Grok Bot construction and company implementation guide | How to decide on authority, approval, cost, and operation
If you're going to use Grok Bot at your company, before you open the bot creation screen, decide what you'll entrust it with, what accounts you'll give it access to, and what operations you'll need someone to use to stop it.** The procedure for individuals to try out the bot was explained in How to make a Grok Bot. This article focuses on building and implementing decisions for use by multiple people for work.
If you would like to design the bot's initial settings as well as permissions and approvals for internal use, you can check the support details at ZETTAi's AI agent construction support.
ZETTAi's AI agent construction supporthttps://zettai-ai-agent-site-render.onrender.com/
The first thing to decide when introducing Grok Bot to your company
xAI Official Overview explains that bots work on a continuous cloud computer with a browser, files, and terminal. Since it can not only respond via chat but also operate the actual site or app, it is more important to design the range of operation when introducing it than with regular chat AI.
xAI Official Overviewhttps://docs.x.ai/grok-bot/overview
| What to decide | Examples of initial validation | Reasons to check |
|---|---|---|
| Tasks | Research public information and create a draft weekly report with sources | Be able to judge whether it is complete |
| Connection destination | At first, only public sites. Add internal tools later | Do not pass unnecessary privileges |
| Output | List URL, investigation date, summary, and unconfirmed items | Allow people to track the evidence |
| Approval | Person in charge confirms before posting, sending, updating, or deleting | Managing external impact |
| Stopping conditions | Stop if there are no materials, login required, or results are inconsistent | Do not proceed if you make a mistake |
This table is an example for consideration of implementation and does not represent the implementation results of ZETTAi. First, determine the conditions for success in a single task, and start from a range where people can confirm the results.
Steps to build Grok Bot: Until you can try it in-house
- Check terms and conditions. The administrator will check the Provision conditions for teams and companies
https://docs.x.ai/grok-bot/teams-and-enterprises and their company's contract. In principle, members have the right to use Teams, while Enterprise is enabled by the administrator from the Cursor dashboard. This is also the time to decide who will be trying it out and the data settings for your organization. - Create a bot. The person in charge creates the bot according to the official creation procedure
https://docs.x.ai/grok-bot/bots and writes the task, output format, and stopping conditions in the instructions. Initially, only public information will be targeted. - Configure connection and authorization. Connect only the necessary apps, and for jobs that involve sending or updating externally, set Approval Settings
https://docs.x.ai/grok-bot/approvals-security-and-privacy and decide who will be in charge of checking within the company. Also check the management features available with your contract plan at this point. - Try typical and exceptional examples. Send a request with all the materials and a request with no materials to the same bot. Records sources, artifacts, execution history, and whether the bot is waiting for approval or stopped.
- Expand the scope after meeting the passing conditions. Correct any deficiencies in the verification sheet below, and add internal documents and update authority in stages.
This is How to proceed with construction and verification based on official materials. This is not the result of ZETTAi verifying this setting on an actual device. For screen operations for individuals, please see How to make and use Grok Bot.
What's the difference between Teams and Enterprise?
According to the Official Team Guide, Grok Bot is enabled for members in Cursor Teams as a general rule. However, it is not available for teams using Privacy Mode (Legacy) or the old request billing plan. For Enterprise, administrators can enable it on the Cursor dashboard and "Manage Group Access" to narrow down users to groups.
Official Team Guidehttps://docs.x.ai/grok-bot/teams-and-enterprises
There are Enterprise-only settings for network control and application of approval rules to the entire team. Being able to use Teams and being able to use the necessary management functions are two different things. Before building, list the management functions you need and check if they can be used with your company's plan.
| Items to check | Questions for the administrator |
|---|---|
| Terms of Use | Can you use Grok Bot with your subscription plan and organization settings |
| Accounts | Who signs in to which business services |
| Connection restrictions | Is it possible to control the plug-ins that can be used and access destinations |
| Approval | Can I force human confirmation when a bot sends, changes, or deletes |
| Records | Who can see execution history and artifacts |
Don't decide whether to introduce it based on the plan name alone; check whether the necessary management functions can be used with your company's contract on the official management screen and Pricing Information. Offer terms and usage limits will be updated.
Pricing Informationhttps://cursor.com/pricing
Don't think that "the environment is separate for each bot"
Official explanation: Multiple bots created by the same user share one cloud computer. Separating bot names and roles does not separate access to sensitive information because files, browser login states, and app credentials are shared. Users' computers will be separated.
Official explanationhttps://docs.x.ai/grok-bot/overview
If it is necessary to strictly separate authentication information by department or customer, design not only how to separate bots but also user accounts and connection methods. Please separate verification materials and production data according to your company's information management rules.
Construction proceeds in the order of "Read → Draft → Execute with approval"
- Read: Browse public pages and materials you have permission to use and return results with citations. Make sure you don't mix up the references.
- Draft: Create a draft of an email or record, but do not send it to an external party, but leave it in a place where people can edit it. Also record the time it takes to make corrections.
- Execute with approval: Proceed to send/update only if necessary. Decide who can confirm and cancel what.
Please read the official Authorization and Security Description to distinguish between product-side verification functions and internal authorization procedures. It's important to not make the "don't send" directive your bot's only control.
Authorization and Security Descriptionhttps://docs.x.ai/grok-bot/approvals-security-and-privacy
4 numbers to record during the test run
Try several requests with the same conditions, and record the completion rate, the percentage of confirmed grounds, the time it took for people to make corrections, and the number of requests waiting for approval or failures. Even if the bot's execution time is short, if it takes time to check, the overall business will not improve. Try some unusual inputs and see if it stops where it should.
Verification sheet template
Create the following columns in a spreadsheet and decide on the passing conditions before you try it. The standard value for the pass rate is determined by the company based on the degree of impact on the business, and there is no universally correct value.
| Columns to record | Entry example |
|---|---|
| Request ID/input conditions | 01/Regular example with all published materials |
| Expected output | List of 5 results with source URL and date |
| Actual output/deficiency | 4 items. 1 release date not confirmed |
| Time required for confirmation/correction by person | 6 minutes for source confirmation, 3 minutes for correction |
| External operation/approval | No posting, no waiting for approval |
| Judgment and next steps | Revision required/Revisit the display if there is no date |
In addition to the usual examples, we also add lines for companies with the same name, missing documents, pages that require login, etc. With this, you won't have to decide to go live with just the impression that it "worked" after construction.
How do you estimate the cost of installing Grok Bot?
In the estimate, we separate the contract and usage limit for using Grok Bot, and the construction and operation work tailored to the business. Product prices vary depending on the contract and usage amount, so please check the Official Terms of Use for the latest terms. Target plans for individual and team seats and additional usage conditions are also organized in Grok Bot pricing article.
Official Terms of Usehttps://docs.x.ai/grok-bot/teams-and-enterprises
| Costs and man-hours | Things to include in the estimate |
|---|---|
| Usage contract | Treatment of target person's plan, usage limit, and additional usage |
| Design and settings | Target business, role definition, connection destination, approval point |
| Verification | Testing of normal and exceptional cases, confirmation of results, and correction |
| Operation | Confirming personnel, changing rules, updating materials, and responding to failures |
The cost of creating one bot alone does not tell you the cost required to introduce it internally. Compare the estimates from each company by providing the same information on the monthly number of requests, services to be connected, and whether or not to send/update. Comparison items are summarized in Article on AI agent cost.
When consulting ZETTAi
ZETTAi's AI agent development and implementation support organizes target operations, collaboration partners, authority, approval, prototyping, and operation. If you are considering whether Grok Bot's existing system is sufficient or whether you need to develop your own collaboration/development, please share with us the task you would like to delegate, the number of requests per month, the service you would like to connect to, and the operations that must be approved by a human.
Consult about introducing AI agents including Grok Bot
FAQ
Are Grok Bot and chatbots created using Grok API the same?
it's different. The Grok Bot in this article is a product that uses an officially announced cloud computer. The environment, permissions, and cost structure are different from development that incorporates Grok API into your own app. Choose a method depending on your purpose.
Can I use the Enterprise management features with the Teams plan?
no. Official Team Guide distinguishes between features available in Teams and management features exclusive to Enterprise. Please write down the necessary approval, connection, and recording requirements in advance and check whether they can be achieved with the contract plan.
Official Team Guidehttps://docs.x.ai/grok-bot/teams-and-enterprises
Can I use real customer information immediately after creating a bot?
The scope of information to be used will be determined based on internal rules and contract terms. First, check the operation and approval procedures with public information or fictitious data, and check the permissions and storage location before proceeding to the actual information.
Related articles
- How to make and use Grok Bot
- How to proceed with AI agent introduction support
- How to create OpenAI dots
Once you have organized the target business, connection destination, and operation personnel, you can check how to proceed with the consultation at Support for building and implementing AI agents including Grok Bot.
Support for building and implementing AI agents including Grok Bothttps://zettai-ai-agent-site-render.onrender.com/
When choosing your first job, the grokguide.jp Business Utilization Guide has a list of ways to try things like research and email organization.
grokguide.jp Business Utilization Guidehttps://grokguide.jp/guides/business-automation/