xAI and Cursor Deliver Grok Bot Cloud Computer to Rival ChatGPT

Peter Yang

Grok Bot introduces a persistent cloud computer operating system that replaces manual browser tabs with voice-driven agents, though its steep two hundred dollar monthly price tag prevents it from replacing ChatGPT as a daily driver.

AI-assisted summary of Grok Bot: 5 Must-Try Use Cases for Work and Life (Full Tutorial) — original source. Check the original for full context.

Source published: 2026-08-17

Key ideas

Grok Bot runs autonomous agents inside persistent cloud computers hosted on SpaceX/xAI infrastructure, eliminating the session disconnects of ChatGPT Work and the hardware maintenance required by local setups like Hermes. (1:38)
By understanding holistic travel itineraries rather than monitoring rigid routes, Grok Bot identified an alternative round-trip flight booking to Japan that saved approximately $2,700 compared to an open-jaw plan. (15:09)
Equipped with a full virtual desktop operating system, Grok Bot can autonomously download, install, and execute native software like retro PC games, though virtual mouse drivers and input latency currently hinder real-time playability. (17:26)
Despite a superior and less cluttered UI than ChatGPT, Grok Bot's steep $200 per month cost for heavy use and user hesitation around entering primary credentials into a cloud OS remain major barriers to replacing daily driver assistants. (21:27)

Detailed notes

## Background Published on YouTube by creator and founder Peter Yang on August 17, 2026, this tutorial explores Grok Bot as the emerging future of personal AI agents. According to the video description, the guide covers five practical bot workflows—including an orchestrating advisor, a YouTube researcher, an X scout, a digital Marie Kondo email and subscription cleaner, and a travel concierge—alongside retro gaming tests and an evaluation of privacy, pricing, and whether it can replace ChatGPT.

## TL;DR Peter Yang demonstrates the practical power of Grok Bot, emphasizing its persistent cloud computer architecture on SpaceX/xAI servers compared to local setups like Hermes or fragmented tools like ChatGPT Work. Through a core three-step prompting workflow, he builds autonomous agents to curate niche content, declutter inboxes and cancel unused subscriptions, and structurally optimize flight itineraries to save thousands of dollars. After testing its full operating system capabilities by running retro DOS games, he concludes with an honest critique of user trust barriers regarding remote credentials and notes that a steep $200 monthly price tag keeps it from immediately replacing ChatGPT as his everyday driver.

## Atmosphere & style The presentation carries the polished, energetic tone of an experienced product builder delivering a hands-on live demo to fellow power users. Yang balances genuine, geeky excitement—laughing at retro gaming quirks and delighting in custom UI coworker animations—with sharp, pragmatic skepticism about cloud security and pricing models. Watching it feels like looking over the shoulder of an early adopter who is eager to push frontier technology to its limits while keeping viewers grounded in real-world utility.

## Bulletin - [01:38](https://www.youtube.com/watch?v=MkVcHbviYOw&t=98s) Grok Bot runs autonomous agents inside persistent cloud computers hosted on SpaceX/xAI infrastructure, eliminating the session disconnects of ChatGPT Work and the hardware maintenance required by local setups like Hermes. - [15:09](https://www.youtube.com/watch?v=MkVcHbviYOw&t=909s) By understanding holistic travel itineraries rather than monitoring rigid routes, Grok Bot identified an alternative round-trip flight booking to Japan that saved approximately $2,700 compared to an open-jaw plan. - [17:26](https://www.youtube.com/watch?v=MkVcHbviYOw&t=1046s) Equipped with a full virtual desktop operating system, Grok Bot can autonomously download, install, and execute native software like retro PC games, though virtual mouse drivers and input latency currently hinder real-time playability. - [21:27](https://www.youtube.com/watch?v=MkVcHbviYOw&t=1287s) Despite a superior and less cluttered UI than ChatGPT, Grok Bot's steep $200 per month cost for heavy use and user hesitation around entering primary credentials into a cloud OS remain major barriers to replacing daily driver assistants.

## Key takeaways - [00:00](https://www.youtube.com/watch?v=MkVcHbviYOw&t=0s) Grok Bot shifts personal computing to persistent cloud machines on SpaceX/xAI servers, avoiding the need to run local hardware 24/7 or suffer browser session logouts. - [01:38](https://www.youtube.com/watch?v=MkVcHbviYOw&t=98s) Compared to ChatGPT Work's fragmented interface across Chat, Work, and Codex tabs, Grok Bot offers a unified coworker-style UX designed with distinct animated avatars. - [03:27](https://www.youtube.com/watch?v=MkVcHbviYOw&t=207s) A meta-agent Advisor can be used to orchestrate an entire personal AI ecosystem by generating, configuring, and combining specialized sub-bots. - [04:42](https://www.youtube.com/watch?v=MkVcHbviYOw&t=282s) Effective agent orchestration relies on a three-step cycle of initial prompting, iterative output formatting, and scheduled automated execution rather than one-shot prompting. - [07:12](https://www.youtube.com/watch?v=MkVcHbviYOw&t=432s) First-party integration with X allows Grok Bot to aggregate niche viral trends and bookmarks, summarize them into actionable ideas, and email digests directly upon voice command. - [10:26](https://www.youtube.com/watch?v=MkVcHbviYOw&t=626s) Implementing strict guardrails and numbered approval lists enables an autonomous agent to safely audit inboxes and cancel bank-linked subscriptions in roughly five minutes. - [15:09](https://www.youtube.com/watch?v=MkVcHbviYOw&t=909s) Context-aware agents outperform traditional price trackers by analyzing entire itinerary documents to uncover structural routing optimizations that save thousands of dollars. - [17:26](https://www.youtube.com/watch?v=MkVcHbviYOw&t=1046s) Having an operating system inside the agent's environment allows it to install native desktop software and retro games, signaling potential for cloud-streamed software management. - [20:20](https://www.youtube.com/watch?v=MkVcHbviYOw&t=1220s) Entering sensitive credentials into a remote virtual OS triggers psychological resistance, making user trust the critical bottleneck for mainstream cloud agent adoption. - [20:20](https://www.youtube.com/watch?v=MkVcHbviYOw&t=1220s) Grok Bot addresses privacy by adopting Cursor SSO authentication, encrypting sessions in transit and at rest, and excluding user cloud activity from model training. - [21:27](https://www.youtube.com/watch?v=MkVcHbviYOw&t=1287s) A steep price gap—$200 per month for heavy Grok Bot usage versus $20 per month for ChatGPT—prevents it from immediately displacing existing mass-market AI subscriptions. - [21:27](https://www.youtube.com/watch?v=MkVcHbviYOw&t=1287s) Cursor's model-agnostic strategy positions it as an agile front-runner in personal computing agents against closed ecosystems from OpenAI and Anthropic.

## Detailed walkthrough ### [00:00](https://www.youtube.com/watch?v=MkVcHbviYOw&t=0s) Introduction and the Cloud Computer Advantage

Peter Yang opens by declaring that "Grok Bot is the future of personal agents" and previews the five core bots he will build in this tutorial: an advisor to create and manage other bots, a YouTube researcher to spot outlier videos, an X Scout to surface insightful and funny posts, a digital Marie Kondo to clean up email and cancel paid subscriptions, and a personal travel concierge to hunt down cheaper vacation prices. As a fun bonus, he also plans to build a gamer bot to test whether Grok Bot can play classic titles like *Doom* and *Red Alert*, before wrapping up with an honest evaluation of privacy and whether Grok Bot can serve as a new daily driver.

The backstory of how Peter got his hands on Grok Bot began a few weeks prior, when he tweeted that ChatGPT was just one feature away from building his dream AI product: "a dedicated personal computer in the cloud for running AI agents." Shortly after that post, Lee from the Cursor team sent him a DM inviting him to test Grok Bot. Peter jokes about the well-known meme of people walking around with their laptops half-open just to keep their local agents running. With Grok Bot, that hassle disappears because the bot lives inside a persistent computer sitting on SpaceX/xAI servers rather than in your home, complete with its own operating system and a browser where you can stay logged into your favorite applications.

[01:38](https://www.youtube.com/watch?v=MkVcHbviYOw&t=98s) Comparing Hermes, ChatGPT Work, and Grok Bot

To illustrate why a dedicated cloud computer matters, Peter compares his experiences across Hermes, ChatGPT, and Grok Bot. He currently runs Hermes 24/7 on a Mac Mini at home; while that provides a persistent machine, it requires buying physical hardware and configuring everything yourself (or doing heavy setup work on a virtual private server). Meanwhile, ChatGPT Work relies on plugins and a cloud browser, which suffers from two major drawbacks: the cloud browser currently cannot stay signed in to your favorite apps, and the user interface feels fractured across Chat, Work, and Codex tabs. Displaying his screen, Peter notes that the interface is "just kind of like a mess of chat threads" that is confusing even for a power user like him.

In contrast, Grok Bot provides a persistent cloud computer right out of the box and is by far the easiest of the three to set up. Its interface feels focused and delightful, featuring small bot avatars that play subtle, charming animations when they start working—such as when looking up the weather. Each bot has its own distinct personality and visual style, which Peter credits to Cursor's design team, specifically calling out Jenny Wen and her colleagues whom he has interviewed in the past. Overall, he remarks that the UX "feels more like talking to a coworker than getting lost in a hundred chat threads," setting up the central test of whether it can dethrone his current daily driver combination of ChatGPT and Codex.

[03:27](https://www.youtube.com/watch?v=MkVcHbviYOw&t=207s) Bot #1: The Advisor Bot (Orchestrating Other Bots)

The very first bot anyone should create after installing Grok Bot is an Advisor, which Peter keeps pinned to the top of his bot list. The strategy is to tell the Advisor about your work and personal life, and then ask it to brainstorm other bots to build. Peter shares the exact prompt he fed his Advisor: he introduced himself as a creator and founder of Behind the Craft living in the Bay Area with two daughters, noted that he publishes newsletter posts, practical AI YouTube tutorials, and podcast interviews alongside running a membership portal, and asked for five bot suggestions (with a few sentences explaining each) that could save him time or money.

The Advisor immediately responded with five tailored ideas: - **Daily AI News Scout** - **YouTube Comment Miner** - **Membership Concierge** - **Social Media Poster** - **Podcast Preparer**

Crucially, Peter points out a hidden superpower that many users miss: "you can actually get one bot, in this case our advisor bot, to help us create the rest of our bots." Rather than building them manually from scratch, he instructed his Advisor to combine two of its suggestions into a single bot that both researches other channels and mines viewer comments, christening it the "YouTube Researcher."

[04:42](https://www.youtube.com/watch?v=MkVcHbviYOw&t=282s) Bot #2: YouTube Researcher (Iterative Prompting in Action)

Once the Advisor spun up the YouTube Researcher with an initial instruction to send a morning intel brief, Peter stepped in to refine it. He first asked the bot to monitor specific competing channels in his niche. While the bot successfully pulled his YouTube comments and identified viewer themes, its first attempt at a brief was "very verbose and hard to follow"—a wall of text that was simply too much to process over morning coffee.

Instead of trying to "one-shot" the prompt, Peter demonstrates how to iteratively refine AI output. He gave the bot strict formatting rules and a tighter time window, since AI YouTube content moves fast and is highly topical: - List the **top 3 content ideas** first. - Show the **top 5 outlier videos** from other channels (defined as videos that beat that channel's average performance over the last 14 days). - List the **top performing videos overall** (limited strictly to the last 14 days). - Include the **top comment themes** from his own viewers at the end.

The revised report came back much more concise and actionable. Among the top three content ideas, the bot actually suggested making a video on "Spec Scale" (which Peter plans to film soon) and a video on Grok Bot—the exact video he is currently recording. After verifying the refined full report, Peter set it up as a recurring daily job that automatically delivers fresh angles, watch lists, and comment feedback every morning. He highlights this three-step loop as the core blueprint for every bot in the tutorial:

```mermaid flowchart LR accTitle: Core Grok Bot Creation Workflow accDescr: The three-step iterative process for building and automating bots in Grok Bot. subgraph Setup["1. Initialization"] A["Kick off with initial prompt"] end subgraph Refine["2. Iteration"] B["Review output & refine formatting back-and-forth"] end subgraph Automate["3. Automation"] C["Schedule proactive routine (daily, weekly, or monthly)"] end A --> B B --> C ```

[07:12](https://www.youtube.com/watch?v=MkVcHbviYOw&t=432s) Bot #3: X Scout (Mining Viral Posts and Automating Email Digests)

Applying the same three-step process, Peter builds an X Scout bot. Because Grok Bot is part of SpaceX/xAI, it enjoys first-party access to X (formerly Twitter) data. Peter admits that despite having an "unhealthy addiction to X," he still misses great posts and constantly bookmarks tweets that he never looks at again. To fix this, he prompted the X Bot to find the most viral X posts from the last 7 days in his niche and compile a weekly report of the top 10 tweets from people he follows, engages with, or bookmarks. He asked for the tweets to be grouped into categories—complete with full tweet copy, links, and analysis—and to conclude with three content ideas for him to tweet next.

The bot automatically located his X account, set up a morning routine, and generated a categorized report featuring three main themes: - **Playbooks that people save (not just like):** Highlighting posts from Greg ("23 ways I use AI agents to grow my startup") and a Chinese developer discussing "from Loop to graph engineering." - **Grok Bot, Codex, and ChatGPT Work compared in practice:** Surfacing tweets from Riley and from Ben (who works on the Cursor team) sharing the most loved internal Grok Bot use cases. - **Creator workflows:** Posts tailored to Peter's content production interests, followed by three suggested tweet ideas at the bottom.

Just for fun, Peter also asked the bot to pull the top five funniest tweets from his timeline; naturally, the number-one spot went to OpenAI and Anthropic sniping and joking at each other, alongside other AI-niche humor. Finally, because Peter doesn't want to open the Grok Bot app every time just to read his digest, he turns on voice mode and speaks directly to the bot: *"Can you send me this report to peter@behindthecraft.com? And also, make sure you include a full tweet copy in your report. Send to my email right now."* Seconds later, the formatted report lands directly in his Gmail inbox, complete with full tweet text and three next-tweet ideas—allowing him to wake up and consume the best of X right from his email.

[10:26](https://www.youtube.com/watch?v=MkVcHbviYOw&t=626s) Digital Marie Kondo: Email Cleanup and Subscription Cancellation

Grok Bot can act as a personal decluttering assistant—dubbed the "Marie Kondo" bot—designed to tackle digital chaos across the three primary places it accumulates: email inboxes, Google Drive storage, and recurring paid subscriptions. To set this up, Grok Bot is connected to the Gmail and Google Drive plugins, as well as an MCP (Model Context Protocol) integration for Mercury, a business bank account. The initial prompt instructs the bot to audit Gmail, Google Drive, and inbound email receipts to formulate a comprehensive cleanup plan. Specifically, it searches for newsletters that are rarely opened, large or abandoned Drive files, and active subscriptions flagged from receipts and Mercury banking records. Crucially, the prompt includes a strict safeguard: Grok Bot is explicitly forbidden from deleting, moving, unsubscribing, or canceling anything without explicit prior approval. Reviewing planned deletions beforehand is an essential security measure when letting an autonomous agent manage personal files.

```mermaid flowchart TD accTitle: Marie Kondo Cleanup Agent Workflow accDescr: Diagram illustrating how Grok Bot audits data sources, presents a numbered approval list, and executes deletions and cancellations on a cloud computer. subgraph Sources ["Connected Integrations"] A["Gmail Plugin"] B["Google Drive Plugin"] C["Mercury MCP (Bank)"] end subgraph Audit ["Audit & Approval"] D["Grok Bot Scans & Audits"] E["Refined Numbered Plan<br/>(Max 10 Items per Category)"] F["User Approval<br/>(Action Selected Numbers)"] end subgraph Execution ["Autonomous Execution on Cloud VM"] G["Unsubscribe Email Senders"] H["Trash Drive Files"] I["Cloud VM Automation<br/>(Lovable, Equip Foods)"] end Sources --> Audit Audit --> Execution A --> D B --> D C --> D D --> E E --> F F --> G F --> H F --> I ```

After authenticating on its dedicated cloud computer, Grok Bot successfully establishes connections to Drive, Gmail, and Mercury MCP. Its first output is a massive, sprawling list that proves overwhelming. AI agents rarely deliver the perfect format on the first try, so an iterative prompt is required: restrict the output to a maximum of ten items per category, strip out extraneous labels, and isolate only actionable items—emails to unsubscribe from, Drive files to delete, and subscriptions to cancel. Structuring the report as a clean numbered list is a crucial prompting workflow because it allows the user to issue batch execution instructions simply by referencing numbers rather than retyping names or URLs.

Once the numbered review is finalized, simple commands trigger execution: unsubscribing from emails numbered 3 through 8, deleting large files and an old tax return from Drive, and canceling subscriptions 11 and 15. The bot executes these actions autonomously on its cloud environment, unsubscribing from senders and trashing three Drive files. For account-level cancellations, the user signs into the cloud computer session: - **Lovable**: Upon logging in, the bot verifies that the account was already scheduled for cancellation, requiring no further action. - **Equip Foods**: The user manually signs into the protein powder service, after which the bot completes the cancellation process. As an aside, Equip Foods is an excellent brand, but the cancellation was necessary simply because too much protein powder had piled up at home.

The entire audit and cancellation process wraps up in approximately five minutes, replacing what would normally take thirty minutes to an hour of manual effort. To complete the experience, the bot's persona is updated with a prompt asking it to adopt Marie Kondo's tone. Grok Bot embraces the character immediately, reporting that completed files have been thanked and placed into the trash to rest, while noting that Equip Foods "no longer sparks joy" and releasing the Prime Protein subscription "with gratitude." The bot can be scheduled to run weekly or monthly to keep digital clutter at bay.

- Connected services: Gmail plugin, Google Drive plugin, and Mercury MCP. - Strict guardrail: Grok Bot is instructed never to execute deletions or cancellations without explicit approval. - Prompting best practice: Requiring a numbered list allows rapid execution commands via index numbers (e.g., "unsubscribe to 3, 4, 5... cancel 11 and 15"). - Time efficiency: Cut down an estimated 30–60 minutes of tedious administrative maintenance to roughly 5 minutes.

---

[15:09](https://www.youtube.com/watch?v=MkVcHbviYOw&t=909s) Personal Travel Concierge: Flight Tracking and Route Optimization

The Personal Concierge bot handles end-to-end vacation and travel logistics. For a family vacation planned for December in Japan, an AI-generated itinerary is already saved in a planning document. Rather than manually monitoring flights, Grok Bot is tasked with reading the vacation document, analyzing the exact flight legs, checking Google Flights on a recurring schedule, and surfacing alerts whenever pricing improves.

While standard flight tracking tools exist, Grok Bot offers a clear structural advantage over standard Google Flights price alerts: it understands the broader context and flexibility of the overall travel itinerary. Standard trackers only watch fixed, predetermined routes. Grok Bot, however, evaluates the entire route network against the vacation document.

The initial itinerary outlined open-jaw flights: departing San Francisco (SFO) to Tokyo, traveling onward from Tokyo to Fukuoka, and flying back from Fukuoka to SFO. When analyzing the options, Grok Bot uncovered that booking a standard round-trip ticket between SFO and Tokyo was approximately $2,700 cheaper than the open-jaw routing listed in the document. A basic flight tracker could never identify that level of structural savings because it lacks awareness of the traveler's overall destination plans.

Following the initial analysis, Grok Bot is scheduled to run automated checks every Monday at 9:00 AM to see if fares fluctuate. Beyond ongoing price monitoring, the agent can expand into a full-service travel assistant capable of completing flight check-ins, purchasing airline tickets, securing theme park admissions, and organizing daily on-the-ground activities.

- Core capability: Reads unstructured itinerary documents and queries Google Flights autonomously. - Contextual flight optimization: Discovered a round-trip route to Tokyo saving ~$2,700 over the open-jaw route (SFO → Tokyo → Fukuoka → SFO). - Scheduled automation: Configured for weekly price monitoring recurring every Monday at 9:00 AM. - Long-term utility: Serves as a persistent concierge for booking excursions, activities, and logistics during the trip.

---

[17:26](https://www.youtube.com/watch?v=MkVcHbviYOw&t=1046s) Retro Gamer Test: Cloud Computer Gaming and System Capabilities

Because Grok Bot operates on a dedicated virtual cloud machine, it can execute arbitrary local tasks beyond standard web browsing and APIs, including installing and playing retro PC games. In a direct test of the agent's system capabilities, Grok Bot is instructed to source and install *Command & Conquer: Red Alert*, *Doom*, and *Commander Keen*.

The agent successfully locates the required files, installs them inside the cloud environment, and launches the games on command: - **Doom**: Launching Doom inside the virtual computer allows selection of the classic episode menu and the "Hurt Me Plenty" difficulty setting. However, the experience falls short due to control limitations: the virtual mouse driver is improperly configured within the cloud session, causing the player camera to stay locked pointing directly at the floor without any way to pitch up. - **Commander Keen**: The classic DOS platformer fares slightly better than *Doom*. Using the Control key to jump, the platformer mechanics load up and run, but noticeable latency and input lag on both the keyboard and mouse make precise platforming controls difficult.

While Grok Bot is by no means ready to replace a dedicated gaming PC or cloud gaming platforms like GeForce NOW, the experiment illustrates the remarkable flexibility of having an AI agent operating inside its own persistent OS. Exploring the file manager reveals a desktop environment reminiscent of Windows 3.1 or a lightweight Linux desktop, complete with Chrome, a system terminal, the newly installed retro games, and travel files from the Japan flight search. Having an autonomous assistant that can install desktop software, configure environments, and manipulate local programs points to substantial potential—hinting at a future where xAI data centers could eventually stream full desktop applications and AAA games straight through cloud-hosted virtual agents.

[20:20](https://www.youtube.com/watch?v=MkVcHbviYOw&t=1220s) The Trust Hurdle: Virtual Cloud Computers and Privacy Concerns

The single biggest roadblock to widespread Grok Bot adoption comes down to user trust. Peter shares a candid moment of hesitation during his testing: while seeing a standard Google authentication prompt on a personal laptop rarely raises an eyebrow, encountering that exact login screen inside a remote, virtual cloud computer feels entirely different. He found himself pausing and wondering, "how do I know that nobody else is seeing the screen on the virtual cloud computer?" Handing over primary credentials to a machine running in the cloud introduces a psychological barrier that most everyday software never triggers.

```mermaid flowchart TD accTitle: Traditional vs Cloud Agent Workflows accDescr: Diagram comparing traditional manual computing on a local machine with the voice-driven cloud agent paradigm and its associated trust friction.

subgraph Local["Traditional Paradigm"] A["User Input: Keyboard & Mouse"] --> B["Local Laptop Machine"] B --> C["Direct App Login: Established Trust"] end

subgraph CloudAgent["Grok Bot Cloud Paradigm"] D["User Input: Voice Orchestration"] --> E["Virtual Cloud Computer"] E --> F["Automated Agent Execution"] F -.-> G{"Trust Barrier: Credential Entry in Cloud"} end ```

Looking for answers, Peter points to the privacy disclosures on the official Grok Bot website under the section addressing "How does Grokbot handle my privacy?" The documentation specifies that the service relies on the same Cursor SSO authentication and privacy mode that developers already trust. Furthermore, the virtual cloud computer environment is encrypted both in transit and at rest, and user data is explicitly excluded from AI model training. While Peter is comfortable granting these broad permissions as an early AI adopter, he emphasizes that mainstream users will likely struggle to grasp the concept of an ephemeral cloud desktop and will hesitate before logging into sensitive personal accounts. Ultimately, the AI agent platform that successfully solves this trust and verification problem will be the one that unlocks mass adoption.

- **Authentication and security architecture:** Grok Bot implements Cursor SSO authentication, utilizing end-to-end encryption for the cloud computer both in transit and at rest. - **Data privacy policy:** User activity, sessions, and data on the virtual cloud machine are not used for AI model training. - **Adoption bottleneck:** Mainstream non-technical users face friction when asked to enter primary credentials (such as Google or email logins) inside an unfamiliar virtualized cloud OS.

---

[21:27](https://www.youtube.com/watch?v=MkVcHbviYOw&t=1287s) The Shift to Cloud Agents, Pricing Realities, and Final Verdict

Grok Bot represents a fundamental shift in personal computing. Rather than manually clicking through tabs with a mouse and keyboard on a physical laptop, users are moving toward using natural voice instructions to orchestrate teams of agents operating across dedicated cloud instances. Grok Bot stands out as the very first consumer-facing product to make that autonomous multi-agent vision usable in practice. However, Peter notes that it is not quite ready to replace ChatGPT as his daily driver. Cost remains the primary reason: ChatGPT provides a vast toolkit for $20 per month, whereas running Grok Bot on an ongoing basis demands a steep $200 per month subscription tier.

Despite the prohibitive price tag for heavy regular use, Peter praises Grok Bot's interface, noting that it features a "much cleaner UI than ChatGPT right now" while delivering remarkable execution power. He finds it refreshing to see companies like Cursor and xAI emerge as serious front-line competitors against OpenAI and Anthropic. Cursor in particular may hold an edge in the agent race if it maintains its model-agnostic approach by supporting top-tier foundation models from every major provider. Wrapping up, Peter strongly encourages viewers to try the free tier to catch a glimpse of where computing is heading, noting that "the team really cooked here," and teases an upcoming deep-dive interview with the Cursor engineering team detailing how they engineered the platform.

- **Pricing comparison:** ChatGPT offers broader daily value at $20/month, compared to the $200/month commitment required for intensive, continuous Grok Bot usage. - **User experience:** Grok Bot delivers a cleaner, more focused user interface than ChatGPT while managing automated browser and desktop tasks seamlessly. - **Competitive landscape:** Cursor's ability to remain provider-agnostic and support multiple frontier models gives it a strategic advantage against OpenAI and Anthropic in the agent race. - **Availability:** A free tier of Grok Bot is available for users to test cloud computer agent workflows before committing to high-tier plans.

## Notable quotes > [00:00](https://www.youtube.com/watch?v=MkVcHbviYOw&t=0s) Grok Bot is the future of personal agents — Peter Yang

> [00:00](https://www.youtube.com/watch?v=MkVcHbviYOw&t=0s) a dedicated personal computer in the cloud for running AI agents. — Peter Yang

> [01:38](https://www.youtube.com/watch?v=MkVcHbviYOw&t=98s) just kind of like a mess of chat threads — Peter Yang

> [01:38](https://www.youtube.com/watch?v=MkVcHbviYOw&t=98s) feels more like talking to a coworker than getting lost in a hundred chat threads — Peter Yang

> [03:27](https://www.youtube.com/watch?v=MkVcHbviYOw&t=207s) you can actually get one bot, in this case our advisor bot, to help us create the rest of our bots. — Peter Yang

> [04:42](https://www.youtube.com/watch?v=MkVcHbviYOw&t=282s) very verbose and hard to follow — Peter Yang

> [07:12](https://www.youtube.com/watch?v=MkVcHbviYOw&t=432s) Can you send me this report to peter@behindthecraft.com? And also, make sure you include a full tweet copy in your report. Send to my email right now. — Peter Yang

> [10:26](https://www.youtube.com/watch?v=MkVcHbviYOw&t=626s) no longer sparks joy — Grok Bot

> [20:20](https://www.youtube.com/watch?v=MkVcHbviYOw&t=1220s) how do I know that nobody else is seeing the screen on the virtual cloud computer? — Peter Yang

> [21:27](https://www.youtube.com/watch?v=MkVcHbviYOw&t=1287s) much cleaner UI than ChatGPT right now — Peter Yang

> [21:27](https://www.youtube.com/watch?v=MkVcHbviYOw&t=1287s) the team really cooked here — Peter Yang

## Mentioned Peter Yang, Grok Bot, ChatGPT, ChatGPT Work, Codex, Hermes, YouTube, X, Marie Kondo, Gmail, Doom, Command & Conquer: Red Alert, Commander Keen, Lee, Cursor, SpaceX, xAI, Mac Mini, Jenny Wen, Behind the Craft, Spec Scale, Greg, Riley, Ben, OpenAI, Anthropic, Google Drive, Mercury, Lovable, Equip Foods, Google Flights, GeForce NOW, Windows 3.1, Linux, Chrome, Google

## Video description Copy my best AI workflows to save time and automate busywork: https://www.behindthecraft.com 🤝 Subscribe to my practical AI newsletter read by 150K+ professionals: https://creatoreconomy.so

I think Grok Bot is the future of personal AI agents. In this tutorial, I show you how to build five useful bots, including an advisor that creates other bots, a YouTube researcher, an X scout, a digital Marie Kondo that cleans up Gmail and saves money, and a travel concierge that finds cheaper flights. I also test using Grok Bot to play retro games and share my honest take on privacy, pricing, and whether it can replace ChatGPT as my daily driver.

Timestamps: (00:00) How Grok Bot is different from ChatGPT and Hermes (03:30) Advisor: Create and orchestrate other bots (04:30) YouTube Researcher: Find outlier videos and content ideas (07:00) X Scout: Surface the best posts from your feed and bookmarks (10:30) Marie Kondo: Clean your email and save money on paid subscriptions (15:00) Personal Concierge: Get alerts for better flight prices for your trips (17:30) Gamer: Install and play retro games (20:30) My honest take on privacy, pricing, and whether GrokBot replaces ChatGPT

Try Grok Bot here: https://x.ai/bot

Where to find me: Newsletter: https://creatoreconomy.so/ X: https://x.com/petergyang LinkedIn: https://www.linkedin.com/in/petergyang/

📌 Subscribe to this channel for more practical AI tutorials.

[Source](https://www.youtube.com/watch?v=MkVcHbviYOw)

中文简报

xAI联手Cursor推出Grok Bot云电脑以叫板ChatGPT Grok Bot推出了一个持久化云电脑操作系统，用语音驱动的智能体取代手动浏览器标签页，不过其每月200美元的高昂售价使其难以取代ChatGPT成为日常主力工具。 Peter Yang展示了Grok Bot的实际性能，强调了其基于SpaceX/xAI服务器的持久化云电脑架构，相比Hermes等本地配置或ChatGPT Work等碎片化工具更具优势。通过核心的三步提示词工作流，他构建了自主智能体来策划利基内容、清理收件箱并取消未使用的订阅，以及在结构上优化航班行程以节省数千美元。在通过运行复古DOS游戏测试了其完整的操作系统功能后，他坦率地指出了用户在远程凭据方面的信任阻碍，并指出每月200美元的高昂价格使其无法立即取代ChatGPT成为他的日常主力工具。 Grok Bot在托管于SpaceX/xAI基础设施的持久化云电脑中运行自主智能体，消除了ChatGPT Work的会话中断问题以及Hermes等本地配置所需的硬件维护。 (1:38) 通过理解整体旅行行程而非死板地监控固定路线，Grok Bot找到了一种前往日本的替代往返航班预订方案，相比缺口程计划节省了约2700美元。 (15:09) Grok Bot配备了完整的虚拟桌面操作系统，能够自主下载、安装和运行复古PC游戏等原生软件，但虚拟鼠标驱动程序和输入延迟目前影响了其实时可玩性。 (17:26) 尽管其UI优于ChatGPT且更为整洁，但重度使用每月200美元的高昂成本，以及用户对在云操作系统中输入主凭据的顾虑，仍然是其取代日常主力助手的重大障碍。 (21:27)

Reading summary

Peter Yang argues that Grok Bot signals a major shift toward autonomous personal agents by hosting them on persistent cloud computers, though high subscription costs and cloud credential concerns currently prevent it from replacing daily workflows. Persistent Cloud Architecture Versus Local Setups (1:38) Grok Bot operates inside a persistent virtual computer hosted on SpaceX and xAI infrastructure, differentiating it from both local hardware and standard browser plugins. Running agents locally through platforms like Hermes requires dedicated physical hardware, such as a 24/7 Mac Mini, or manual virtual private server configuration. Conversely, ChatGPT Work relies on cloud browsers that cannot remain signed in to third-party applications and divide user attention across disconnected chat, work, and coding tabs. By contrast, Yang reports that Grok Bot provides an out-of-the-box operating system with a persistent browser that preserves active sessions. Designed in partnership with Cursor's team, the interface replaces fragmented chat threads with animated bot avatars. Yang states that the resulting user experience feels more like managing digital coworkers than navigating conversational threads, resolving the operational friction of keeping physical laptops open to sustain background tasks. Iterative Orchestration and Content Intelligence (3:27) To manage specialized tasks, Yang outlines a three-stage lifecycle: initial prompting, iterative refinement, and recurring scheduled automation. Rather than configuring every assistant from scratch, users begin with a pinned "Advisor Bot" that audits the user's workload to architect and spin up downstream agents. When an initial output proves unwieldy, strict formatting rules transform verbose summaries into actionable intel briefs. Yang illustrates this by deploying a YouTube Researcher instructed to monitor competitor channels for outlier videos that outperform their 14-day baseline, as well as an X Scout bot. Leveraging direct access to X data, the scout categorizes high-engagement playbooks, creator workflows, and niche industry humor, concluding with custom post concepts. Using voice prompts, the bot can format the findings and dispatch the digest directly to an email inbox on a recurring schedule. Cross-Platform Decluttering, Travel, and Desktop Execution (10:26) Because Grok Bot controls a virtual computer, it can execute multi-step administrative workflows across connected integrations and local operating systems. Paired with Gmail, Google Drive, and Mercury banking data via Model Context Protocol (MCP), a decluttering agent audited receipts to cancel an unneeded Equip Foods subscription and trash abandoned Drive files in approximately five minutes. To prevent unintended deletions, Yang introduced a strict safety constraint: the agent must present a numbered list requiring explicit user approval before executing any destructive action. In travel logistics, Grok Bot analyzed an unstructured vacation document for a family trip to Japan. By contextualizing the entire itinerary rather than monitoring isolated route legs like standard flight trackers, the agent identified an alternative round-trip flight to Tokyo that was roughly $2,700 cheaper than the proposed open-jaw routing. Testing system boundaries further, Yang instructed the agent to install and launch classic PC titles like Doom and Commander Keen. While mouse driver bugs and keyboard latency made gameplay difficult, the experiment demonstrated that the agent can manipulate arbitrary desktop software inside its environment. Privacy Hesitations and the Pricing Barrier (20:20) Yang identifies user trust as the principal adoption bottleneck facing cloud-hosted agents. Entering primary credentials into a remote virtual desktop triggers psychological friction that local browser logins do not. According to official Grok Bot documentation, the service relies on Cursor single sign-on, maintains encryption both in transit and at rest, and excludes user sessions from AI model training. Even so, Yang cautions that mainstream users will hesitate to entrust sensitive accounts to virtual environments without clearer verification models. Cost presents a second major barrier to daily use. While ChatGPT offers broad utility for $20 per month, sustained Grok Bot utilization requires a $200 monthly tier. Consequently, Yang concludes that Grok Bot cannot yet replace his daily setup of ChatGPT and Codex, even though its clean interface and Cursor's model-agnostic architecture represent a superior vision for multi-agent workflows. Yang concludes that while Grok Bot successfully validates the concept of autonomous cloud agents orchestrating real-world tasks, widespread adoption depends on lowering the $200 monthly price tag and helping non-technical users navigate the trust barrier of entering credentials into remote virtual machines.

Source quotes

[0:00] Peter Yang: Grok Bot is the future of personal agents [0:00] Peter Yang: a dedicated personal computer in the cloud for running AI agents. [1:38] Peter Yang: just kind of like a mess of chat threads [1:38] Peter Yang: feels more like talking to a coworker than getting lost in a hundred chat threads [3:27] Peter Yang: you can actually get one bot, in this case our advisor bot, to help us create the rest of our bots. [4:42] Peter Yang: very verbose and hard to follow [7:12] Peter Yang: Can you send me this report to peter@behindthecraft.com? And also, make sure you include a full tweet copy in your report. Send to my email right now. [10:26] Grok Bot: no longer sparks joy [20:20] Peter Yang: how do I know that nobody else is seeing the screen on the virtual cloud computer? [21:27] Peter Yang: much cleaner UI than ChatGPT right now [21:27] Peter Yang: the team really cooked here

Original transcript

[0:00] Hey everyone. [0:02] I think Grok Bot is the future of [0:04] personal agents and I'm going to show [0:06] you how to build five of my favorite [0:08] bots in this video. [0:10] We're going to build an advisor to [0:12] create and manage your other bots, a [0:14] YouTube researcher to find outlier [0:17] videos in your interest, an X Scout to [0:20] surface the most insightful and funny [0:22] tweets from your timeline, [0:24] a digital Marie Kondo to clean up your [0:27] email and find paid subscriptions to [0:29] cancel, a personal concierge to help you [0:32] get good prices on travel and vacations, [0:35] and as a bonus, we're going to build a [0:37] gamer to see if Grok Bot can play [0:39] classic games like Doom and Red Alert. [0:43] We'll finish with a discussion about [0:45] privacy and whether Grok Bot should be [0:47] your new AI daily driver. [0:50] All right, but first, [0:52] let me show you what makes Grok Bot [0:54] different. [0:55] A few weeks ago, I tweeted that ChatGPT [0:58] was one feature away from building the [1:00] AI product I want, which is a dedicated [1:02] personal computer in the cloud for [1:05] running AI agents. Lee from the Cursor [1:07] team DM'd me shortly after to test Grok [1:10] Bot. [1:12] What exactly is a dedicated cloud [1:14] computer? You know the meme of people [1:16] walking around with their laptop half [1:18] open to keep their agents running? [1:21] You don't have to do that anymore if [1:22] you're using Grok Bot. Basically, Grok [1:25] Bot lives in a computer sitting [1:27] somewhere in SpaceX AI servers instead [1:31] of in your home. It has its own browser [1:33] and operating system that you can use [1:35] and log in to your favorite apps. [1:38] Now, let's do a quick comparison of [1:40] Hermes, ChatGPT, and Grok Bot. [1:44] Hermes I run on my Mac Mini at home. [1:47] It's on 24/7 and it gives it a [1:49] persistent computer, but you have to go [1:52] buy the machine and set everything up [1:54] for yourself. [1:55] You can technically also run Hermes on a [1:58] virtual private server, but that again [2:00] requires more setup work. [2:02] Now, ChatGPT Work uses plugins and a [2:05] cloud browser, but the problem is that [2:07] browser as it right now [2:09] cannot stay signed in to your favorite [2:11] apps. [2:12] The UX for ChatGPT Work also feels [2:15] scattered across chat, work, and Codex. [2:19] Let's take a quick look here. [2:21] It's just kind of like a mess of chat [2:23] threads here. [2:24] There's ChatGPT, Codex. There's like [2:26] another tab up here for work and chat. [2:29] And it's just all really confusing even [2:30] to someone like me. [2:32] In contrast, Grok Bot gives you a [2:34] persistent cloud computer right off the [2:36] box. [2:37] And I think it's the easiest of the [2:39] three to set up and get going. [2:41] And its interface, as you can see here, [2:43] also feel a lot more focused and [2:45] delightful. [2:46] I love the little bots and the [2:49] animations that they make when they [2:50] start working. Right, let's say [2:53] look up the weather. [2:55] And you can see here that there's some [2:57] pretty cool subtle animations. [2:59] Each bot has a different personality and [3:01] style. And this is the work of Cursor's [3:04] great design team, including Jenny Wen [3:06] and other folks on her team that I've [3:08] interviewed in the past. [3:10] Overall, I think this UX [3:12] feels more like talking to a coworker [3:14] than getting lost in a hundred chat [3:16] threads. [3:18] So, can Grok Bot actually replace [3:20] ChatGPT and Codex, which is my current [3:22] daily driver? Let's run through some of [3:24] my favorite bots right now. [3:27] So, the first bot that you should create [3:28] after you install Grok Bot is an [3:31] advisor. [3:32] Here I have my advisor pinned to the [3:34] very top [3:35] of my Grok Bot list. [3:37] With your advisor, you want to start by [3:39] telling it about your work and life. [3:42] And then actually ask it to give you [3:43] some ideas for bots to create. Here's [3:46] what I told my advisor. I'm a creator [3:47] and founder of Behind the Craft. I have [3:50] two daughters and I live in the Bay [3:51] Area. I publish newsletter posts and [3:53] YouTube videos focus on practical AI [3:55] tutorials and podcast interviews, and I [3:58] also have membership portal over here. [4:00] So, based on all this, suggest five bots [4:02] that can help me save me time or money. [4:05] And for each, explain what it would do [4:07] in a few sentences. [4:09] As you can see here, it created five bot [4:12] ideas. A daily AI news scout, YouTube [4:15] comment miner, membership concierge, [4:18] social media poster, and podcast [4:20] preparer. [4:21] Right? [4:23] Now, what many people don't realize is [4:25] that you can actually get one bot, in [4:27] this case our advisor bot, to help us [4:30] create the rest of our bots. [4:32] Here, I asked my advisor to build a [4:34] single bot that both researches other [4:36] channels and mines comments. [4:39] And let's call it YouTube researcher. [4:41] Right? [4:42] So, now let's go ahead and check out the [4:44] YouTube researcher use case. [4:46] You can see that the advisor kicked off [4:49] the YouTube researcher bot with this [4:51] instruction to create a warning brief to [4:53] send me YouTube intel every morning. [4:56] But, I want to give it more specific [4:57] instructions, right? First, I asked it [4:59] to monitor specific channels in my [5:01] niche. [5:03] And let's go down here, and it's [5:05] starting to monitor channels, and here [5:07] is its first attempt at a brief. [5:10] You can see here that the brief is like [5:11] just like very verbose [5:12] and hard to follow. [5:14] There's like a lot of stuff here. It did [5:16] end up pulling my YouTube comments and [5:18] finding themes there, so that's great. [5:20] But, this is just like too much to [5:22] process, right? [5:23] So, I gave it some more instructions. I [5:25] told it to give me a report in the [5:27] specific format, which is first, list [5:30] your top three content ideas, [5:32] then give me the top five outliers for [5:34] other channels, [5:35] and then give me the top performing [5:36] overall. Limit your research to the last [5:39] 14 days because YouTube with AI stuff is [5:41] very topical. [5:43] And also add the top comment themes back [5:45] to the report. [5:47] And by the way, this is how we should be [5:49] working with AI to refine its output [5:51] instead of trying to one shot something. [5:54] Now you see here that the report is much [5:56] more concise and easier to follow. Let's [5:58] take a quick look. Top three content [6:00] ideas, it wants me to make a video on my [6:03] Spec Scale that I plan to do soon. [6:06] It tells me to make a video on Grok Bot [6:08] that we're making right now. [6:10] And here are the top outliers from the [6:12] other channels. [6:13] Outlier meaning that it beat that [6:15] channel's performance for the last 14 [6:16] days. [6:17] Yeah, pretty interesting topics here [6:19] that we can follow up on. [6:21] And top performing overall. [6:23] And also the top common themes. And it [6:25] did more here and then I told it to send [6:26] the full report again just to make sure [6:28] it looks good. [6:29] And now what I can do is basically [6:33] there's a daily job that sends me this [6:35] report every morning, right? Just this [6:37] morning for example, it sent me a new [6:39] report [6:40] with the top angles, the watch list, and [6:42] comments. [6:44] And I basically have a YouTube [6:45] researcher that proactively finds [6:47] interesting topics for me to make videos [6:49] on and gives me feedback based on my [6:51] users' comments. [6:53] Okay? So this is the process that we're [6:55] going to follow for pretty much the rest [6:56] of the bots that we're going to make, [6:58] which is we're going to kick it off with [6:59] an initial prompt. We're going to [7:01] iterate back and forth with it to make [7:03] its output good. And then we're going to [7:05] schedule a routine or job that runs [7:07] daily, weekly, or monthly to have it [7:09] proactively do work for us. [7:12] Now, let's actually see how we can [7:14] follow the same process [7:15] with our X Scout Bot. [7:18] So Grok Bot is part of SpaceX AI, right? [7:22] So it should have first-party access to [7:25] X and Twitter data. [7:26] And even though I have an unhealthy [7:28] addiction to X, I still miss plenty of [7:31] great content or bookmark tweets that I [7:34] never look at again. [7:35] Here's the prompt that I gave my X Bot. [7:38] Find the most viral X posts from the [7:40] last 7 days in my niche and create a [7:42] weekly report with the top 10 tweets [7:45] from people I follow, engage with, or [7:47] bookmark grouped into a few categories. [7:49] Include a full tweet copy, link, and [7:51] your analysis. And end with three [7:54] content ideas for me to tweet about [7:56] next. [7:57] You can see here that it found my X [7:59] account [8:00] and it automatically created a routine [8:02] to do this every morning. [8:04] And here is the initial output, right? [8:08] So, it found themes around playbooks [8:10] that people save, not just like, from [8:12] Greg, 23 ways I use AI agents to grow my [8:15] startup. [8:16] Chinese developer from Loop to graph [8:18] engineering. And more tweets from Greg. [8:22] Another theme that it found is just [8:23] about Grokbot, Codex, and ChatGPT work [8:26] compared and actually used, right? Tweet [8:28] from Riley and from Ben, who works on [8:30] the Christian team, on the most loved [8:32] internal Grokbot use cases. [8:35] And more. And the third theme is around [8:37] creator workflows that I actually am [8:39] interested in [8:41] And it ended this report with the three [8:44] things to tweet about next. [8:46] Some ideas here. [8:48] And just for fun, I asked it to also [8:50] give me the top five funniest tweets [8:52] from my timeline. [8:53] As they get through it, and the number [8:55] one funniest tweet is, as usual, OpenAI [8:58] and Anthropic kind of sniping at each [9:00] other and kind of joking. [9:03] Right? And then there's some more funny [9:04] tweets down here. [9:05] All kind of in my AI niche. [9:08] You can see here that it set up a [9:09] routine automatically and it sent me a [9:12] new report this morning. [9:14] Now, let's actually tell it to do [9:15] something else. I'm going to turn on [9:16] voice here. [9:18] Can you send me this report to [9:20] peter@behindthecraft.com? [9:23] And also, make sure you include a full [9:26] tweet copy in your report. Send to my [9:28] email right now. [9:30] Right? So, let's see if it actually is [9:32] able to send this report over email cuz [9:34] I don't want to have to open Grokbot [9:35] each time to see it. I want to just in [9:37] my inbox every morning. [9:39] Awesome. Here's a report from GrokBot. [9:42] You can see here that it has these [9:43] themes. It's included a full tweet copy. [9:47] I can scroll down here. [9:49] And yeah, I I think this looks pretty [9:51] good, right? And then it has down here, [9:53] I'm sure, [9:54] the top three things to tweet about [9:55] next. Again, the process here is to [9:59] give GrokBot initial prompt, [10:02] have it pull up the information, iterate [10:04] with it to get the output in the report, [10:05] right? And then either tell it to send [10:07] you a daily job through GrokBot itself [10:11] or through your email. In this case, I [10:12] prefer email cuz I like to wake up with [10:14] the top tweets to consume directly in my [10:16] inbox instead of having to open a [10:18] separate app. [10:19] And by the way, it's able to send me [10:21] email because I connected to a bunch of [10:23] plugins, including the Gmail plugin. [10:26] Now, let's move on to a fun bot [10:30] that I call Marie Kondo that cleans up [10:33] your digital clutter. [10:35] There are three places where clutter [10:36] piles up, right? Your email, [10:39] Google Drive, and your paid [10:41] subscriptions. [10:42] Now, I've connected both Gmail and Drive [10:46] in plugins to GrokBot. [10:48] And then I gave it this prompt. [10:50] Audit my Gmail, Google Drive, any [10:52] current email receipts, and create a [10:54] cleanup plan. Find newsletters I rarely [10:56] open. [10:57] Find large or abandoned Drive files. [11:01] Try to identify paid subscriptions from [11:03] email receipts. And also, you can hook [11:04] up Mercury MCP to get this data. Mercury [11:07] is a bank that I use for my business. [11:09] And then group everything into these [11:11] different categories, use a number list. [11:14] And crucially, I told it to not move, [11:17] delete, or unsubscribe, or cancel [11:19] anything without my approval, right? The [11:21] last line is really important for a bot [11:24] that cleans up your files. You always [11:26] want to review what it plans to delete [11:29] before letting it remove anything. [11:32] And you can see here that first Grokbot [11:34] confirmed that it's connected to Google [11:36] Drive and Gmail. [11:37] And then it actually connected to [11:39] Mercury MCP after signing on the remote [11:42] cloud computer to make this work. [11:44] And then it gave me this initial report. [11:48] This is a massive list of stuff to clean [11:50] up in my email, [11:52] subscriptions to review, [11:54] and other things that I should not [11:55] touch, right? Honestly, this list is [11:58] like pretty overwhelming. It's [11:59] incredibly long. [12:00] And I gave it some feedback of like, [12:02] "Hey, list it properly, categorize it [12:03] properly." It's still an incredibly long [12:05] list, right? Again, this stuff is not [12:07] going to get it right in one shot. [12:10] And eventually, I told it to just show [12:11] me max 10 items in each list and get rid [12:14] of all the random labels that it has. [12:16] Now, this is much more digestible. It [12:18] has a bunch of emails here, [12:21] Google Drive, and paid subscriptions. [12:24] And I gave it one final piece of [12:25] feedback to only show me emails to [12:28] unsubscribe to, Google Drive files to [12:30] delete, and paid subscriptions that I [12:32] want to cancel. And then here's the [12:34] updated list that it came up with. [12:36] Now, it has this list as a numbered [12:38] list. [12:39] And my feedback is to actually get it to [12:41] take action, which is email unsubscribe [12:44] to three and four, five, six, seven, [12:47] eight. Google Drive, delete the actual [12:50] tax return, delete some of these large [12:51] files, and paid subscriptions, cancel 11 [12:54] and 15, right? And just as a general [12:57] tip, it's always good to ask Grokbot or [12:59] AI to give you its response in a [13:02] numbered list like this to make it super [13:04] easy for you to just like tell it to do [13:06] things by just referring to the number [13:08] in the list instead of having to type [13:09] everything over again. I use this [13:11] pattern all the time. All right, so now [13:13] Grokbot is actually taking action, [13:15] right? This is where the magic happens. [13:17] You can see here it's already [13:18] unsubscribed to a bunch of senders. It's [13:22] trashed three Google Drive files that we [13:24] asked it to. And now it's canceling [13:26] Lovable and Equip Foods, which is a [13:28] protein company. [13:30] So now to cancel Lovable, I first have [13:32] to sign to the cloud computer with my [13:34] Lovable credentials, which I did [13:36] manually. [13:37] And we found out that actually Lovable [13:39] is already scheduled to be canceled. So [13:40] Lovable is done. Then it asked me to [13:42] sign in to Equip Foods and I did that. [13:44] And then it canceled uh the protein [13:46] subscription for me. By the way, Equip [13:48] Foods is a great protein company. I'm [13:50] only canceling because I have too many [13:52] protein powders at home already. [13:56] Uh so it basically did all this work, [13:57] right? It's trashed a bunch of stuff. [13:59] It's unsubscribed to a bunch of emails [14:01] and it's canceled a bunch of [14:02] subscriptions to save me money. [14:05] And you can see that it did all this in [14:07] around 5 minutes when it would have [14:09] taken me probably 30 minutes to an hour [14:10] to do all this manually. [14:12] Marie Kondo is very, very useful. Of [14:15] course, [14:16] I think there's something missing with [14:17] this bot. I think we should ask it to [14:20] Can you talk like Marie Kondo from now [14:22] on? Give me an example. [14:25] Because right now it sounds too much [14:26] like a robot and not so much like Marie [14:28] Kondo. And let's see what it comes up [14:30] with. [14:31] All right. So now Marie Kondo is [14:34] actually going to talk like Marie Kondo. [14:36] As you can see, Hermes and the thing [14:38] have completed work. We thank them and [14:40] place them in the trash where they may [14:41] rest. Equip Foods no longer sparks joy [14:44] and release Prime Protein subscription [14:46] with gratitude. [14:46] >> [laughter] [14:47] >> Yeah, so now we can set up Marie Kondo [14:50] to maybe talk to us every week or every [14:52] month to clean up our digital files and [14:55] spark joy, right? So this is definitely [14:57] a bot that I recommend you setting up. [14:59] Just remember to ask it to give its [15:02] output in a number list instead of just [15:03] doing the cleanup for you so that you [15:05] can review it first, right? You don't [15:06] want it to actually delete some [15:08] important file. [15:09] Now let's cover the personal concierge [15:12] bot. This is a bot that I want to use [15:15] for all my vacation and travel plans. It [15:18] has access to my vacation document where [15:20] I've listed in December trip to Japan. [15:22] Here's a quick preview of document. It [15:24] has the full itinerary for Japan [15:26] that I created with AI as well. [15:29] Let's go back to Grokbot. And what I [15:31] want Grokbot to do is I haven't quite [15:34] booked my flights to Japan yet. So, I [15:36] want it to monitor for price changes and [15:38] alert me on what routes are the best [15:41] deal. Here's a prompt that I gave it. [15:43] Read my vacation document and monitor [15:45] the exact flight legs for my family [15:47] trip. [15:48] Get the dates, the best options for each [15:50] flight leg, check regularly, and let me [15:52] know when the price improves. [15:55] You see here that it found my flight [15:56] legs and it's using Google Flights to [15:58] check the prices. Now, you may be [16:00] wondering [16:01] what's the advantage of using Grokbot [16:03] for this instead of just setting up [16:05] Google Flight price alerts? [16:07] And the value here is that Grokbot can [16:09] understand my whole trip based on my [16:11] document and decide what's a better [16:13] option for my family. If you scroll down [16:15] here, [16:17] the big finding is that a Tokyo round [16:20] trip is about $2,700 cheaper [16:23] than the open-jaw flights that my dog [16:26] prefers, right? My dog prefers it to fly [16:29] from SFO to Tokyo and then from Tokyo to [16:32] Fukuoka, which we're going to, and then [16:34] from Fukuoka back to SFO. But actually, [16:36] Grokbot found a better itinerary that's [16:38] about $2,700 cheaper to just fly round [16:41] trip to Tokyo instead of doing this [16:43] itinerary that I laid out. Right? But a [16:45] simple Google price alert would not have [16:48] found this. [16:50] And now I asked it to check every Monday [16:51] at 9:00 a.m. to see the price improves. [16:53] And you can see here that the price is [16:55] still around the same. The Tokyo round [16:57] trip is still much cheaper. Eventually, [17:00] I can ask Grokbot to also just go ahead [17:01] and book the flight for me or check into [17:03] the flight when the time comes. [17:06] And generally speaking, I think it's [17:08] always a good idea [17:09] to have a travel thread or a travel bot [17:13] to both help you plan travel ahead of [17:14] time and also when you're at the [17:16] location to help you book amusement [17:18] parks, to help you figure out what to do [17:20] every single day. [17:22] So, I imagine I'm going to be using this [17:23] personal travel concierge a lot. [17:26] All right, let's test one more use case [17:27] with Grokbot. Because Grokbot gives me a [17:30] dedicated cloud computer, I thought it [17:33] would be fun to ask it to install and [17:35] let me play some retro games. [17:37] I asked it to install Red Alert, [17:40] Doom, and Commander Keen. [17:42] And you can see here that it found the [17:44] files and installed it. So, why don't we [17:46] just ask it now? Let's say, open Doom [17:49] for us to play. [17:51] And let's see how things work out. [17:53] All right, looks like Doom is up. Let's [17:55] open it in our virtual cloud computer. [17:58] And here it is. [18:00] And here we can start a new game. [18:02] Let's pick this episode. [18:05] And yeah, let's say hurt me plenty [18:06] plenty. [18:08] Here's Doom. And here's kind of where [18:10] Grokbot falls apart a little bit. [18:12] Unfortunately, because it's on a virtual [18:14] cloud computer, the mouse isn't quite [18:16] configured right to actually play Doom. [18:19] For some reason, it's looking at the [18:20] floor all the time, and I can't seem to [18:22] adjust the mouse to actually look up. [18:25] All right, so Doom doesn't quite work [18:26] here right now. [18:28] Now, let's actually try playing some [18:29] other game. Let's try asking to play [18:31] Commander Keen. I'm not sure if you guys [18:33] know this, [18:34] but Commander Keen is an awesome [18:36] platform game that I played in my youth. [18:39] And now it's loading Commander Keen, so [18:40] let's see what it comes up with. [18:42] Awesome, this is Commander Keen. Let's [18:44] see if it plays well or not. [18:47] Let's start a new game. [18:49] One player, normal difficulty. [18:52] And here we go. [18:53] Here we go. It's a very basic [18:55] platformer, but I remember really [18:57] enjoying it in my youth. [18:59] How do I jump? I forgot how to jump. [19:01] Also, control is jump. [19:03] And yeah, it plays better than Doom for [19:05] sure, but there is some lag in the [19:08] keyboard and mouse that makes it [19:10] difficult to actually control the [19:11] character. [19:12] Okay? [19:13] So, um um yeah, that's just for fun. [19:16] But basically, Grokbot is not replacing [19:19] your gaming PC or GeForce Now yet. [19:23] But the fact that the agent can install [19:24] and launch these games on its own [19:26] computer gives you a sense of how [19:28] open-ended this could become and how [19:30] much potential there is here, right? So, [19:32] maybe SpaceX AI can use all the data [19:35] centers and GPUs in space to actually [19:39] deliver triple-A games through just [19:41] virtual cloud computer. That would be a [19:43] dream. [19:44] Right? [19:45] And while we're at it, I opened this [19:47] file manager here, and here's all the [19:49] stuff that we installed into our virtual [19:51] cloud computer. It kind of looks like [19:53] Windows 3.1 [19:54] a little bit. I'm not sure what kind of [19:56] thing it is. There's Japan flights. [19:58] There's other stuff here. [20:00] Right? And then there's games that we've [20:02] installed. And also we have Chrome and a [20:04] terminal. That's kind of our virtual [20:06] cloud computer. [20:08] All right. [20:09] Now, [20:19] All right. Now, before we wrap, I want [20:20] to talk about probably the biggest [20:22] hurdle for Grokbot adoption, which is [20:24] trust. When I see a Google sign-in [20:27] screen like this on my laptop, I don't [20:29] really think twice before signing in. [20:31] But because this appeared on the virtual [20:33] cloud computer, I hesitated a bit [20:35] because how do I know that nobody else [20:37] is seeing the screen on the virtual [20:39] cloud computer? [20:40] Now, if you go to the Grokbot website, [20:42] if you scroll all the way down here, [20:44] there's a note about privacy here, which [20:45] is [20:46] "How does Grokbot handle my privacy?" [20:48] Grokbot uses the same cursor SSO auth [20:50] and privacy mode you already trust. Your [20:53] cloud computer is encrypted in transit [20:55] and at rest, and there's no AI training [20:57] on top of it, right? [20:59] So, you know, I'm willing to give [21:02] Grokbot and the virtual cloud computer [21:04] access to all the stuff because I'm an [21:06] early AI adopter. but I can see normal [21:09] people struggling to understand what [21:11] this cloud computer thing even is [21:14] and [21:15] kind of hesitate to sign into their [21:16] favorite apps on this device. [21:19] So, I think the AI agent platform that [21:22] figures out trust will be the first to [21:24] get mass adoption. [21:27] But overall, I think Grok Bot is a clear [21:29] sign of the future. We're moving away [21:32] from manually using our keyboard and [21:34] mouse to do work on our laptops [21:37] to using our voice to orchestrate a [21:40] bunch of agents that live in a dedicated [21:43] cloud computer. [21:44] And Grok Bot is the first product to [21:46] actually enable this. [21:48] Now, it's not quite my daily driver yet [21:50] because I think ChatGPT still offers [21:52] more for $20 a month, while Grok Bot [21:55] requires paying $200 a month to use on a [21:57] regular basis. [21:59] But I think Grok Bot has much cleaner UI [22:02] than ChatGPT right now [22:04] and is also very, very capable. [22:07] And overall, it's just great to be in a [22:08] world where Cursor and SpaceX AI are [22:11] just as viable a competitor as OpenAI [22:14] and Anthropic. I think Cursor may even [22:16] have the edge if it can continue to [22:18] support multiple models from all [22:21] providers. [22:22] So, Grok Bot is available for free [22:25] and I definitely recommend downloading [22:26] and trying it [22:28] to see a glimpse of the future. [22:31] I'll include some of the prompts from [22:32] this video in the pinned comment below [22:34] and I also have an exclusive interview [22:36] with the Cursor team on how they built [22:39] Grok Bot coming up in the next few [22:41] weeks. [22:42] Overall, I'm really impressed by Grok [22:44] Bot. I think the team really cooked here [22:46] and I can't wait to hear the story [22:47] behind how they built this. [22:49] So, please like and subscribe [22:50] if you enjoyed this video and I'll see [22:52] you next time.