Grok+9 Mixed

Operator Insights

What are business and product leaders learning from rolling Grok out across their teams?

6 accounts: 1 failure, incident or rollback, 1 switched and 4 reported by others

Failures, incidents and rollbacks1

GitHubGitHub Status incident reportGitHub
GitHub Copilot, which serves Grok 4.6 and Grok 4.7 next to models from other labs.
GitHub Status, published Oct 24, 2026Read the report (opens GitHub Status in a new tab)
Observation
GitHub serves Grok 4.6 and 4.7 in Copilot beside other models, and an upstream incident sent their error rates up until the provider mitigated it.
Takeaway
Keep a second model one click away wherever people work in Grok: when xAI's models degrade, GitHub's answer was to switch users to another model.
What happened
GitHub's incident report says Copilot users met elevated error rates on Grok 4.6 and Grok 4.7 between 23:28 and 00:23 UTC on October 27, 2026. GitHub traced it to an upstream model provider, showed in-product warnings on the affected models, and told users to pick another model or Auto until the provider's mitigation held.
“Between 23:28 and 00:23 UTC on October 27, 2026, GitHub Copilot experienced degradation affecting Grok 4.6 and Grok 4.7 models. Users encountered elevated error rates when using these models.”
GitHub
- Public organization write-up (GitHub Status incident report)
- Incident window from GitHub's report
Deployment details
GitHub Copilot, which serves Grok 4.6 and Grok 4.7 next to models from other labs.
- Public organization write-up (GitHub Status incident report)
- Incident window from GitHub's report
Team
GitHub Copilot
Duration
23:28 to 00:23 UTC, Oct 27, 2026
Workload
Copilot chat and IDE model picker
Product version
Copilot AI model providers
Model
Grok 4.6 and Grok 4.7
Measurement
Elevated error rates between 23:28 and 00:23 UTC (Oct 27, 2026)

Switched to or from it1

Claire VoFounderChatPRD
ChatPRD's support inboxes, swept every hour by an agent named Holly Helpdesk, one of nine bots the founder runs on Grok Bot.
How I AI (ChatPRD), published Sep 3, 2026Read the write-up (opens How I AI (ChatPRD) in a new tab)
Observation
ChatPRD moved its customer support agent from OpenClaw to Grok Bot, with a person approving every Stripe refund.
Takeaway
Give a support agent the right to ask and a person the right to pay: ChatPRD's refunds go through one approval per transaction.
What happened
Claire Vo migrated her agent stack from OpenClaw to Grok Bot, including ChatPRD's support agent, which sweeps the Intercom and email inboxes every hour and handles refund and technical requests from a support playbook. The agent may request a refund but not issue one: Vo gets an Approve button that opens a single Stripe transaction for her to grant. In the first week, several customers asked where to leave a review of the support (How I AI, Sep 3, 2026).
“I migrated our ChatPRD customer support agent, Holly Helpdesk, over to Grok Bot, and the results have been astonishing. In the past week, I’ve had multiple customers ask where they can leave a review for our exceptional support. That has never happened before.”
Claire Vo
- Public practitioner write-up (How I AI episode page)
- Customer reaction reported; no metrics
Deployment details
ChatPRD's support inboxes, swept every hour by an agent named Holly Helpdesk, one of nine bots the founder runs on Grok Bot.
- Public practitioner write-up (How I AI episode page)
- Customer reaction reported; no metrics
Team
ChatPRD support
Duration
About one week at the time of writing
Workload
Hourly sweeps of Intercom and email, refunds and technical support
Product version
Grok Bot, with a Stripe approval step for refunds

Reported by others4

Eric BandholzBuilding Beardbrand's AI phone-support service, as AI Adopters Club reportsBeardbrand
A public beta of AI phone support for Beardbrand, built with Grok Bot's help and not yet listed on the contact form.
AI Adopters Club, published Oct 15, 2026Read the report (opens AI Adopters Club in a new tab)
Observation
Beardbrand's phone-support beta on xAI's voice service read out a customer's address and order history in an early test, so its founder is narrowing what the agent can see.
Takeaway
Decide what customer data a voice agent may read before it answers its first call, and verify the caller before it reads any of it.
What happened
Eric Bandholz joined Shopify order data, Recharge subscriptions, xAI's voice service, Google Workspace email follow-up and a Cloudflare-hosted middle service, using a Grok Bot agent to guide the build. In an early test the voice agent read out a customer's address, products, name and email without enough restraint, and he is moving to a flow that verifies the caller first.
“Eric described moving toward a narrower flow that verifies the caller before supplying relevant information. That is work in progress, not proof of a security review.”
Kamil Banc
- Second-hand news report (AI Adopters Club newsletter, from a conversation with Eric Bandholz)
- Setup as the founder described it; code not inspected
Deployment details
A public beta of AI phone support for Beardbrand, built with Grok Bot's help and not yet listed on the contact form.
- Second-hand news report (AI Adopters Club newsletter, from a conversation with Eric Bandholz)
- Setup as the founder described it; code not inspected
Team
Beardbrand founder
Duration
Public beta as of Oct 13, 2026
Workload
Order and subscription questions by phone
Product version
xAI voice service, with Grok Bot used during the build

GSA officialSep 25, 2025FedScoopGSA signed a delayed OneGov deal for Grok in September 2025 once its safety team found Grok 4 met its guidelines.
GSA officialAgency official, quoted by FedScoopU.S. General Services Administration
OneGov, GSA's government-wide buying program for AI tools.
FedScoop, published Sep 25, 2025Read the report (opens FedScoop in a new tab)
Observation
GSA signed a delayed OneGov deal for Grok in September 2025 once its safety team found Grok 4 met its guidelines.
Takeaway
Run your own safety test on the version you would buy: GSA signed only after its own team tested Grok 4 against the issues reported in public.
What happened
FedScoop reported on Sep 25, 2025 that xAI reached a OneGov agreement with GSA after reported delays: Wired had reported that a planned deal fell through after Grok posted antisemitic content on X. GSA's AI safety team, with data scientists from its Office of the Chief Data Officer, analyzed Grok 3 and Grok 4 and tried to reproduce the reported issues, and the agency determined Grok 4 met its guidelines. Agencies can buy Grok 4 and Grok 4 Fast for 42 cents until March 2027. FedScoop did not name any agency that bought it.
“The GSA’s AI safety team determined Grok 4 met its guidelines after testing, an agency official said.”
Miranda Nazzaro
Second-hand news report (FedScoop)
Price reported; no adoption figures
Deployment details
OneGov, GSA's government-wide buying program for AI tools.
Second-hand news report (FedScoop)
Price reported; no adoption figures
Team
GSA OneGov and its AI safety team
Duration
Agreement through March 2027
Workload
Government-wide purchasing of AI models
Product version
Grok 4 and Grok 4 Fast
Model
Grok 4, after tests of Grok 3 and Grok 4
Cameron StanleyJun 17, 2026The Hindu (AFP)The Pentagon's AI chief testified that the Grok Gov model is in use within Project Maven, its AI-assisted targeting program.
Cameron StanleyPentagon AI chiefU.S. Department of War
Project Maven, the military's AI-assisted targeting program, first powered by Anthropic's Claude.
The Hindu (AFP), published Jun 17, 2026Read the report (opens The Hindu (AFP) in a new tab)
Observation
The Pentagon's AI chief testified that the Grok Gov model is in use within Project Maven, its AI-assisted targeting program.
Takeaway
Ask for the workload and the metric behind a sworn statement of use before treating a reference customer as proof.
What happened
In a Department of Justice brief reported by AFP on Jun 17, 2026, Cameron Stanley stated under oath that Grok is in use within Project Maven, and credited the Grok Gov model with greatly increased operational efficiency. No Grok-specific measurement was published.
“To support the argument, federal prosecutors presented testimony from Pentagon AI chief Cameron Stanley in which he states, under oath, that Grok is already in use within Project Maven, the U.S. military's AI-assisted targeting programme that was initially powered by Anthropic's Claude model.”
AFP
Second-hand news report (AFP via The Hindu, from a court filing)
No metrics reported
Deployment details
Project Maven, the military's AI-assisted targeting program, first powered by Anthropic's Claude.
Second-hand news report (AFP via The Hindu, from a court filing)
No metrics reported
Team
Project Maven
Workload
AI-assisted targeting
Product version
Grok Gov model
Chief Digital and AI OfficeJul 14, 2025Breaking DefenseThe Pentagon's Chief Digital and AI Office awarded xAI a contract with a US$200 million ceiling for agentic AI workflows, alongside Anthropic, Google and OpenAI.
Chief Digital and AI OfficeContract award, reported by Breaking DefenseU.S. Department of Defense
CDAO awards to four labs for agentic AI workflows.
Breaking Defense, published Jul 14, 2025Read the report (opens Breaking Defense in a new tab)
Observation
The Pentagon's Chief Digital and AI Office awarded xAI a contract with a US$200 million ceiling for agentic AI workflows, alongside Anthropic, Google and OpenAI.
Takeaway
A contract ceiling buys the right to try: the same office funded four labs at once, so treat the award as an evaluation, and look for the deployment that follows.
What happened
On Jul 14, 2025, the CDAO split US$600 million evenly among Anthropic, Google and xAI, after a similar US$200 million award to OpenAI the month before, and xAI launched Grok for Government the same day (Breaking Defense, Jul 14, 2025). Options are not guaranteed to be exercised, and no outcome has been published.
“This morning, the Pentagon’s Chief Digital & AI Office (CDAO) announced that it would split $600 million in contracts evenly among Anthropic, Google, and xAI, following on a similar $200 million award to OpenAI announced last month.”
Sydney J. Freedberg Jr.
Second-hand news report (Breaking Defense)
Contract ceiling reported; no outcomes
Deployment details
CDAO awards to four labs for agentic AI workflows.
Second-hand news report (Breaking Defense)
Contract ceiling reported; no outcomes
Team
Chief Digital and AI Office
Duration
From Jul 14, 2025
Workload
Agentic AI workflows
Product version
Grok for Government

Failures, incidents and rollbacks1

GitHub serves Grok 4.6 and 4.7 in Copilot beside other models, and an upstream incident sent their error rates up until the provider mitigated it.

Switched to or from it1

ChatPRD moved its customer support agent from OpenClaw to Grok Bot, with a person approving every Stripe refund.

Reported by others4

Beardbrand's phone-support beta on xAI's voice service read out a customer's address and order history in an early test, so its founder is narrowing what the agent can see.

GSA signed a delayed OneGov deal for Grok in September 2025 once its safety team found Grok 4 met its guidelines.

The Pentagon's AI chief testified that the Grok Gov model is in use within Project Maven, its AI-assisted targeting program.

The Pentagon's Chief Digital and AI Office awarded xAI a contract with a US$200 million ceiling for agentic AI workflows, alongside Anthropic, Google and OpenAI.