What you’ll leave with
A focused Bot role, a vendor-comparison brief, and a handoff format that separates source evidence from recommendations.
Before you begin. An eligible Cursor plan or supported linked SuperGrok subscription, according to the current vendor access matrix. Verify account requirements before installing.
What the documentation establishes
The vendor documents persistent Bots that use a cloud computer and can keep working while the local app is closed. Bots under one account share files, browser sessions and logins on that computer. Source ↗
The setup guide lists desktop and mobile access, Cursor sign-in, and eligible plans or linked SuperGrok access. It also notes that Legacy Privacy Mode must be changed to a supported data setting before Grok Bot can start. Source ↗
The walkthrough below is an original worked exercise. We have not independently tested this product’s performance.
Multiple Bots are not automatically a team. If two Bots investigate the same vague question, you receive two narratives and still have to reconcile them. Start with one owned output. Add another role only when you can describe the handoff in a few lines.
This walkthrough compares three customer-support vendors for a small company. One Bot gathers evidence. You decide which facts matter. A second reviewer, if useful, checks the shortlist against the original requirements instead of rewriting the first Bot’s prose. No vendor is contacted during the exercise.
1. Create a role with a finish line
Use the official setup guide to install and sign in. Name the first Bot by its job, such as ‘Vendor research.’ Attach a short requirements file: team size, required channels, data-location requirements if relevant, and a monthly budget ceiling. If you do not know a requirement, mark it undecided instead of filling the gap with a default.
Keep the initial trial on public product pages. You do not need to connect a CRM, mailbox, or payment system to compare documentation. More access is only useful if it changes the work the Bot can complete.
Job: Vendor research for our support-tool selection.
Use the attached requirements as the decision boundary.
Return a comparison table with exact source links and a checked date.
Separate documented facts, missing facts, and your assessment.
Do not contact vendors, start trials, accept terms, or change accounts.
If pricing requires a quote, say so; do not estimate it as a fact.2. Ask for a decision-ready comparison
Choose three candidates yourself for the first trial. This removes discovery quality as a variable while you assess evidence quality. Ask the Bot to compare the same dimensions for each: eligibility, integration fit, limits, setup effort, and costs that the public source actually establishes.
A comparison should expose missing information. ‘Supports Slack’ may mean a notification integration, a complete support channel, or a marketplace plugin. Make the Bot explain the concrete action the integration supports and cite the appropriate page, rather than treating every matching logo as equivalent.
Compare [vendor A], [vendor B], and [vendor C] for our attached support workflow.
For each requirement, return supported / not supported / not established,
with a primary-source link and the exact feature or limit it establishes.
List plan-specific restrictions separately. Do not mix monthly and annual
prices. End with three questions we need answered before a trial.
Save a concise findings document and identify every unresolved assumption.3. Add a reviewer only after the contract is clear
If a second Bot is justified, give it the requirements and the first Bot’s findings, with a different task: identify unsupported comparisons and omissions. It should not simply vote on which vendor sounds best. Ask it to return disputed rows and the reason each row fails the requirement.
Because Bots share computer resources, put outputs in named folders and state which files each role may change. A second conversation does not imply a separate browser account. Avoid simultaneous tasks that require incompatible logins or editing the same spreadsheet.
| Field | Example |
|---|---|
| Input | requirements-v1.md and vendor-comparison-v1.md |
| Owned output | review/disputed-rows.md |
| Review rule | Every positive capability claim needs a source matching the required action |
| Do not do | Change the comparison, contact vendors, or start trials |
| Completion | All rows checked, with unresolved issues explicitly listed |
4. Turn a good run into a reusable routine
Run the corrected brief a second time on a different set of vendors. If the same omissions recur, improve the brief before scheduling it. A routine should specify the inputs, output location, maximum scope, and what happens when a source disappears or requires login.
Track review time as well as generation time. An assistant that produces fifty plausible rows can create more work than one that returns six carefully supported comparisons. Expand the number of Bots only when the combined result is easier to inspect than the single-Bot version.
Sources & next steps
- Grok Bot overview ↗
Cloud execution and shared-computer model.
- Official setup guide ↗
Installation, sign-in, access and data-setting requirements; checked October 10.
Guide history
Revisions preserve the original instructions. A source recheck alone does not create a new revision.
Revision 3 · Added a documented Bot creation path, complete fictional six-seat vendor packet, cost and evidence answer key, and a reproducible missing-input test.
Revision 2 · Re-read all sources; added a reusable template contract, complete/missing-input fixtures, clean handoff and transparent pilot context.
Revision 1 · Initial documented guide
Continue learning
Knowledge work · 7 min read
OpenAI dots: build a useful ongoing project brief →
Give a persistent agent a clear project, define meaningful updates, and test whether it can maintain a source-linked decision log.
Updated 2026-10-16
Techniques · 11 min read
Skills, subagents, loops, graphs: choose the mechanism your task needs →
Choose skills, workers, retries and explicit state using a worked brief-writing task, then test whether your workflow notices a missing check.
Updated 2026-10-16
Everyday assistants · 7 min read
Meta Muse: give a personal agent a job you can verify →
Check the October 12 Muse model availability, then evaluate a personal-agent task with source evidence and a deliberate memory correction.
Updated 2026-10-13