01What a Grok Bot Cloud Agent Is and How It Works

A grok bot cloud agent is a code agent that Grok Bot starts for you on a computer in the cloud, not on your laptop. With the latest update, that agent can start in a custom mode you name, and it stays in that mode for the whole job.

In Grok Bot, these cloud agents come from Cursor, the coding app a lot of builders use. When an agent starts, a card shows up in the chat, and there is an open in Cursor button right on that card.

The big idea is simple. Without a mode, an agent does whatever it feels like. With a mode, it works your way, every single time.

You write your rules once, give them a name, and then send one message to Grok Bot. The cloud agent picks up every rule inside that mode.

In this episode, Julian Goldie walks through the update, then shows five modes built around the biggest things he has made with Grok Bot. He also shows the one follow-up message that proves the mode really sticks.

02What's New: Custom Modes That Stay On

The change comes from one line in the latest Grok Bot release notes. It says bots can start cloud agents in a custom mode you name, and keep replying to them in that mode.

Here is the example from the release notes. Someone asks Grok Bot to start a cloud agent on a project called Storefront in Review Mode. A card pops up with the job, which is to review the open pull request.

The status moves from creating to running.

Then they send a follow-up asking it to also check the tests. Grok Bot replies that it sent this to the agent, and that the agent is still in Review Mode. That last reply is the whole update.

The release notes also add two more things:

  • Follow-ups to a running cloud agent now reach it right away.
  • If you tell your bot to stop, it also stops the cloud agents it started.
  • Every follow-up you send lands in the same mode as the first message.

So your fifth message gets the same care as your first one.

03What a Mode Is and Why the Old Way Was Painful

A mode is a set of rules with a name on it. Those rules cover a few key things:

  • How careful the agent should be.
  • What it should check.
  • What it is allowed to touch.
  • What the finished thing should look like.

Before this update, you could already start a cloud agent from a Grok Bot chat. The problem was that the agent did not keep your rules. So on every job, you pasted the rules in again.

You reminded it again not to touch the live site. You ended up babysitting the agent on every single task.

The new way flips that around. You start a cloud agent on your project in a named mode, like a brand mode. From then on, every follow-up you send lands in brand mode too.

You do not need to repeat yourself or remind the agent what matters.

This is the main reason the update matters. It is the gap between an agent that drifts and an agent that follows your way of working from the first message to the last.

04How to Set Up Your First Grok Bot Cloud Agent Mode

Getting started follows the same pattern Julian uses in every example. You do not need a long setup. You need clear rules and one message.

  1. Pick one kind of job you do again and again, such as reviewing work or adding to a build.
  2. Write your rules once: how careful to be, what to check, what the agent may touch, and what done looks like.
  3. Give that set of rules a short name, like Review Mode or Studio Mode.
  4. Send one message to Grok Bot asking it to start a cloud agent on your project in that mode, with the task you want done.
  5. Watch the card appear and the status move from creating to running.
  6. Send follow-ups in the same chat. Each one goes straight to the running agent and stays in your mode.
  7. If you need to, open the job in Cursor from the button on the card.

Julian suggests a Review Mode as a good first one. It matches the release notes example, where the agent reviews an open pull request and then checks the tests when asked.

05Five Grok Bot Cloud Agent Modes to Copy

Julian built five modes, one for each of his biggest Grok Bot builds. Each one has a simple launch message.

  • Studio Mode for his AI TV network. The message asks for a cloud agent on his TV channel in studio mode to add a show for tonight's reviews.
  • Jarvis Mode for his Jarvis voice assistant that talks to you. The message asks for a cloud agent on Jarvis in Jarvis mode to add a panel for today's calls.
  • Operator Mode for his full agent operating system. The message asks for a cloud agent on his agent OS in operator mode to add a page for client follow-ups.
  • Template Mode for his 31 ready-made business templates. When he wants template 32, he asks for a cloud agent on his templates in template mode to make one for a dog groomer.
  • Mythic Mode for his team of mythical bots, which he runs as a pantheon.

The pattern is always the same. You name the project, name the mode, then say the task. The mode carries all the rules, so the message itself stays short and easy to type.

06The Follow-Up Test That Proves the Mode Sticks

Near the end of the episode, Julian shows the follow-up messages that prove the mode holds. They are short, everyday requests:

  • Also check the tests.
  • Make the headline bigger.
  • Add a second show.

Every one of those follow-ups goes straight to the running agent. And each time, the agent stays in the same mode it started in. You do not have to restate your rules or start a new job.

This lines up with the release notes, which say follow-ups to a running cloud agent now reach it right away. It also means you can keep a project moving with quick notes, the way you would with a person on your team.

There is also a safety point. If you ever tell your bot to stop, it stops the cloud agents it started as well. So you are never left with agents running in the background that you forgot about.

07Common Doubts and Mistakes to Avoid

Julian answers a few doubts people may have about this update.

One doubt is whether this is only for coders. The examples in this episode are written as plain messages, like asking for a show, a panel, a page or a new template. The mode holds the rules, so the message you send can stay simple.

Another doubt is whether you really need five modes. The answer in the episode is no, you don't. Five modes is what covers every one of Julian's builds.

Most people can begin with one, and Review Mode is a natural first pick.

A few mistakes to avoid:

  • Skipping the rules. A mode is only as good as the rules you write into it, so spell out what to check and what not to touch.
  • Pasting rules into every message. That is the old way. Put them in the mode once.
  • Starting a new agent for each small change. Send a follow-up instead, since it reaches the running agent right away.

08Who the Grok Bot Cloud Agent Update Is For

This update is for anyone who builds things with Grok Bot and is tired of repeating themselves. If you keep a project going over many messages, modes keep the agent on track the whole time.

It helps most if you have repeat jobs:

  • Adding new shows or content to a channel.
  • Adding panels to an assistant like Jarvis.
  • Adding pages to an agent operating system.
  • Making new business templates from a set you already have.
  • Running a team of bots with a shared style.

It matters because consistency saves time. When the agent works your way every time, you stop checking and correcting the same things.

Julian Goldie also mentions that the full Grok Bot agent mode pack, with all six modes written out and the message that launches each one, is inside the AI Profit Boardroom. It sits alongside the systems behind every build in this episode, a roadmap for setting up Grok Bot with cloud agents, and weekly coaching calls where members set up their first mode on their own project.