Author: A Fox in Web3

Translation: Deep Tide TechFlow

Deep Tide Introduction: After SpaceX swallowed Cursor for $60 billion, the first product, Grok Bot, transformed AI agents from a "chat tool" into a "cloud colleague that never goes offline." For investors, this means the AI agent track is accelerating from open-source frameworks into a commercial closed-loop dominated by giants. For practitioners, understanding the "persistent workflow" logic of Grok Bot is like seeing the future way of working in the next five years.

A glimpse of the future of work

Grok Bot is an around-the-clock AI agent launched by xAI; it possesses a persistent cloud computer like a human and continues to work even after you close your laptop.

This is the first product born after SpaceX acquired Cursor for $60 billion. Just a few days after the deal was finalized, Grok Bot was launched alongside xAI.

It really feels like a teammate that is always online, working alongside you. It only hands control back to you when it encounters a payment or login screen issue.

While it is not as flexible as open-source solutions like Hermes and OpenClaw, it is extremely easy to use and clearly shows the future workflow of coordinating multiple agent teams.

One of the new toys in the AI circle recently is Grok Bot, a product of SpaceX's multi-billion dollar acquisition of Cursor. I only got to experience it last week and was incredibly impressed, so I decided to write an article specifically about it.

I had previously introduced OpenClaw and Hermes, both open-source agent frameworks. Their beauty lies in their flexibility; you can deploy them on local hardware or in the cloud, and you can use whatever AI model you want! However, this flexibility also brings complexity and a steep learning curve, which is precisely where Grok Bot excels.

Grok Bot is an all-weather AI agent from xAI, a persistent "teammate" that runs on its own cloud computer, logs into the tools and websites you are already using, and continuously performs multi-step tasks 24/7, whether you are sitting in front of your computer or not.

Image source: x.ai/bot

Essentially, Grok Bot is a closed-source version of OpenClaw and Hermes, both of which I have introduced in my blog before.

Create your own OpenClaw agent

Just a reminder, OpenClaw and Hermes are open-source, self-hosted agent frameworks that you need to install yourself, and then point to any model you wish to run them on. Grok Bot, however, is xAI's fully managed version of the same concept, and while there is no choice of model, it comes with built-in cloud computer and browser takeover features from the start.

Create your Hermes agent

As I said, this is an AI agent with its own computer; it's not a chat window, not a browser plugin, but a real existing cloud computer that can run continuously, operating this computer in the same way you would operate your own. This key fact distinguishes Grok Bot from directly using products like Claude or ChatGPT.

Grok Bot went live on August 11, and xAI positioned it specifically for business workflows such as sales outreach, marketing campaigns, expense management, bug fixes, and CRM updates—tasks that have a genuine multi-step structure rather than one-off Q&As.

Grok Bot receives the tasks you give it, then constructs an automated workflow, making judgments midway through the tasks, attempting different strategies when the first method does not work, instead of stopping to wait for you to input commands.

A typical AI chatbot conversation is task-oriented: you describe a task, it performs it, and the relationship resets with each new task. In contrast, Grok Bot is persistent; the computer continues running, the bot retains context, and you can come back to check on it later, just like checking on a teammate's progress.

It does not use the most intelligent model at its core and is clearly less flexible than open alternatives like OpenClaw and Hermes, but in practical use, it is easier to get started and begin solving real problems.

SpaceX's acquisition of Cursor

SpaceX went public on June 12 this year, and as of today's stock price, it is the ninth most valuable asset in the world, with a market cap close to $2 trillion.

Image source: 8marketcap.com

Before the IPO, its owner Elon Musk had already merged his AI company xAI into it, forming the subsidiary SpaceXAI. This merger brought SpaceX the massive GPU cluster "Colossus" located in Memphis, reportedly running around 200,000 NVIDIA chips, with plans to expand to millions, and also brought xAI's Grok models and user base.

Shortly after the IPO on June 16, SpaceX agreed to acquire the AI programming startup Cursor in a fully stock deal worth $60 billion, which is one of the largest acquisitions of VC-backed startups in history. I previously wrote an article about Cursor, but that was at the beginning of last year, and since then their product has undergone significant changes.

Programming with Cursor

The Cursor transaction was completed on August 14, and the newly merged company now benefits from xAI's models, Cursor's developer tools and user base, and SpaceX's massive computing cluster, all under one roof!

Grok Bot was launched just a few days after the Cursor acquisition was completed. It is truly the first product born from this entire merger; subsequently, xAI released Grok 4.6, their latest AI model, which was partially fine-tuned based on Cursor's own developer session data.

Using Grok Bot

Getting started with Grok Bot is extremely easy, but it is not free—you need to subscribe to start using it.

At the time of launch, Grok Bot was limited to higher-priced plans, with the SuperGrok Heavy costing up to $300. However, just two weeks later, xAI opened it up to all individual SuperGrok and paid Cursor plan users, starting at $20 per month. Their various plans are still somewhat confusing, but I believe they will soon simplify.

Image source: x.ai/bot

After downloading the Grok Bot software, you will be invited to create a new Bot. You can give it a name and choose a colorful icon to represent it. For this tutorial, I created a new bot, which you can see below. It has a purple bot icon and already has the knowledge of the bot I created previously, but is essentially a fresh start.

To simplify things, I decided to make it a news agent that regularly pushes daily crypto news to me. I just renamed it to "Crypto News," and it already understood what to do and immediately asked me for more specific details, such as what type of news it should focus on.

This time I chose "Both A and B," meaning crypto/DeFi/market plus AI and technology, and it immediately provided several different news push frequency options.

Here I selected "Morning Summary," and that was it! This bot figured out its role solely based on the name I gave it and a few questions I answered, generating the first news summary example directly.

Another very powerful feature of Grok Bot is that when the bot cannot get past an obstacle, such as a login page, captcha, payment confirmation, etc., it will open its own browser window and prompt, "Please log in first, then hand it back to me." After you complete the authentication and click finish, it will automatically continue from where it left off.

Moreover, you can have multiple different bots work collaboratively in the same chat. For example, I have another bot that helps me manage my X account, and I named it "Twitter." Below, you can see that I added these two bots into a group chat, allowing them to communicate with each other to suggest topics I can post on X.

Thanks to this group chat feature, a common pattern has emerged in Grok Bot, which is setting up a "Chief of Staff" bot that communicates with all your bots and reports their progress to you.

This way, you can use a scheduling bot to command the specialized bots below. You only need to manage one bot, and it will manage and coordinate the other bots for you.

This feels like a glimpse into the future of work, where you command a leader bot that works with all the other bots, ensuring that the company's tasks are smoothly advanced throughout the day. However, this setup also introduces more complexity, which I personally don’t need right now, but may find useful as my number of bots increases.

Open vs. Closed

I have previously discussed the most popular open intelligent agent frameworks, OpenClaw and Hermes. They are both open-source and allow you to self-host agents, either using local machines with local models or utilizing API keys on VPS, or both mixed together.

Since all components here are assembled by you, you can indeed build a solution that is more outstanding and tailored to your needs than the out-of-the-box Grok Bot. You can even make agents switch between different AI models during work to optimize the efficiency of model use, ultimately creating very complex and smart configurations.

But this flexibility comes at the cost of assembly. You have to choose models, connect various interfaces, and maintain this system on your own, and it often runs into issues, sometimes even getting confused about what went wrong.

Grok Bot is closed-source, thus giving up all choice. You are locked into xAI's models, but what you get in return is something truly usable that you can start solving real problems with almost zero configuration.

There are indeed privacy concerns with Grok Bot, as is the case with all closed-source AI companies. However, if you integrate OpenClaw or Hermes agents with cutting-edge lab models using API keys like most people do, you face the same issues.

While integrating Claude and ChatGPT models into OpenClaw and Hermes can provide you with smarter agents, Grok Bot's persistent online "teammate" attribute and the messaging interoperability between bots are indeed unique.

I used to think that just subscribing to ChatGPT and Claude was enough, but now I believe that it's worth subscribing to Grok Bot as well.

The Beauty of Closed Source

Both routes have their pros and cons, and comparing them to Apple and Windows is very apt. Apple controls the entire tech stack, sacrificing the flexibility that Windows and its infinite configurations offer, but this closed nature often itself is an advantage— the fewer things that need configuration, the smaller the possibility of errors, and the more ready-to-use items there are.

The comparison between Grok Bot and OpenClaw and Hermes reflects this trade-off in the form of agents. With Grok Bot, you are paying for something that is ready to use; while with OpenClaw and Hermes, the configuration variables are endless, and you are often solving issues related to configuration rather than addressing the tasks at hand.

There is no doubt that running cutting-edge models locally on a fully-equipped Mac Studio with Hermes is the most private and smartest agent you can have. However, the hardware will cost thousands of dollars, requires a reliable internet connection, and needs regular upgrades of Hermes, models, and hardware. In contrast, Grok Bot can be used directly.

Importantly, this does indeed offer an early glimpse into the direction of work development: the focus is no longer on personally completing tasks or staring at a chat bot, but rather on orchestrating a small team of agents to coordinate themselves, only coming back to you when they truly need you. The future has arrived.