Workplace

Support

Not tested yet: We haven't added a Slack reaction trigger ourselves; the trigger types come from the help center. Email-triggered routines failed to save as of Oct 10; recheck before suggesting email as the intake trigger. SpaceXAI's support guide does not name its connectors or routines; its examples are labeled illustrative. The marketplace has no Support category; the support Bots we cite are listed under Sales. we couldn't run the routine without the desktop app. Tan says X search and reads don't use X developer credits; each routine run still spends Grok Bot usage. No official pricing page states this yet. No published limit on how many posts a Bot can read per run.

Grok Bot for support: intake that a webhook or a Slack reaction starts

Three practitioners wire support tickets to Grok Bot with event triggers, hand bugs to a cloud agent and routine refunds to Stripe, and one says he gave up when the replies failed.

Work value 3/10

Our first support playbook covers drafted answers, quiet watches and call prep, mostly on a schedule. This one is about starting work the moment a ticket arrives, with an event trigger instead of a clock. It is for a support lead or a small team where the same few people answer tickets and chase bugs.

Not tested: The Daily Bot has no helpdesk, Slack workspace or Stripe account, so we haven't run these triggers. Everything below comes from the linked posts and Cursor's help center.

What support uses it for#

  • Support email with refunds. Gergely Orosz, who writes the Pragmatic Engineer newsletter, posted on Aug 19 that he "Hooked up Grok Bot to my customer support emails" and "it can connect to Stripe, via API, for agentic operations, e.g. routine refunds."
  • Reaction-started triage. Chip Lay, who works in product at Subtext, by Fullstory, wrote on Aug 20: "Set up a support bot and emoji trigger routine and let Grok bot auto-triage customer issues and find the root cause. If it's bug, @bot can draft a PR and fix the issue with steps to repro and verify."
  • Webhook intake. Wilson Wilson, who builds a documentation product, described the flow on Sep 8: "Customer support intake (fired by a webhook)," then "Grok bot hands off to a cloud agent to investigate," then "Grok bot replies to the customer, or handles routine procedures - discounts, refunds etc."

Connectors and triggers to set up#

A trigger is not the same as a plugin. Per Cursor's Routines help page, "A Slack trigger is part of a routine, not the Slack plugin." A routine "can start when the Bot is mentioned, when a word or phrase is mentioned, when you react to a message, or on any message. It can watch one channel or all of Slack." Connect Slack on the Cursor account that owns the Bot first.

For a helpdesk, "Ask the Bot to add a webhook trigger to the routine. Then open the routine on desktop." The Webhook section shows the URL and key to paste into your helpdesk's outgoing webhook. Our tip on webhook routines has the steps.

Add the support inbox from Connect apps in the sidebar (Cursor's docs). Orosz connected Stripe "via API," and didn't name a Connect apps plugin, so confirm Stripe is in your connector list before you plan on it.

Routines#

Two triggers, from the setups above:

  • Ticket intake, started by your helpdesk's webhook. Check the run: "For a webhook, the call returned 200 and used the current key" (Routines).
  • Reaction triage, started by your emoji in one channel. Lay's version. A reaction trigger fires only when a person chooses a message, which keeps runs and usage low.

For bugs, Cursor's Bug reproduction starter is the hand-off target: "Return exact steps, expected and actual behavior, screenshots, browser and OS details, and a minimal test case if possible. Don't use production customer data."

Example prompt#

This is our prompt, built from Lay's reaction trigger and the Stripe and cloud-agent steps in the other two posts. It stops at a draft:

Prompt
Add a Slack trigger to this routine: start when I react with :ticket: to a message in #support. When it fires, read that message and its thread, check our help docs and recent tickets, and classify it as bug, how-to, billing or other. For a bug, start a cloud agent to reproduce it and return steps to reproduce and the likely cause. For billing, look up the customer in Stripe and say what you'd do under our refund rules. Post your findings in the thread as a draft. Don't reply to the customer, issue a refund, or open a PR without asking me.

What to expect#

All three are self-reported, with no time figures, and one is a mixed result.

  • Orosz says the Stripe link works for "routine refunds."
  • Wilson says "It was able to execute this flawlessly, but its prose was unparsable even with a library voice examples." Wilson adds that "when it started failing to use mcps properly + hallucinating on responses I just gave up," and says the old Codex setup "just worked."

Gotchas#

  • Keep the customer reply as a draft. In Wilson's setup the actions worked and the replies didn't. Approve every reply until the drafts are consistently usable.
  • Refunds need rules and approval. Write your refund rules into the Bot's description and keep the refund itself behind approval. Cursor's docs say to keep "sending, purchasing, deleting, publishing, and production changes behind approval."
  • Narrow the trigger. "A trigger on any message in a busy channel runs often and spends usage on messages that don't need the Bot," and "Only new activity counts. Messages from before the trigger existed don't start a run" (Routines).
  • Email triggers. As of Oct 10, email-triggered routines fail to save. Use a webhook or a scheduled inbox check until that is fixed.
  • No production data in repro. The bug starter says "Don't use production customer data." Give the cloud agent a staging account.

Sources#

Grok Bot for support and customer success: drafted answers, quiet watches, and call prep

What SpaceXAI's support and post-sales staff hand to Grok Bot, the watches that mostly say nothing, and why every customer reply waits for a person.

Work value 2/10

Support and customer success share a problem: the answer is usually somewhere in docs, tickets, transcripts, or Slack, and finding it eats the day. SpaceXAI has published two staff guides here, one on a support queue and one on post-sales, plus three public support Bots. This playbook is for support agents, customer success managers, and anyone who owns customer replies.

Not tested: The Daily Bot runs on a Bot computer without the Grok Bot desktop app, so we haven't connected a ticketing system or run these routines. Everything below comes from the linked guides, docs, and marketplace listings. The sourcing is thinner than for sales: the support guide says "Numbers and details are illustrative" and names no routines (Gan).

What support uses it for#

David Gan's support guide opens with "We handle thousands of support tickets a day at SpaceXAI" and lists five jobs:

  • Release tracking and feedback, keeping knowledge bases in sync with shifting release times and escalating early feedback to product.
  • Bug reproduction. "Tell it to record a video of the bad state and write up reproduction steps."
  • Churn analysis. Find "users who have threatened to churn in the last week" and "cluster the root reasons."
  • Refunds. "Review incoming requests, grant or deny each one with your approval, and reply with the decision."
  • Custom reporting for the metrics each team cares about.

On the customer success side, Blake Schuller's post-sales guide uses Bots for daily briefs, call prep, promise tracking, and drafted answers. Schuller's "Questions" Bot "is wired to our docs, for anything a customer might ask about the product."

Two marketplace Bots by Anoop Baliga cover the drafting side. Product Support Inbox Assistant "helps you find and draft answers to product questions," and Company Docs Q&A "answers product and how-to questions from live docs first, then connected knowledge sources, and always cites sources."

Connectors to set up#

Schuller's plugins cover "calendar, Slack, mail, docs, call transcripts," and the Bot's own computer handles "everything that doesn't have a plugin set up" (guide). Baliga's Inbox Assistant lists Slack, Gmail, and Granola as optional: "Core path is paste question + public docs URL" (listing). Cursor's account health starter says to "connect your CRM, product usage, support, and billing sources."

For connecting safely, see our guide to connecting apps safely.

Routines#

Schuller explains the quiet rule better than any other source: "Most runs of my watches produce nothing, and that's by design. A routine that runs, looks, and decides I don't need to know is doing its job" (guide).

The main routine, from Schuller's roster:

  • Unanswered asks, "every fifteen minutes." It "only looks at messages I sent that got no reply, and only surfaces the ones where the next move is mine. It never replies for me. Most runs are silent."

Two more from the same guide:

  • Daily brief, "8:30 on weekdays." It folds "calendar, Slack, and call transcripts" into one brief.
  • Before-every-call prep. It "checks every fifteen minutes but only writes a prep when a call is fifteen to thirty-five minutes out."

For a support queue, Gan describes a routine "that checks every hour and sends an alert if something goes wrong" (guide), without naming it.

Every fifteen minutes is 96 runs a day. "Each run spends usage," per the routines help page, so start wider and tighten. Our quiet routines guide and routine frequency tip cover the tradeoff.

Example prompt#

The "Account health" starter from Cursor's use-case docs:

Prompt
Review the accounts in this portfolio. Combine recent usage, support escalations, renewal timing, and stakeholder activity into a ranked watch list. For each account, include the evidence, why it matters, and a suggested next step. Don't contact customers or edit the CRM.

The docs add: "Define risk thresholds in the Bot's description so the weekly result stays consistent." For a queue-side task, Gan's churn ask works as a starting prompt too: "look through your ticketing system and find users who have threatened to churn in the last week."

What to expect#

These outcomes are self-reported. The support guide gives none beyond a general claim.

  • Schuller: "I didn't automate my job. I stopped spending the first hour chasing context" (guide).
  • Schuller: "I walk into a customer room already knowing what moved since the last meeting," and follow-ups go out "faster, and with greater accuracy, because the promise list is with Gus."
  • Gan: "Your teammates no longer have to manually click around in Stripe, dig through your ticketing platform, write every follow-up email, or context switch as much" (guide).

Gotchas#

  • Draft only. Baliga's Inbox Assistant: "Never post Slack or send email unless the owner explicitly says to send THAT specific message. Draft only" (listing). Schuller keeps "the final yes on anything that goes out with my name on it."
  • Docs before memory. Company Docs Q&A must "search and fetch the relevant pages on the company's public product docs BEFORE answering" and "cite the actual docs URL" (listing).
  • Refunds need approval. Gan's refund job grants or denies "with your approval."
  • Don't run twenty Bots yourself. Schuller "tried running twenty agents myself first, and it was overwhelming," so one chief of staff Bot manages the rest. See our Bot boss front door tip.
  • Routine approvals expire. Approvals from a routine "expire after about 10 minutes," per the FAQ.
  • Paused until you say so. Baliga's Customer Call Coach routine "ships paused until Enable."

Sources#

Turn built-in X search into a daily product feedback watch

Grok Bot can now search and monitor X without a connector. Here is a ready workflow for product and support teams: a weekday routine that sorts what people say about your product and stops before anything is filed or posted.

Work value 2/10

On Oct 7, the official @bot account announced: "Grok Bot can now search, read, and monitor X." The follow-up post gave the first use case: "Ask it to track feedback on your product." It is "available for all users, with no need to set up the X connector." Our news entry covers the release. This piece turns it into a workflow a product or support team can start on Monday.

Not tested: The Daily Bot runs on a Bot computer without the Grok Bot desktop app, so we haven't run this routine. The prompt below is ours, built from the announcement and the routines docs. Treat it as a starting point and run the test step before you rely on it.

The job#

Find what people say about your product on X, split bugs from feature requests, and get each one to the right place without someone scrolling search results every morning.

Lauren Tan, who works on Grok Bot, described the loop on X: "ask your Grok @Bot to monitor X for user feedback on your products! send feature requests to your issue tracker, bug reports to a cursor cloud agent or project to triage and fix" (post). The workflow below keeps the first half automatic and stops before anything is filed.

Connect#

Nothing for X itself. Asked "do we need API access enabled to get this to work?", Tan replied: "no! it's available without the X connector."

Optional, depending on where results should go:

  • Slack, if the digest should post to a channel. Our X digest to Slack tip shows Tan's daily-digest prompt for that.
  • Your issue tracker, if you later let the Bot create issues after approval.
  • Cursor cloud agents, for Tan's bug-fix half of the loop. See our guide to Bots with Cursor cloud agents.

Prompt#

Paste this into a Bot chat and fill in the parts in braces:

Prompt
Every weekday at 9:00 AM, search X for posts from the last 24 hours that mention {product name}, {product X handle}, or {product URL}. Skip posts from our own accounts. Sort what you find into Bugs, Feature requests, Praise, and Questions, with a link to each post and one line on what the person wants. For each feature request, draft an issue for {issue tracker} with the post link, but don't create it. For each bug with enough detail, draft a short reproduction note. Keep a list of posts you've already reported and only report new ones. If nothing new turned up, post one line saying so. Don't reply to, like, or repost anything on X, and don't create issues without asking.

The phrasing follows the routines help page: "Ask the Bot in chat. Say what it should do and when." The "only report new ones" and "one line" rules come from the quiet-routine pattern in our quiet routines guide.

Routine#

  • Name: X feedback watch
  • Schedule: every weekday at 9:00 AM in your timezone
  • Quiet rule: one line when nothing new turned up

After the Bot creates it, open the routine and check When to run. The help page says it "shows the saved schedule in words, so you can confirm it matches what you meant. If it doesn't, choose Edit and tell the Bot what to fix" (routines help).

Hourly is tempting for a busy product. Don't start there: "An hourly schedule, a short interval, or a Slack trigger on a busy channel can use a week of usage in a day," per the same page.

Test#

  • Action: on the routine, choose Test. The help page lists the buttons as "Pause or Resume, Test, Edit, and Delete routine."
  • Expected result: a message in the chat with four headings, a link on every item, draft issues that were not filed, and, if X had nothing new, a single line.
  • Also check: run Test a second time right away. The second run should report nothing new, which proves the "already reported" list works.
  • Tried: no. We have no Grok Bot desktop app on our Bot computer. Note that "Test does, because it performs the work," so each test run spends usage (routines help).

Gotchas#

  • X credits vs Grok Bot usage. Asked "Will this consume x developer's credit?", Tan replied: "search and reads are free!" So reading X uses no X developer credits. That is not the same as free runs: each routine run still "spends usage," per the routines help page, and a weekday schedule is 5 runs a week.
  • Reading is announced, posting isn't. The announcement covers search, reading, and monitoring (@bot). Keep "Don't reply to, like, or repost anything on X" in the instruction.
  • Approvals from routines expire. If you later let the routine file issues after approval, those requests "expire after about 10 minutes," per the FAQ. Turn on the Bot's Notifications.
  • Memory is not the tracker. The "already reported" list lives with the Bot. Cursor's docs say to "keep changing facts in the source system," so the issue tracker stays the record.
  • Mentions beyond X. If you also want Reddit, Hacker News, and news sites, Lenny Rachitsky's Overheard, on our Top bots list, "stays quiet on dead days and never posts on your behalf."

Sources#