Guide

Grok Bot for Work: our templates, plus the official guide in brief

By Atlas AI Resources · Published October 19, 2026 · How we make these guides

Most office jobs fight the same three battles: too many messages, a calendar other people keep filling, and information that lands in places you never read. This page summarizes how one author covers those with three specialist Grok Bots (an Inbox Manager, a Calendar EA, and an Intel Scout) and then adds a fourth, Bot Boss, so all three report through one front door instead of pinging separately. After that come our own fill-in templates for an escalation rule, a twice-daily brief, and a front-door bot. In the guide, every send and calendar change waits for the author’s approval, and we suggest you keep it that way.

It is based on the official guide Grok Bot for Work (September 24, 2026) by Josh Kim. The workflows below are summarized in our words, with a few short quotes; read the guide for the author’s full bot descriptions.

1. From chat helper to orchestrating bots

The author writes that the shift happened quickly: a few weeks earlier, the author was still doing most of the work, directing questions to chat agents and sometimes using AI on the last mile. Now the author describes orchestrating dozens of bots, and picks out the four that changed their output at work. The guide opens with “I can’t do my job without Grok Bot.”

2. The three specialists

• Inbox Manager (communication). The hard part is not the volume of email, Slack, and DMs; it is deciding which few messages actually need a reply. This bot reads every inbound message and escalates only what is both important and urgent, about three items, along with misses such as a dropped promise or a VIP left hanging. The author keeps it read-only: “While it could respond to everything, I’ve set mine to read-only.” Its description tells it to open the full live thread before asserting anything, to treat a thread the author already replied in as closed, to stay silent when there is nothing new, and to keep any drafts labeled as drafts that never go out until the author types “send.” It runs on weekdays at 9:05 and 4:05.
• Calendar EA (time). The author books meetings next to each other to protect long focus windows, and, as a parent, has a hard stop to pick up their kids. The Calendar EA flags conflicts, focus cuts, and pickup collisions, offers named options with a default, and drafts moves and invites only after the author confirms. Its description says to start every run from a fresh calendar pull and never to write to the calendar silently. Over time it has learned the author’s quirks, such as keeping meeting rooms close together.
• Intel Scout (information). The cost of missing information is real, such as learning a day late that a decision was made in a channel you don’t read. The scout reads inbox, Slack, and meeting notes on weekdays at 8am and 5pm and delivers a short brief in three parts: company moves that affect you or your team, open follow-ups, and what you need before upcoming meetings. Each item names its source, why it matters, and one thing to do, and the description tells it to demote noise, verify before asserting, and stay quiet when nothing is new.

All three descriptions in the guide say the bot’s scheduled runs stay disabled until the author turns them on. Our advice: read the first few runs of any new routine before you let it keep a schedule.

3. Bot Boss: one front door

The three bots made the author more productive, but they also added three new things asking for attention. The fix was a fourth bot. Instead of three bots pinging about an email, an afternoon brief, or a meeting conflict, they all report to Bot Boss, which passes updates on, and fetches answers from the right bot when the author has a question. The author no longer talks to the specialists directly.

Bot Boss’s description keeps its own lane narrow: it routes and relays, checks each item before passing it on, stays quiet when nothing changed, and drafts but waits for approval on any send or calendar write. The guide describes it as a hub-and-spoke EA, “not a builder and not an auditor.” The specialists still own their jobs; Bot Boss routes the traffic between them and the author. The payoff, in the author’s words, is mental clarity: time to zoom out and find where the leverage is.

Our guide on small AI fleets and handoffs covers giving each stage a single owner, and what a one-job Grok Bot is explains why narrow bots are easier to trust.

4. Our advice: patterns worth copying from the guide

• Read-only first. The Inbox Manager could reply to everything, and doesn’t. Its job is to find the fires.
• Silence when nothing changed. Every one of the four descriptions says to stay quiet on a run with nothing new.
• Never invent. No made-up urgency, counts, or attendees; an empty search is reported as “UNVERIFIED as a search result,” not as proof that nothing is there.
• Options plus a default. The Calendar EA hands back named choices with a default, so you can accept the default or pick another option.
• A small, fixed shape. About three escalations, a three-part brief, a morning board, and an afternoon note with only what isn’t obvious. A short, predictable format is easier to read every day.
• One front door once you have several bots. Add the hub after the separate pings start to hurt, not before.

5. Atlas’s advice: three fill-in templates

What follows is Atlas’s own advice, not the guide’s. Adapt the wording to your job and your tools.

Atlas’s advice: an escalation rule for an inbox bot

• Always escalate: messages from ____ (people or roles), and anything mentioning ____ (projects, deadlines, customers).
• Escalate if waiting: a question to me with no reply after ____ hours.
• Never escalate: newsletters, automated alerts from ____, threads I already replied in.
• Cap: at most ____ items per run, most urgent first; the rest go in an FYI line.
• Each item: sender, one-line why, a link to the thread, and “unverified” if it couldn’t open the full thread.
• Drafts: allowed, labeled as drafts; nothing is sent, replied to, or reacted to without my yes.

Atlas’s advice: a twice-daily work brief

• Sources: ____ (inbox, Slack channels, meeting notes).
• Part 1, what changed: decisions or announcements that affect me or my team, each with its source link and date.
• Part 2, open follow-ups: who owes what to whom, and since when.
• Part 3, before my next meetings: one or two lines per meeting on what I should know.
• Each item: one suggested next step, or “context only” if I already own it.
• If nothing is new: say nothing.

Our prompt library has ready-made versions of nearby jobs: morning inbox sort, meeting prep card, and chief of staff prompts.

Atlas’s advice: a brief for a front-door bot

• Reports to me from: ____ (list each specialist bot by name).
• What it does: passes on updates, merges duplicates, and asks the right bot when I have a question.
• What it doesn’t do: redo a specialist’s job, change its rules, or act on its behalf.
• Morning: top priorities and anything waiting on me; afternoon: only what is new since the morning.
• When specialists disagree or a source looks wrong: show me both, with links, rather than picking one.
• Stop points: any send, calendar change, or purchase waits for my yes.

A front door works best as a relay with clear lanes behind it. If one bot starts deciding everything for every lane, you lose the reason you split the work up in the first place.

6. Keep sends and calendar writes behind your yes

In the guide, every bot drafts and waits: the Inbox Manager never sends without the author typing “send,” the Calendar EA confirms before it moves or invites, and Bot Boss waits for approval on any send or calendar write. Grok Bot’s guardrails are plain-language rules checked by a review agent, not code. The 101 guide puts them in Settings > General > Agent, with allow and block lists steering the reviewer. The current docs call this Settings → General → Auto-review: Ask first rules always stop matching actions for you, and Allow automatically rules let matching actions proceed only when the automated review does not find another reason to stop. Auto Review is model-based, so the docs say to use it alongside least privilege and explicit approval boundaries, not instead of them. An approval controls the action it is attached to; it does not undo work already done.

Our advice: a bot that reads all your email and Slack reads plenty of text written by other people, and some of it can contain instructions aimed at your bot. The Inbox Manager’s description in the guide treats screenshots as data, not instructions; apply the same rule to every message. Our guide on keeping AI agents safe from prompt injection covers this in more depth.

General architecture advice, not a description of how Grok Bot works: if you also run your own scripts against your mail or calendar, enforce hard limits such as read-only access or a send allowlist with exact checks in that code.

7. Our advice: where to start

• Pick the battle that costs you most, whether that’s messages, time, or information, and start with that one bot.
• Connect it read-only and let it report for a week or two before it drafts anything.
• Keep its schedule off at first. Run it by hand, read the output, and turn the schedule on once it is useful. Use safe inputs for test runs, since a test run does real work.
• Save each correction as a standing rule, so you don’t make the same fix twice. Our guide on skills before routines shows how to turn a good run into a saved skill and then a schedule.
• Schedule at the times you read, on weekdays, as the guide does with 9:05 and 4:05 or 8am and 5pm. See scheduling AI agent routines for cron patterns that stay cheap.
• Add the front door only after the second or third bot, when separate pings start to add up.
• Judge it by what you stop missing: fires caught early, follow-ups closed, meetings you walked into prepared, not how many messages the bots sent.

Read the author’s full bot descriptions in Grok Bot for Work.

Useful public resources

Related: Grok Bot for PMs · Grok Bot for post-sales · Grok Bot for personal use · Small AI fleets: handoffs · Skills before routines · Complete Grok Bot setup guide · Grok Bot for GTM