Governance arrives after the product
xAI opened Grok Bot to enterprises on 3 September, adding access, network and audit controls so organisations can run the agents at scale. Grok and Cursor Enterprise customers get free usage for two weeks and can invite their entire organisation, including people without an existing seat.
Grok Bot went into public beta on 11 August. The pitch has not changed: these are agents that sign into applications, work on their own around the clock, and pass context between each other. “You delegate real tasks to them and they carry the job through end to end, working autonomously around the clock inside the same tools you use,” xAI said in the announcement.
What has changed is that a company can now see what they did.
What the controls actually cover
The enterprise tier adds an organisation-wide enable switch, network controls, team setup, action recording and computer management. Audit logs cover administrative, security and authentication events, and an OpenTelemetry export streams the record into a company’s own monitoring stack rather than leaving it inside the vendor’s console, according to xAI’s release notes. xAI also says each user’s work runs in its own isolated environment, separate from every other user’s.
The OpenTelemetry piece is the one that matters most and gets the least attention. An agent that acts inside a company’s systems produces evidence that belongs in the same place as everything else the company logs. A vendor dashboard is not that place: it cannot be correlated with the rest of the estate, and it disappears if the contract ends.
The customers named, and the claims made
xAI named Legora, Supermicro and ServiceTitan as organisations using Grok Bot, and said millions of bots have been created in recent weeks — a figure from the company, with no external verification. It gave sales, recruiting, marketing, finance and engineering as the working use cases, and said finance teams have used specialised bots to find tens of thousands of dollars in savings through procurement monitoring and vendor negotiation.
Those examples share a shape worth noticing. Draft outreach left for review in the morning, a candidate shortlist assembled overnight, a flag raised on a contract renewal — each ends with a human deciding. That is the version of agent autonomy enterprises are actually buying: work performed unattended, decisions still made by staff.
The competitive position
Every major lab is now selling long-running agents to the same buyer, and the differentiation has moved off capability. Anthropic, OpenAI and Google all have agent products in the enterprise; what separates them for a chief information officer is the boring surface — identity, network egress, retention, logging, and whether the audit record can leave the vendor.
The free two weeks is aimed squarely at that evaluation. Letting a customer invite an entire organisation, seats included, is how you get a pilot that produces enough log volume to be judged, and it is a cheap way to find out whether the agents survive contact with real workflows.
What to watch is renewal after the trial ends, and whether xAI publishes anything about how often bot actions are reversed by the humans reviewing them. That number, if it ever appears, would say more about whether these agents work than any capability claim.
