xAI’s Grok Bot teammate cleared 204 support tickets in 10 minutes

xAI says its support team now uses two AI teammates built on Grok Bot to handle product incidents. In an October 20 post, the company reports that one of them, IncidentSweep, cleared 204 tickets in about 10 minutes during a September 18 incident, with a support engineer approving the batch.

Two bots, one watching and one cleaning up

The first bot, IncidentWatch, monitors the support queue. According to xAI, it runs every five minutes and scans the last quarter hour of incoming tickets for clusters about one problem. When a cluster is not already linked to a known incident, it notifies the support engineer on call, with a description of the apparent issue, a count of matching tickets and sample cases. A human then decides whether it is a real incident and works with engineering. Meanwhile IncidentWatch holds the tickets that clearly match.

Once the team confirms an incident, it is posted to the public status page. Support Bot, the company’s main customer-facing bot, then goes back to the held tickets, checks them against the incident and replies to customers whose reports match. It can come back with updates as the incident develops.

The second bot, IncidentSweep, works after resolution. An engineer tells it which tickets to consider, which to exclude and the exact message to send. It finds matching tickets and leaves aside any that it cannot tie to the incident with confidence. Before anything goes out, it shows the proposed batch, the included tickets and the exact text, so the engineer can spot-check and edit. After approval it sends in small batches and logs each action. xAI says it never adds tickets after approval, and it stops and reports if something fails.

The company contrasts this with the two older options: answering tickets one by one over hours, or a bulk macro that is fast but gives little control over who receives the message.

Who can use it, and what the post leaves out

The post is an internal case study, not a product launch. IncidentWatch and IncidentSweep are bots xAI built for its own team, and the page does not say they ship as ready-made products. It does carry download and sales buttons for Grok Bot itself. The numbers come from xAI alone: one incident, one batch size and one duration, with no data on how many tickets IncidentSweep set aside, how often engineers changed a batch, or how accurate the matching was. It also does not name the Grok model behind the bots, the ticketing system they connect to, or the languages and channels covered.

Access to the underlying platform is narrower than the post suggests. According to Constellation Research, Grok Bot launched in beta in August for SuperGrok Heavy, Cursor Ultra and Cursor Teams Premium subscribers, starting at $120 a month, with enterprise plans described as coming. It runs on macOS, Windows and iOS.

What to watch as the human approval step is tested

The design keeps a person in the loop at the two points where an error is costly: deciding that an incident exists, and approving the exact message and recipient list. That is a deliberate choice, and it is the part worth tracking, because the speed gain depends on engineers spot-checking quickly rather than rubber-stamping.

Earlier xAI figures give the scale of the problem. Tesla North reported, citing xAI’s earlier post on its support deployment, a 175% jump in tickets after Cursor joined the company on August 14, and a claim that Grok Bot costs $0.20 to $0.30 per resolved ticket against $1 to $4 for traditional AI support tools. We could not check those numbers against the original post, so they should be read as company claims relayed by a third party. Our own earlier coverage shows how fast the group is growing around these tools: it covered SpaceX’s $60 billion bet on Cursor.

Signals that would confirm xAI’s account: independent customers describing similar approve-then-send workflows, published error rates for ticket matching, and a clear statement of whether these incident bots become part of the enterprise plans. Until then, the 204-ticket figure shows what one team reports for one incident, not a benchmark. The post’s closing argument is that teams should assign bots whole jobs rather than single tasks, and the company says it is looking for more workflows to hand over.