Grok Bot for Admins

Rohan describes a shift from chatbots that answer questions and agents that handle individual tasks to AI teammates that can run workflows end to end. He gives a sales example: a lead bot researches and scores inbound leads, drafts replies, prepares a brief, and brings in an account executive to approve. He also says SpaceX’s internal Support Bot helped the team scale 175% without hiring, with $2 million in salary savings.

John says Grokbot runs in the cloud, with each user’s bots working in a Linux VM; admins can restrict network destinations and choose which MCP tools are available. Team rules apply across bots, while Auto Review evaluates bot actions and sensitive actions require user approval. Inference Hooks can send prompts to a company endpoint for allow or deny decisions; John says the feature is in early access.

Admins can use startup scripts to install tools such as a company VPN or endpoint detection software, and configure secrets that bots cannot access. They can restart, recreate, or wipe VMs, set inactivity retention policies, and export bot activity to an OpenTelemetry collector. John recommends starting with a smaller pilot group, then expanding access; shared bots let a functional leader build a workflow once and make it available to a team or department.

Usage analytics categorizes tasks by team and automation level, from conversational use to workflows that run independently. Admins can track spend by team, group, and user, and set limits, including separate controls for on-demand use. In Q&A, John explains that team bots maintain shared memory for group conversations and separate personal memory for individual interactions; by default, they use the connectors of the person interacting with them.