一个人的对冲基金：Wes Roth 用 Grokbot 编制一张 24/7 AI 研究台（SpaceX 赞助）

信源说明：本文基于 Wes Roth 频道视频《my bots just BROKE Wall St…》（视频 ID rY8FuDONHxw，发布于 2026-10-30，全长约 27.6 分钟，英文转写 14 段覆盖 0–1658 秒，已完整读取）整理。本期为赞助内容：Wes Roth 明确说明视频由 SpaceX 赞助（他自称无订阅提成，且从 Grokbot 发布首日就在用）；他在片中称 Grokbot 为 SpaceX 产品、SpaceX 已收购 Cursor、并预告 Grokbot 将按请求路由到 Anthropic/Cursor/Grok 等多家模型——以上均为赞助语境下的主播陈述，本文未独立核实，按「Wes Roth 称」归属。全部交易演示为纸面交易（paper trading），主播反复声明不构成任何投资建议。同频道前文：《Blood Bench 与「导演式建造」范式》、《「指数时间假设」被 AI 证伪之后》。ASR 疑似错词以「转写疑作」标注，文末附校正表与完整转写原文。

先说结论

标题里的「BROKE Wall St」是营销话术，视频本体是一份完成度很高的多智能体组织教程：Wes Roth 用 Grokbot 从零搭了一张「一个人的对冲基金研究台」——幕僚长（chief of staff）做唯一联络人，十来个专科 bot 组成研究、记账、交易流水线，7×24 小时在云上跑。真正的教学内容不是某个功能按钮，而是三件可迁移的组织学：其一，把日常工作拆成「可交接的工序」（谁来接电话、谁来读文件、谁来当反方），再逐件自动化；其二，信息降噪靠编制——十几个 bot 只有幕僚长和主 bot 有权推送通知，其余一律静音；其三，让 bot 全程写日志，数据资产一年后才兑现成「哪种推理方式通常带来赢面」。边界也清楚：这是赞助视频，交易全程纸面，他的结论是「我仍然做最终决策」。

设定：云端「远程雇员」与一张纸面交易研究台

Wes Roth 的定位词是「drop-in remote employee（即插即用的远程雇员）」：每个 bot 有自己的云端电脑和工作区，你关掉所有设备它们仍在干活，手机、桌面随时接入。他提醒的第一条心智纪律是别一上来就搭大编制：「从一件每周都做的重复活开始」，把日常工作流拆成小块、逐块自动化——「一旦做到，生活会变得非常美好」。为了演示，他把对冲基金的一整套部门搬了进来：拆解财报的分析师、读 FTC 与政府文件的文书、法务、税务、会计，以及「在你错了的时候告诉你」的人——这些过去要么烧资本要么烧时间。

编制表：从幕僚长到「怀疑论者 Bear」

这张研究台的组织结构值得完整记录（均为 Wes Roth 演示口述）：

• Grokbot（主 bot）：新增的通用总管，「执行助理」角色；
• 幕僚长（Chief of Staff）：唯一联络人，向其他 bot 派活、汇报、改名换头像、创建新 bot——但不能删除 bot（防止误删数据，删除必须由人手动做）；它还负责把五十条 bot 消息过滤成「值得让你知道的那几条」；
• Hard Code：构建者，搭出了带实时行情与交易流的「Hedge Desk Arena」仪表盘，还主动提出「给我一个 GitHub 仓库或 Cursor Origin（他称 Cursor 版 GitHub、为 agent 规模化设计）会干得更好」；
• Clerk：盯 watch list 公司的 SEC 文件，标记内幕减持、大交易等红旗（演示中提到 Nvidia、AMD 的收购动态）；
• Mapper：供应链图——AI 芯片、代工设备、内存、数据中心、电力、云；
• Calendar：市场日历——财报、联储会议、通胀数据；
• Bear：专职怀疑论者，专门挑战其他 bot 的结论，且给出「什么情况能证明我错了」（可证伪声明）；
• Analyst：财报分析师，负责拉远视角做综合；
• Keeper：记账员——持仓表、对标的目标权重测量漂移、把缺口变成待你签核的再平衡方案（不擅自动手，只向幕僚长报事）；
• Taxman：税务与税收优惠账户；自封的「业余工作」是抱怨加州的文书与税；
• Paper Trader：按研究流下单、跑回测，唯一基准是「能不能跑赢标普 500（即转写疑作 SPY）」——演示里它一边晒跑赢大盘的战绩一边留现金，「等 analyst 或 bear 给我理由再进场」。

工序上，Clerk/Mapper/Calendar/Bear/Analyst 的产出汇入 Paper Trader，幕僚长居中连线；Wes Roth 把这些 bot 拉进一个群聊（group chat），让交接在工作流内完成。其中一个细节很像真团队：Google Sheets 统一数据格式这件事最初没交代，有的 bot 用了逗号分隔文件——他的解法不是挨个教训，而是给幕僚长下一句话「让所有人接上 Google Drive、上传表格、格式弄好看」。

让编制自己转：例行程序、技能文件与邮件身份

三个让自动化「无人化」的机制：

1. 例行程序（routines）：「每天开盘前一小时、收盘前一小时查一次 subreddits，重大新闻才通知我」；等待突发事件（AI 实验室发布、预期中的下跌）时可加密到每两三分钟刷一次指定网站/推特。他强调这不是死的定时提醒——bot 知道你的 watch list，会自己判断「值不值得打扰你」。
2. 技能（skills）：浏览器里「教一个任务」（teach a task）——录一遍你的操作，bot 用视觉回看视频、写成步骤化的技能文件。他教的例子是扫雷（还翻车了：「Grok 扫雷跟我一个水平，但 Excel 强我十倍——后者显然更重要」）；他真用的技能是多源信息聚合上 KPI 仪表盘，一天跑两次，把原先 20–30 分钟的活变成一条命令。关键属性：技能就是 .md 文本文件，可跨模型/跨 agent 迁移，其他平台攒的技能可以直接搬来。
3. Agent Mail：让 bot 自己开一个专属邮箱（免费、给 AI agent 用的邮件服务）——收通讯、注册服务，不必冒你的名义发信。

权限模型值得单独记一笔（他自己提醒的）：bot 各有工作区但共用一台机器，接入 Gmail/日历后所有 bot 都有访问权，「给了一个，另一个技术上也能用」。家用无妨，放到企业数据上这就是首先要重新设计的一层。

数据日志：一年后才兑现的护城河

视频里最「反直觉有用」的一条建议：让每个 bot 把交易、想法、分析、你和它的沟通全部写进日志存 Google Drive，再设一个每月例行的 bot 检查数据备份与可做的分析。Wes Roth 的论证：前期没有任何回报，但一年后你会拥有一份「没有人类会费劲手写、也没有人类会去翻」的决策数据集，届时可以问出「哪一类推理通常导向赢面」。他的一句话总结很准：「这个几秒钟写完的 prompt，可能值不少钱。」

边界与启发

四条可迁移判断：①多智能体的第一性原理是编制而非模型——单点联络（幕僚长）、可交接工序、专职反方（Bear 的可证伪声明值得抄）、人类保留删除权与最终决策权；②通知权限是新的管理杠杆，默认全员静音、只放行把关人；③自动化收益非线性——单件外包省小钱，「最后一块拼图装上传送带整机转动」才是设计目标，所以要从一件每周重复的事起步；④数据护城河在使用中生成，日志与例行归档是零成本期权。限制条件：本期为 SpaceX 赞助、Grokbot 生态迭代极快（他称「几天前刚有一次大更新」、MCP 连接等大量功能未讲），具体功能以产品现状为准；X/Twitter 数据接入需自付 API 费用；所有行情战绩均为纸面交易，不构成投资建议。与他在 Blood Bench 一期展示的「导演式建造」互为印证：人的角色收敛为定目标、定规则、验收与拍板。

转写校正表

转写原文（修正版）

[0s - 120s]

Hey, my name is Wes, and in today’s video, I’m going to show you how to use Grokbot. We’re going to go from beginner all the way to mastery. I’ll show you how to set up your very first bot. I’ll show you how to give each bot a specialty, how to automate all your repetitive work, and build your very own AI workforce. We’ll connect them to your own actual tools and delegate entire projects to them. We’ll teach them important skills. We’ll put recurring tasks on autopilot, and we’ll make sure by the time you get up in the morning that they have already done a full day’s work for you. Now, at this point, you’ve probably heard about how people are using Grokbot to automate a lot of their work. The reason for that is Grokbot is made for that task specifically, from the ground up. It’s built to be kind of like a drop-in remote employee, and not just one. Now, for the purposes of this demonstration, I set up basically an entire hedge fund. You know how hedge funds, those investment funds, they have entire research desks. There are analysts that tear apart every earnings report, clerks who read all the filings with the FTC and all the government, whatever. There’s people that are dealing with legal, dealing with taxes. You have the accountants. You have the people that are kind of are there to tell you when you’re wrong. And it used to be that this would either take a lot of capital to be able to hire all this talent, or it would take you a lot of time if you’re willing to do it all yourself. With Grokbot, you have your own department. You have a chief of staff that’s in charge of actually dealing with the underlings and making sure that they’re getting their work done. The different bots are responsible for different things. They complete their project and they hand off their work to another bot. They run 24/7 in the cloud. They have their own machine that’s in the cloud, so you can talk to them on your desktop or phone or laptop or whatever. Even if you have all your devices off, they are still working. So we’re going to start with the very basics. Feel free to skip ahead. We’re going to have chapters down below. Now, I’ve been using Grokbot since the very first day it came out, and I’ve

[120s - 240s]

I’ve been using Grok for a long time. Before that, so when the SpaceX team reached out to do this video, I was pretty excited. I’m not gonna lie. This video is sponsored by SpaceX. Now, I don’t get any commissions for getting people to subscribe or download Grokbot, but for people that have been watching this channel, you know that I use it, that I like it, and I do recommend it. If you want to try it out, links will be in the description and pinned comment. Go to the website, download it for whatever machine you’re on, set up the account, and just start with one repeatable job. If you’re just getting started, sometimes automating stuff it it seems exciting, but once you kind of get there, it can be a little bit difficult. There’s a certain mindset that goes with it. You have to start breaking up your workflow, the stuff that you do on a day-to-day basis, into little pieces, and then automate each one of those pieces. And once you do, life becomes pretty awesome. Now, I already have a bunch of bots set up and working. So when you get to the screen, this is Grokbot. This is what that window looks like. To get started, it’s very simple. Just click the plus button to create a new bot. Create new bot. Give it some very cool name. Tell it what its life’s mission is. Pass the butter, etc. And you’re ready to go. One thing I do recommend, and this is what I do for most these bots, I do give them a little bit of a personality. I give them a little avatar picture, and hopefully imbue a bit of a personality because yes, they are bots, but they don’t have to be robotic. Sometimes they go too far. This one, for example, is trash talking about beating the S&P 500, but I guess that’s fine. At least it’s motivated. Now, next, you’re gonna want to connect some of the apps that these bots are going to need. Quick note on how this works. So each bot has their own sort of workspace, but they are all on the same machine. So similar how you can have multiple users all on the same sort of like Windows or Linux or Apple, but they each kind of have their own slice of that machine. This is very similar. When you connect your apps.

[240s - 360s]

Gmail, calendar, etc. All the bots will have access to it. Now, in general, if you give one bot access, another bot might technically be able to do it. But if you’re not using it for that, if you’re having to do some separate tasks, it’s not like it’s going to necessarily use that. But it’s important to understand that once you give that permission, once all the bots should have it. Connecting the apps is super simple. Click here on Connect Apps. I’ve already connected a lot of the stuff that I need, like Google Calendar, Google Drive, Gmail. But there are tons and tons of apps that are pre-built that you can just kind of connect with, like Notion, Canva. You have apps for design, code, and data, and sales and finance. Whatever you want, they have it. Now, I’ll delete my new bot for now. Goodbye, new bot. We hardly knew ya. Now, one of the new additions that they’ve recently added is kind of the main bot, Grokbot. It’s your primary general-purpose bot. We’ll talk about that in just a second. One thing I would recommend with whatever bot you start, not a lot of people know that this is a feature. You just ask it to create an email and whatever username you want. What it then does completely automatically is it it goes ahead and sets up Agent Mail with its own email address. Agent Mail is a free service, not for humans, but for AI agents. They get their very own little email address, and they’re able to receive email, send email. This is great because you’re not always going to want them sending email on your behalf. This allows you to have them receive newsletters and information and sign up for stuff and have their own little private email box. Again, you don’t have to set up anything here. Just tell them, and they will do it. This will save you headaches down the road. Now, Grokbot, this main bot is a new addition, and I kind of think of it as an executive assistant. It’s your general-purpose bot. But there’s another very important bot that I think should be either the first bot you set up or the second after Grokbot. Just make sure it’s one of the first ones that you make, and that is your chief of staff. When you just launch Grokbot, it actually might prompt.

[360s - 480s]

To do it, or might suggest that it’s definitely a good idea. As you can see, I have a lot of little bots here, all of them talking to me, and all of them doing different stuff. I generally don’t go and communicate one-on-one with each and every one. I have pretty much one point of contact, and that’s my chief of staff. I tell him what has to get done. He or it, whatever, delegates to everyone else. Also, instead of getting bombarded with fifty million different messages from all the bots, he kind of filters out the stuff that I need to know about. I don’t need to get pinged for every thought that pops into their head. Once you make your chief of staff right-click his face and click pin, that’s how you get him to be up here. So, for example, I can pin this one as well or unpin. That way, he will always remain here, and the bot at the top will generally be the last one that had some notification. And your chief of staff will be able to create whatever new bots you need, manage them, delegate work to them, change their names, give them portraits. At this point, he won’t be able to delete them, which kind of makes sense. You don’t want to accidentally lose that data. You have to, if you want to delete a bot, you have to do it yourself. But outside of that, pretty much everything else should be and can be done by your chief of staff. So once that’s set up, let’s give one of the bots a task to do. I’ll use the Grokbot for this one, and I’ll say, “Go to Reddit and see what are the top investing and stock subreddits talking about.” And I’m going to click go. Once I click go, Grokbot will, on its own computer up there in the cloud, go and do that work. We can actually see it work if we click on its name up here. I’m going to be a little bit all over the place here, but basically on the right-hand side, a panel shows up that gives you some more options like details, library. These are all the files, pictures, if it makes any, and you can also click on computer, and you can see Grokbot screen. Whoa! Flashbang alert! I apologize about that. I need to talk to my Grokbot about dark mode, so it’s on Wall Street bets because, of course, it is. You know what? I’ll put a little sticky note in front of whatever.

[480s - 600s]

They are saying because certain subreddits, you know, you know what I’m talking about. It also briefly visited the subreddit stocks, so it’s going through and basically checking out what are the headlines for today. It’s doing it in real time on its own computer, so you have to have your laptop open. It has its own computer. It’s able to browse the internet how it sees fit, and we can close out of this like that, and it’s going to continue working. You can continue to monitor the situation in the right-hand panel, but while it’s working, we’re going to actually take a look at some of the other bots. So we’re going to take a look at every one of these other bots that I’ve created, and when I say I’ve created, I’ve told the chief of staff to create them. And right before this video started, I’ve asked the chief of staff to have each bot kind of give me a rundown of what it does, what it’s been doing, etc. So let’s quickly go around the room and introduce ourselves. The first one is Hard Code. Now, of course, as you might know, so SpaceX they’ve acquired Cursor. Recently, Elon Musk announced that they’ll also will be incorporating other models into Grokbot. So depending on your request, it might get routed to one of the Anthropic models, for example, or a Cursor model, or Grok, or whichever one will best serve that query. So for example, you say, “Hey, how are you?” That might get served to you know a smaller model that is perfectly capable of saying, “Hi, how are you?” But you can also say, “Build me this crazy functionality, code up a platform that shows all these fancy charts and real-time data tickers, whatever.” And so that’s exactly what Hard Code does for me. This is the thing that it created. So it called it Hedge Desk Arena. So notice you’ve got the stock ticker symbols going across the top. You have the starting cash. Now I gotta mention at this point that obviously none of this is financial advice. All that we’re doing here, none of it is quote-unquote real. I’m not suggesting you trade, buy or sell stocks. I’m not telling you anything in the financial.

[600s - 720s]

Financial realm whatsoever. I’m just teaching you how to use Grokbot. None of this is financial advice. With that out of the way, so this is a paper trading account. We also have kind of a flow of each one of the bots as they surface something that is interesting or relevant that will appear at the bottom. We have our live trade feed, and this is basically a live view of everything that’s happening, kind of at a glance heads-up display that just kind of shows me how everything’s going. So again, since I’ve had the chief of staff kind of summon all these bots on my behalf and give them the prompt, usually right after everything set up, I’ll usually go in there and I’ll kind of ask, okay, what do you do? What are your responsibilities? What do you need to perform better? So here’s hard coded saying my job is to build things. It’s a builder. Right now, that’s the dark mode trading desk dashboard for your video and more broadly scripts, web apps, automations, and repo work. A GitHub repo or Cursor Origin to put this code in would help. So I ask, ask it like, what would help you do a better job? So it’s it’s kind of going well. We could do GitHub or Cursor Origin, which is Cursor’s version of GitHub, specifically built and able to scale for AI agents in mind. And let’s just quickly run through the rest of them. Next one is Clerk. It watches the SEC filings for all of the companies that are on my watch list, and it logs everything. It also specifically watches for certain things that needs to flag if they’re kind of weird looking, if there’s certain red flags, things like insider sales or big deals or anything like that. It also mentions who it works with. So certain bots, they kind of have their own little clusters, and they work closer together than some of the other ones. And I’ll show you in a second how to kind of cluster them together so that they’re able to work more efficiently. But Clerk works with Analyst, Bear, Mapper, and Calendar Mapper keeps track of the supply chain since a lot of this kind of is AI related and the various AI chips, foundry equipment, memory, data centers, power, cloud. Those are important. Mapper.

[720s - 840s]

Is a specific agent in charge of that. Next, we have calendar, the market calendar clerk on this research desk. Its job is to just keep track of all the real-time events like earnings, Fed meetings, inflation reports. Next is bear. Bear is the skeptic. Bear’s job is to call out others when they might be wrong or when they’re missing something. Notice here it’s challenging mapper. So there’s something in that report it didn’t like and it’s flagging it. Next we have analyst, the earnings analyst on our research desk, and there’s just a few other ones that are kind of like more like the supporting roles, but ones that I found very useful. One is keeper. Keeper keeps the various portfolio books. It’s the bookkeeper, the accountant. It keeps the holding sheet, measures drift against target weights, and it turns the gaps into rebalance plan for my sign off. So again, it doesn’t just do whatever things should be done, but it does watch the situation and flag to me or specifically to the chief of staff things that might be an issue, and then the chief of staff brings it up to me if it’s worthwhile. Next is taxman. So this is the bot that does a lot of the taxes and also making sure that we’re using tax advantaged accounts. That is its official job. It also gave itself an unofficial job apparently, which is complaining about California and the various paperwork and taxes that we have to file. And last one, it’s the paper trader. The paper trader’s job is simple. It makes the trades based on what my research team is doing, paper only, meaning it’s not real money. It runs various back tests, so testing different strategies to see how they would have performed if we were running them. And the metric against which we judge it is the S and P five hundred, which is kind of like the big portion of the market, all the big companies. Basically, we’re seeing if we can beat the market. So basically, we just bought S P I and held it. S P I is S and P five hundred. So you can think of the chief of staff on one side as kind of connecting all the dots, connected to the chief of staff, but also.

[840s - 960s]

On the other side, it’s the paper trader because a lot of the the calendar mapper clerk analyst, a lot of the work that they do kind of flows in to the paper trader for the paper trader to make intelligent decisions about what to buy and or sell. So notice here it is kind of talking trash about how much it’s beating the S and P five hundred by, and he’s saying he has some sitting in cash until analyst or bear gives him a reason to put them somewhere, right? So he’s not just kind of throwing it all willy nilly. He’s waiting for the research desk to come up with intelligent ideas and an investment thesis that paper trader will then execute. Now, one thing that I find very helpful is to actually get all the right bots that need to communicate with each other constantly. Get them all onto a group chat. So if we scroll up here and click the plus button, we can then click create group chat, and we’re going to take clerk mapper calendar bear and analyst and paper trader. Now they’re all in one group chat. Now as they finish their work, they’re able to share with everybody else, pass off the work to the next person needs to use it. I’ll say discuss what is the next big trade. Let me see kind of what ideas they have for us. Now they’re going to think about it, so we’ll come back. Meanwhile, Grok bot came back with all of the stories from Reddit. Now that was just to show you how to dispatch them to go do their job. Normally we would have one of these bots doing it so that everything is kept within that kind of tightly knit group of bots that are working together. So notice clerk is pointing out what the different filings say. Specifically, it’s talking about various Nvidia acquisitions, AMD’s acquisitions, etc. Some insider selling, but it’s small and it’s scheduled. Mapper is talking about the various bottlenecks, what looks good, what doesn’t. Calendar posts about all the upcoming earnings, etc. Bear comes in with the pessimistic look, but also saying this is what would prove me wrong. So kind of here’s the things that would kind of invalidate his thesis. And analyst is kind of a little bit more the.

[960s - 1080s]

Zoomed out of view, taking a lot of the other feedback, suggestion, and data into consideration, and based on this, PaperTrader proposes the next big trade, rotating out of Nvidia and to apply the material. So we’ll see how that goes. Now here’s the thing: since we connected everything to Google Drive, they’re all using Google Sheets to collect and label data. So I’m going to say publish all Google Sheets that we have. So I told the chief of staff to make sure that we have all this data available and we can share it with each other to do it through Google Sheets. And this is actually a great reminder why we have the chief of staff, because some of them, Bear, you know, does have all the Google Sheets and has all the checklists, but some of the other ones just were using you know comma delimited files. So they were they had the sheets, they just weren’t putting it into Google Sheets, which is my fault. I didn’t specify. Now instead of talking to all of these agents, I’ll just click to the chief of staff. Also make sure everyone is hooked up to Google Drive and all spreadsheets are uploaded to Google Sheets. Make sure they all format it well. Has to look good. So we’re going to click go and the chief of staff. Now that’s his job to you know wrangle all the cats and to make sure that they’re all doing what they’re supposed to do. Now next, really fast, let’s take a look at routines for a lot of the routines for my research desk. The chief of staff handles all of it and making sure that it runs and setting it up with each individual bot. Like that’s his job responsibility. But if you want to have a little bit more hands on control and just so you know how to do it, here’s how you would do it manually, right? So you select, let’s say, the Grok bot. You click on its name, so the right hand panel opens up, and here in the details you will have various routines that are running. And I’m going to say check the subreddits one hour before market open each day and one hour before market close. Notify me if there are any major news relating to any stocks we are watching, and I’ll click go. Now notice the new routine will show up here in just a second. And by the way.

[1080s - 1200s]

You can have this more often than twice a day. When I’m waiting for an imminent drop, something that I’m expecting, maybe one of the AI labs to release something, I might have it every two minutes, every three minutes, basically checking a website or a blog post or a Twitter account. And the good news is, is this isn’t just a little ping that you get for some predetermined thing. These bots are smart, so it knows what’s on your watch list, and it can kind of judge what’s worthy of your time and what isn’t. So now, basically, with this set up, I know I will get a notification if there’s anything that I have to see, something urgent. This prevents me from being stuck on Reddit all day. Now, by the way, I’m using this on my desktop, but Grokbot is also installed on my phone. Pro tip: make sure that your notifications are set for on only for the bots that are your main bots. If you have 10 different agents, you probably don’t need to hear from like nine of them. Right now, I just have notifications that ping on my phone if it’s something by Chief of Staff or Grokbot. The other big thing to think about is how are you collecting data over time? Because collecting data over time can be a moat, meaning that if you’re routinely mapping everything and analyzing everything and noting down all the signals and saving them somewhere, like for example in Google Drive and the various spreadsheets, over time, you know, when six months or 12 months go by, that data might become very valuable. It might make sense to have one of these agents check, let’s say, every month to go through all the data we have, make sure that it’s there, that it’s getting backed up, and also to see if there’s any analysis that you can run on it. It’s one of those things where it doesn’t pay dividends till much later. In the beginning, there just won’t be anything there. But a year from now, you might have a lot of data collected that you personally you would not have bothered to sit there and write down your little observations and all the data points that, for example, went into making a particular decision. Once you have months or even years of data, you can say, okay, what type of reasoning usually leads to a lot?

[1200s - 1320s]

So what sort of reasoning usually leads to a win? All these little bots will be more than happy to journal all of their thoughts and also kind of what you were thinking, what you were communicating with them, and just jotting it down alongside the investment thesis. It’s kind of a brand new world in a sense because before no human would bother to write all of that stuff down, and even if they did or they had some automated process of just saving everything, no human would bother going through the tons and tons of data to try to pull out some insights. With this system, all of that is automated. See how simple it is. I’m going to tell the chief of staff. I’m going to say, make sure every bot writes down their trades, thoughts, analyses, and anything else that is relevant in a journal. Now we’re going to save in Google Drive. We will use this later to analyze what has been working historically. This one prompt that took just a few seconds to write might be worth quite a bit now that it’s all automated. You really don’t have to do anything else other than just wait. Notice how in our group chat the different bots are working together. Since they’re sharing their Google Drive, each one can take a look at the other one’s work. Notice here, Bear is addressing Mapper directly, and we can do a lot, a lot more. These bots can access Twitter slash X. You’ll need to pay for the API calls. They will just go on X and pull various tweets for you. So I’ll ask, what are the top tweets about Google? What is the sentiment? So it’ll go on X and it’ll kind of take a look around and try to figure out what the sentiment is. And again, if you want that to be running routinely, if you want to fully automate that research, you can ask it to run it every few hours or a few times a day, whatever you think is best. People are pretty bullish on X about Google carbon model leak, and also if you wanted to get notified about certain big events like leaks, AI model leaking out of these AI labs, or a specific VIP tweeting about something, you can set up that here. Now what I’ve showed you here is just kind of like one example of an application that you could use this for, but this could be used to run a business or help you.

[1320s - 1440s]

At work or help you with your college studies. Grokbot can also manage files on your local computer, so not only is able to work on its own computer in the cloud. If you prefer, it can have a little folder on your desktop, and it can put various files there. It’s excellent at organizing stuff. So if you’ve got a folder with a billion receipts in it or old photos that you want to label, just unleash your army of Grokbots on it. If you find that there’s some workflow that you’re doing over and over again, you might consider creating a skill for it. Now, if you wanted to teach Grokbot to do something within the browser, you can, and you do it just by using the browser. It watches what you do and learns from it. To do that, click on the plus button right here, and then teach a task. So it’s rapidly expanding the size of the columns, actually, so because I asked it to make sure that all the information is visible. So actually, it’s working really, really fast. This is happening in real time, so I’ll give it a second here. But with this feature, you can do a task. You can record yourself doing a task, and this bot, chief of staff, learns the steps and can run them again on its own. It’s frantically formatting all the sheets that all the bots have submitted. So let’s let’s interrupt it. And let’s say we wanted to teach it something like how to play Minesweeper. Now, obviously, you probably want something a little more productive, but let’s see if we’re able to teach it this very important to life skill. We’ll click on the beginner mode, and this is how you place the little markers. Well, I kind of screwed up, but hopefully, it’s able to do this better than me. So it’s using vision to go through the entire video, understand what I did. So it’s saying, looks like you searched Google for Minesweeper and started a beginner game. So it’s going to go through the play-by-play and then write it up as a skill. And when you’re making a skill file, those are actually very transferable between the various large language models and agents. So if you have some skills on another agent that you’ve developed over time, you’re welcome to just grab those. They’re basically text files, they’re .md files, and just use them here.

[1440s - 1560s]

So it learned this brand new skill. It’s saying, “Do we want to try a test? " We’re going to say yes and hard mode because why not? A 30 by 16 board with 99 mines. So wish me luck. One of the skills I often use with Grokbot is there’s a couple places online where it has to collect information. So I show it how to go there and collect that information and aggregate information from multiple places online and then put it on this kind of heads up display, this internal dashboard that shows certain key performance indicators. I end up running that probably twice a day, sometimes more. So taking the time to develop a really good skill, and again, it’s just written text. Just like with any other automation, in the beginning, it might take you a little bit more time to train the bot to do it, similar to how you’d have to train a new employee to understand how to do a new task at the company. There’s a bit of a time investment, but once that process is automated, you end up saving a lot of time. Normally, it would take me maybe 20 minutes to half an hour to gather that data. Now I run one command and I just let it run for a little bit, and it does it flawlessly. So Grok is about as good as I am at Minesweeper, but it’s about 10 times better at Excel, which at the end of the day, I think is much more important. And right now, you just saw how I built an entire fully functional, fully autonomous research desk skills, routines, tool use. I still make the final decisions, but these bots basically run 24/7 doing research, analyzing information, and even beating Minesweeper. Apparently, that’s showing some grit right there, good boy Grok. But the point is that I’ve given you the entire toolset that you need to set up something like this for yourself. It doesn’t have to be finance or stocks; it could be anything. If you want a team like this, try Grokbot. I have a link in the description and pinned comment. Start with just one job that you do every week. Just give it to a

[1560s - 1658s]

Bot and give it a name. Once that’s working, add the second piece and then the next. When you’re doing automation, as you’re putting the pieces in place, sometimes it’s hard to see the results early on. I mean, yeah, you’re able to hand off certain tasks that the bot takes care of for you. You have to do a little bit less, but the real beauty of automation comes in when you put in that last piece, because that last piece means that that sort of conveyor belt of automation is finished. And now the entire machine comes humming into life. It’s working and it’s doing real work, whether you’re sleeping or relaxing or whatever. We’re going to be doing a lot more with Grokbot, so please let me know if you want to see any case studies for automation or anything else kind of built out. Is there something that you’re struggling with? Is there something you want to automate but you’re not sure how? Just drop a comment down below and give your bots cool names and cool portraits. I know that might sound silly at first, but it really helps. This is a great ecosystem to start doing the stuff on because of how quickly they’re adding stuff. SpaceX is going at breakneck speed. Just a few days ago, they had a massive, massive update. By the way, we’ve barely scratched the surface of functionality here. All the connected apps, various MCP connections you can have, and the entire thing is getting better and smarter with each new model release. Huge thanks to SpaceX for sponsoring this video. Congratulations to the Cursor team for becoming a part of that amazing company. If you made it this far, thank you so much for watching. My name is Wes Roth, and I will see you in the next one.