Five Official Grok Bot Workflows to Copy

What the five guides actually show

xAI's current Grok Bot guide library is more useful as a set of operating examples than as a feature tour. Five practitioners show how they divide work between persistent bots, connect those roles to real tools, and decide what still needs a person.

The common pattern is narrow ownership. A manager bot routes work. Specialist bots inspect data, produce a defined output, or operate one part of a system. Consequential actions such as sending external messages, making purchases, deleting information, and changing production stay behind review.

GetBots compared every named role in the five guides with the live directory. Twenty-three were genuinely missing. Five GTM roles were already covered by Krista Letz's existing profiles: GTM Chief of Staff, Account Expert, Account Media Rundown, Brand Deck Keeper, and Call Follow-Up Drafter.

Eric Zakariasson: a project system for bot teams

Eric Zakariasson's experiment starts with a coordination problem. Separate Coder, Writer, and Researcher chats become hard to manage when several projects run at once. His answer is a Projects Manager connected to two Notion databases, Projects and Tasks.

The manager creates a project record and a channel, then proposes a small roster from the bots already available. It can suggest a new specialist when the roster has a real gap, but creation waits for approval. The working bots claim tasks, move their cards, and mark work blocked when they need Eric.

The useful part is the staffing rule. Reusing an existing bot keeps ownership legible and stops every project from producing another near-duplicate agent. The board and channel also give the manager a concrete place to record state instead of holding the whole project in chat memory.

• New directory profile: Projects Manager.
• Core system: one project record, one channel, one small roster.
• Human gate: approve any new specialist before the manager creates it.

Ryan Perry: six bots running a mobile game studio

Ryan Perry describes a six-seat team for Rank'em. Mobile Orchestrator owns store operations and routes work. Analytics Agent owns the evidence. Creatives Agent turns a supported finding into ad variants. Rank'em Engineer changes the app and backend. GCS Agent controls deployments and rollbacks. Bug Fix Agent reviews Sentry overnight and escalates uncertain failures.

The handoffs are the design. Analytics is the only role allowed to declare that the data supports a finding. The creative bot can make and upload variants but does not buy media. The application engineer does not bypass the infrastructure gate. The bug bot fixes clear problems and stops when the diagnosis is ambiguous.

Perry reports that this loop helped move cost per install from $15 to $1 and improved D7 retention by about four times. Those results belong to his Rank'em workflow, not to Grok Bot use in general. The repeatable lesson is that each bot has one decision right, one downstream handoff, and a clear place where it must stop.

• New profiles: Mobile Orchestrator, Analytics Agent, Creatives Agent, Rank'em Engineer, GCS Agent, and Bug Fix Agent.
• Core system: analytics finding → creative or product change → controlled deployment → measured result.
• Human gate: review media spend, uncertain bugs, and production changes.

John Bai: design agents that make ideas testable

John Bai uses four design bots with different jobs. Experiments turns early interface ideas into working prototypes. Motion God builds motion studies around real production assets. Figma Bro handles repeatable file production from existing components and measurements. Devbot answers implementation questions for the rest of the design team.

Bai's examples keep the design loop intact. Experiments built several versions of ambient Grok Bot access so he could try them during normal computer use. Motion God exposed animation controls around the real character asset, then responded to comparative feedback. Figma Bro inspected the actual Figma file instead of guessing at spacing or using the wrong logo.

This is a strong use of agents because the work becomes easier to judge. The bots increase the number and fidelity of versions a designer can see. They do not decide which idea feels right or which one belongs in the product.

• New profiles: Experiments, Motion God, Figma Bro, and Devbot.
• Core system: make a real version early, review it in context, then revise from evidence and feel.
• Human gate: the designer owns taste, product fit, and the final production decision.

Krista Letz: a GTM team that works between calls

Krista Letz's Enterprise GTM setup covers the work surrounding customer conversations. Her Chief of Staff prepares meetings, manages inbox and post-call drafts, and calls specialist bots. Other roles research prospects, monitor strategic accounts, track usage, answer technical product questions, prepare manager syncs, update forecast notes, build decks, and review calls.

Five of those jobs were already in GetBots from Letz's earlier public examples. The missing roles fill the rest of the operating loop: a Prospecting Bot, Data Analysis Agent, Product Expert, 1:1 Agent, Forecasting Bot, and Sales Coach.

The setup works because each bot reads a defined source and returns a defined work product. Prospecting ends in reviewed Gmail drafts. Forecasting ends in formatted Salesforce notes. The Product Expert answers from the codebase or an approved knowledge source. The Sales Coach uses actual Gong calls rather than generic sales advice.

• New profiles: Prospecting Bot, Data Analysis Agent, Product Expert, 1:1 Agent, Forecasting Bot, and Sales Coach.
• Already listed: GTM Chief of Staff, Account Expert, Account Media Rundown, Brand Deck Keeper, and Call Follow-Up Drafter.
• Human gate: review external messages, customer commitments, and sensitive account changes.

Kevin Niparko: a PM team built around attention

Kevin Niparko's starting point is an attention list. His Chief of Staff reviews calendar, Slack, email, and meeting context to track the projects receiving his real attention. That live view helps it filter inbox noise and expose the gap between stated priorities and where time is going.

The rest of the team is organized by decision type. Emily, the engineering manager, breaks down work and validates it but does not code. Engineering IC agents run cloud coding agents and test the results. Ashley answers repeatable warehouse questions with charts. PM Pete turns product discussions into research, RFCs, and reviews. A Recruiter searches for candidates and helps manage recruiting loops.

Niparko reports that Grok Bot accounts for a double-digit percentage of internal merged pull requests at xAI. That is a company-specific result. The part another PM can copy is the management shape: one generalist tracks attention, specialists keep separate memory, and people retain review over messages, purchases, deletes, and judgment-heavy product decisions.

• New profiles: Chief of Staff, Emily, Engineering IC Agent, Ashley, PM Pete, and Recruiter.
• Core system: live attention list → specialist delegation → checked output.
• Human gate: final review for messages, purchases, destructive actions, and product judgment.

What to copy first

Do not start by recreating all five teams. Pick one queue that already repeats and has a visible definition of done. Name the intake source, the bot that owns the job, the output it must return, and the action that still needs approval.

The Projects Manager is the best first pattern when several bots already exist and coordination is the bottleneck. If you only have one recurring job, copy a specialist instead. Give it one source, one output, and one review boundary. Add another bot after the first handoff works reliably.