it needs, then share it. Colleagues open the same Grok Bot in their own conversations or in a shared Slack channel, and they all draw on the same instructions and skills. Anyone needing an AI coworker in Slack gets one that already knows the team.
Public beta runs on Teams and Enterprise plans. This is a business product, not something you sign up for on a personal account. It's a bet that xAI agents belong inside company workflows, not just on individual laptops, and xAI Grok agents inside Slack are the clearest sign of that direction.
The four layers inside every Grok Team Bot
According to xAI, each Team Bot bundles four kinds of operating context. Together they're what separate a shared team agent from a chatbot you message.
Layer | What it holds |
|---|---|
Context | Files, instructions, and skills such as brand guides, internal docs, and team procedures |
Plugins | Connections to apps like Salesforce, Notion, and GitHub, set per person or for the whole team |
Credentials | Secure access to third-party APIs that don't have a plugin yet |
Memories | What the Bot retains so it adapts to its assigned role over time |
Context is the layer most people undervalue. A shared bot is only as useful as the material you feed it, so a brand guide or an internal playbook does more for output quality than a clever prompt.
Plugins and credentials cover how the Bot reaches your tools. Plugins are the clean path for apps that already have a connector. Credentials are the fallback for anything without one, letting the Bot call an API directly instead of waiting for an integration that may never ship.
How Grok Team Bots keep private chats separate
The obvious worry with a shared agent is privacy. If everyone uses one bot, does your deal notes end up in a colleague's chat?
Worth asking. xAI says no.
xAI says no. Each person's direct conversations stay private, with separate context and memories for every user, while shared skills and instructions still apply across the team. So a salesperson's private notes about a deal sit in that user's context, and the account playbook stays available to the rest of the group. It's a sensible split: the knowledge is shared, the session isn't.
Slack is the other half of the story. Every Team Bot gets its own handle, which means you can invite it into a channel where anyone can ask questions, add context, and read its answers under the channel's normal visibility rules. Think of grok slack as the front door: the bot shows up where work happens rather than in yet another tab. That's what makes a Slack AI agent in a shared channel so much easier to adopt than a separate dashboard.
What xAI says Team Bots are already doing internally
xAI backs the launch with examples from its own teams. Four are worth noting because they show the range, not just the marketing.
In sales, every major account has a dedicated Bot shared by the account executive, customer success manager, solutions architect, and sales leader. Each night it reviews company news, Gong call recordings, Notion docs, and relevant Slack threads. Each morning it posts a briefing in the account channel with what changed, what each person should do next, and drafts tailored to their role. As people rotate on and off an account, the Bot becomes the record that brings newcomers up to speed.
On engineering, one Bot works out of a project's Slack channel and connects to Notion, Linear, Hex, Datadog, and Cursor. It follows decisions, triages bugs, files tickets, and launches Cloud Agents for well-defined fixes. xAI says a five-person team used this setup to steer a Cursor Project coordinating hundreds of Cloud Agents, shipping more than 100 pull requests a day and launching Team Bots itself within a few weeks.
Marketing gets a Bot that checks drafts against brand guidelines and takes approved website and SEO changes from review to preview, so regional teams can ship on their own schedule. For data, a Bot answers one-off questions from approved Databricks tables with read-only credentials, drawing on a skills library the data team built over two years across more than 45,000 tables.
One company outside xAI is named in the launch post: Harper, an insurance provider for small businesses, built a Team Bot to find customers with lapsed policies and help them reinstate coverage.
Where Team Bots still leaves questions open
The marketing is clear. The operational detail is thinner.
xAI's engineering case study credits the workflow with more than 100 pull requests a day, but the announcement doesn't publish pull-request size, merge or revert rates, review load, or defect data. Without those numbers, it's hard to compare this to a conventional team or another agent-assisted setup. Treat the figure as a company claim, not a benchmark.
A few other gaps matter for anyone weighing this for a real team. Credential scoping, secret rotation, approval flows, and audit logs aren't spelled out in the announcement. Earlier Grok Bot guidance said Bots on one account share a cloud computer with no security boundary between them, and the Team Bots post doesn't say whether the shared mode adds tenant isolation. Memory governance is also fuzzy: xAI says shared corrections improve answers for everyone, yet each user is also described as having separate memories, and it isn't clear which corrections become shared skills, config, or memory.
For teams in regulated industries, those are the questions to ask before rollout, not after.
What Grok Team Bots change for how teams work
The interesting shift here isn't the model. It's the unit of adoption.
Until now, AI agents have mostly been personal tools. One person configures a bot, learns its quirks, and keeps it to themselves. Team Bots treats the agent as shared infrastructure with a role, a memory, and an on-ramp for new colleagues. That lines up with a pattern across the industry, with Anthropic's enterprise push and OpenAI's business connectors heading the same direction, all chasing the same idea of a team AI agent.
The practical benefit is continuity. When a Bot becomes the place where a project's decisions live, hiring someone means giving them access, not a three-week handover. The risk is the mirror image: a shared bot that learned from bad context will spread that context to everyone.
For now this is a public beta aimed at companies already on Teams or Enterprise plans. If you run a team that lives in Slack and juggles recurring work, it's worth watching how xAI answers the security and governance questions before you connect it to anything sensitive. Start it with one workflow you understand well, so you can judge its output against work you could already do yourself.
