How to Build a Team of AI Agents with Grok Bot

TL;DR

Grok Bot lets you create specialized AI agents with separate cloud computers, shared connections, teachable skills, scheduled routines, and message-based triggers. Agents remain accessible from computers and phones, can delegate tasks to one another, and improve recurring workflows through feedback, although access requires the Cursor Ultra plan and the current trigger selection is limited.

Transcript

Okay, so we just got Grockbot, which is probably the easiest way that I found to be able to spin up teams of different AI agents that you can use from your phone on your computer. Everything stays synced and they're always on. It kind of feels like you've got Cloud Code, Codeex, and Hermes Agent all in your pocket. It's super easy to set up. So, le... Read More

Key Insights

• Grok Bot is a cloud-based workspace for creating teams of specialized AI agents that stay synced across computers and phones. Its chat interface allows users to switch between bots, communicate with each separately, and let agents exchange messages when collaboration or delegation is useful.
• Each Grok Bot agent has its own computer with access to a browser, terminal, and file manager. A user can take control of that environment, sign into websites, and preserve the authenticated session so the agent can later perform authorized browser-based work.
• The teach-a-task feature is a method for turning a recorded demonstration into a reusable agent skill. The creator demonstrated this by teaching an agent to find and like qualifying community posts, after which the agent could repeat the workflow without receiving every individual step again.
• Agent specialization is defined through a name, title, and specific description of responsibility. When an agent receives a task, it can inspect the descriptions of the other agents, select a relevant specialist, and send that agent a message requesting help with the appropriate portion of the work.
• Plugin connections are shared across the user's agents, so a service connected for one bot can be available to the rest of the team. The demonstrated integrations included GitHub, Gmail, and Google Calendar, while Slack, Google Drive, ClickUp, and other services were presented as additional connection options.
• Scheduled routines run in the cloud even when the user's computer and phone are turned off. A morning-planning agent was configured for 7 a.m. on weekdays to retrieve calendar events and actionable Gmail items, then deliver a concise daily plan inside its chat.
• Grok Bot skills can improve after execution because an agent may evaluate what worked, identify ambiguity, and revise its procedure. In the community-post example, the agent noticed that existing likes were difficult to verify and added an instruction to open each post before liking it.
• Grok Bot supports event-triggered workflows in addition to scheduled routines. The demonstrated trigger options included Slack messages, GitHub events, and Teams messages, but the creator noted that the current trigger selection was limited and expected the available choices to expand.

Questions & Answers

Q: How do you create a specialized AI agent in Grok Bot?

Create a new bot and describe what you want it to handle, such as research, writing, daily planning, engineering, or another specific responsibility. Then assign it a name, title, and precise one-line description. The description matters because other agents inspect it when deciding which specialist should receive delegated work. You can also customize the bot's color, shape, or image.

Q: How do Grok Bot agents delegate tasks to each other?

Agents use the names, titles, and descriptions assigned to the other bots to determine which teammate fits a task. When an agent encounters work that matches another bot's stated specialty, it can send that bot a message and ask for help. This allows an executive-assistant agent, for example, to delegate product-development work to an agent described as an AI engineer.

Q: How can Grok Bot turn a recorded workflow into a reusable skill?

Open an agent's computer, select the teach-a-task option, and record yourself completing the desired workflow. Grok Bot converts that demonstration into a skill the agent can run again. The example involved navigating a community and liking posts associated with a seven-day challenge. After running the skill, the agent reviewed the outcome and refined an unclear step in its procedure.

Q: How do you create a scheduled morning planner with Grok Bot?

Create an agent for morning planning, tell it to review your day, and connect the services containing your schedule and actionable messages. In the demonstration, those sources were Google Calendar and Gmail. Set the routine to run at the desired time, such as 7 a.m. on weekdays, then review and edit the generated instructions or request an immediate sample run.

Q: What information can a Grok Bot morning-planning agent provide?

A morning-planning agent can retrieve the day's calendar events and important Gmail items, then combine them into a concise plan. The sample identified meetings, messages requiring responses, a chat invitation, and preparation work for an upcoming call. It also suggested when to handle important items, warned about a crowded afternoon, and recommended protecting an evening dinner reservation.

Q: Can Grok Bot agents work when your devices are turned off?

Yes. Scheduled routines run in the cloud, so an agent can execute them even when both the user's computer and phone are off. The demonstrated morning routine was configured to check Google Calendar and Gmail every weekday at 7 a.m. and send its plan into the agent chat, removing the need to keep a local device running continuously.

Q: How do connections and context work across multiple Grok Bot agents?

Connected plugins can be shared by the user's agents, so linking a service such as GitHub for one bot can make that connection usable by other bots. Users can also connect an existing repository containing a second brain or AI operating system, giving agents established context. Without such a system, each newly created agent guides the user through onboarding questions about its intended role.

Q: What limitations should users know before adopting Grok Bot?

Trying Grok Bot requires the Cursor Ultra plan, according to the walkthrough. The available event-trigger selection was also limited at the time shown, with examples including Slack messages, GitHub events, and Teams messages. The creator expected more triggers to appear, but that expectation was presented as an assumption, so workflows should be designed around the currently available options.

Summary & Key Takeaways

• Grok Bot provides a synced interface for creating multiple specialized agents that remain available through computers and phones. Each bot can receive a name, title, description, visual identity, and focused responsibility. Clear descriptions also help other agents identify the right specialist when they need to delegate part of a larger task.

• Each agent receives its own cloud computer with a browser, terminal, and file manager. Users can retain signed-in sessions, connect repositories, and record tasks that Grok Bot converts into reusable skills. Existing second-brain or AI operating-system context can be connected, while new users can instead rely on the guided onboarding process.

• Automations can run on schedules or supported events, even when the user's devices are off. The demonstrated morning planner reads Google Calendar and Gmail, produces a daily plan, and suggests priorities. Agents can also review completed skill runs, identify unclear steps, revise their own instructions, share plugin connections, and collaborate on projects.