A hands-on tour of a synced, phone-controllable AI agent team — agent computers, teachable skills, scheduled routines, event triggers, and where it stops making sense versus Claude Code or Codex.
Posted
5 days ago
Duration
Format
Demo
educational
Views
39.7K
1K likes
57 · 43
Big Idea
The argument in one line.
Grok Bot's edge over a single AI chatbot is giving every agent its own persistent computer and a narrow, well-described job, so agents can operate real accounts, teach themselves new skills from one demonstration, and hand off work to each other without a human relaying context.
Who This Is For
Read if. Skip if.
READ IF YOU ARE…
You already run more than one AI tool for different jobs and want them to feel like one coordinated team instead of separate silos.
You do real work away from your desk, on your phone or in transit, and want the same agents and running sessions there when you open the app.
You're comfortable connecting real accounts, like Slack, Gmail, or GitHub, to an AI agent and want automations that react to events, not just fixed schedules.
You want a concrete, timed comparison of setting up the same kind of automation in n8n versus Claude Code versus this tool.
SKIP IF…
You do deep, single-session build work at a desk — the creator himself says he still reaches for Claude Code or Codex for that, not this.
You need to stay on a specific model provider; Grok Bot runs on Grok models regardless of your Codex or Anthropic subscription.
You're looking for a mature integration library; the trigger list is currently limited to a handful of event types like Slack, GitHub, and Teams.
TL;DR
The full version, fast.
Grok Bot lets you spin up a team of specialized AI agents, each with its own persistent computer, that stay synced across desktop and phone and never go fully offline. Give an agent a name and a one-line job description, and it can be taught a skill by watching you do it once, run on a schedule or in response to an event like a Slack message, and delegate to other agents whose descriptions match the task. Two agents split a real website-restyle job in the video, correcting their own output after one follow-up round. The creator's take: use it for light coordination and checking in on the go, and keep deep building work in Claude Code or Codex at your desk.
Free for members
Chat with this breakdown — free.
Sign in and you get 23 free chat messages on us — ask for the hook, quote a framework, find the exact transcript moment, generate a markdown action plan. Bring your own key when you want unlimited.
Cold open and product framing: Grok Bot pitched as a synced team of always-on AI agents, shown via its 'Meet Grok Bot' onboarding screen.
00:56 – 03:05
02 · Bot Setup & Computer Use
Walkthrough of the app layout, a sidebar of agents (Klaus, Dev), each agent's own operable computer, and how a recorded skill (liking Skool community posts) gets reused.
03:05 – 07:06
03 · Connectors & Morning Routines
Creating a new specialized agent from scratch, connecting Gmail and Google Calendar, and having it build and run a weekday 7am morning-briefing routine.
07:06 – 11:51
04 · Self-Improving Skills & Triggers
The Skool like-skill runs a dry run, catches its own ambiguity around like/liked states, and rewrites its own instructions without being asked.
11:51 – 13:37
05 · Shared Skills & Agent Messaging
Setting up an event-based Slack trigger for a new media/sponsorship agent, and showing how skills and connectors are shared across every agent in the account.
13:37 – 14:36
06 · Mobile Control
The same agent conversations and running computers appear on the phone as on desktop; Klaus and the newly named Money agent keep coordinating a status check on a video project.
14:36 – 16:01
07 · Limitations & Best Use Cases
The creator draws the line on when he'd actually use this: light coordination and checking in while mobile, not deep flow-state building, which stays in Claude Code or Codex.
16:01 – 18:56
08 · Multi-Agent Website Build
A live two-agent build: Klaus delegates a waitlist-page build to Dev, who ships a default-branded prototype, referencing the creator's own AI Automation Society waitlist page.
18:56 – 20:08
09 · Avoiding Agent Hype
After Klaus fetches the real brand guidelines and Dev restyles the page to match, the creator warns against adopting agent tooling just because it's fun, urging viewers to start from an actual pain point.
20:08 – 20:32
10 · Final Thoughts
Sign-off and a like/subscribe ask.
Atomic Insights
Lines worth screenshotting.
Grok Bot gives every AI agent its own persistent computer, so it can browse, log into services, and stay signed in exactly like a human would.
Recording yourself doing a task once is enough for Grok Bot to turn it into a reusable skill the agent can run again on command.
Connecting one third-party account, like Gmail, to a single agent makes that connection available to every other agent in the workspace automatically.
A morning-briefing routine that pulls today's calendar and flags actionable email took about two minutes and one prompt to set up.
Agents can trigger off events, not just schedules: a new Slack message in a specific channel fired a notification with zero manual steps after setup.
The same kind of automation that took roughly fifteen minutes to wire up in n8n and about seven minutes in Claude Code took about two minutes in Grok Bot.
After running a skill once, the agent reviewed its own output, found an ambiguity, and rewrote the skill's own instructions without being asked.
Every agent should have one narrow job and a specific description, because that description is what other agents use to decide who to delegate a task to.
Agents can message each other directly, and neither agent can keep that conversation private from the human tracking the work.
Everything runs in the cloud, so a phone session shows the exact same agent conversations and running computers as the desktop app, with no local sync step.
For deep, flow-state building work, the creator still reaches for Claude Code or Codex; Grok Bot is what he opens when he's away from his desk.
Grok Bot runs on a Grok model under the hood, not the user's existing Codex or Anthropic subscription, which matters if you have a strong model preference.
When an agent wasn't told which brand assets to use, it picked a plausible-looking default rather than stopping to ask, and needed a follow-up correction to match the real brand.
As of this recording, Grok Bot ships a narrow set of trigger types, Slack, GitHub, and Teams among them, a much smaller set than a tool like n8n offers.
The creator's standing advice: don't adopt a multi-agent tool because it's fun to play with, adopt it because it removes a pain point you're already feeling.
Takeaway
How a synced team of one-job AI agents actually gets work done.
AGENT WORKFLOW
Grok Bot's real value isn't the chat window, it's giving each agent its own persistent computer and a narrow, well-described job, so agents can improve their own skills and hand off work to each other without a human relaying context.
01Meet Grok Bot
The core pitch is a team of agents, not a single assistant: each one is on call, synced across devices, and never offline.
Setup starts with a name, color, and shape for the first agent, framing every agent as its own persistent identity rather than a disposable chat thread.
02Bot Setup & Computer Use
Each agent gets its own computer it can operate, and the human can also open and drive that same computer directly.
Recording a manual walkthrough once is enough for the agent to convert the steps into a reusable skill it can run on its own later.
Signing into a service inside an agent's computer keeps that login persisted, so the agent doesn't need to re-authenticate on every run.
03Connectors & Morning Routines
One connected account, say GitHub or Gmail, becomes available to every agent in the workspace, not just the one that connected it.
A working morning-briefing routine, pulling calendar and email into a short daily plan, took about two minutes to define in plain language.
Because routines run in the cloud, they fire on schedule even if the phone and computer that created them are both off.
04Self-Improving Skills & Triggers
Naming an agent something specific and giving it a one-line job description is what lets other agents find and delegate to it correctly.
After a skill ran once, the agent noticed its own mistake, an ambiguous button state, and rewrote its own instructions to fix it, unprompted.
Event triggers, a Slack message, a GitHub event, let an agent react in real time instead of waiting for a scheduled check-in.
05Shared Skills & Agent Messaging
Skills work like slash commands across every agent in the account, so one agent's trained skill is available to all the others.
Agents route work to each other based on their descriptions: a generalist agent pings the specialist whose job matches the request.
Agent-to-agent conversations aren't private from the human, asking one agent what happened surfaces what it discussed with another.
06Mobile Control
The phone app mirrors the desktop state exactly, same conversations, same running agent computers, because everything lives in the cloud rather than on one device.
You can drive an agent's browser session directly from a phone, which matters for quick checks away from a desk.
07Limitations & Best Use Cases
The creator reserves deep, flow-state building work for Claude Code or Codex at his desk, and uses Grok Bot for lighter checking-in and coordination while mobile.
Grok Bot runs on a Grok model, not the user's own Codex or Anthropic subscription, worth knowing if you have a strong model preference.
The honest self-assessment of what you will and won't actually use a tool for is more useful than a feature list when deciding whether to adopt it.
08Multi-Agent Website Build
Two agents split a build task, one delegating and reviewing, the other producing the output, and handed context back and forth without the human relaying it manually.
Left without explicit brand guidelines, the agent picked a plausible-looking default rather than stopping to ask, which produced a result that missed the mark.
Once pointed at the real brand assets, the same agent pair corrected the output in a single follow-up round.
09Avoiding Agent Hype
The standing caution: don't force agents and routines into a workflow that doesn't actually have the problem they solve, or the setup time becomes the cost.
The better filter is naming a specific, felt pain point first, then checking whether the tool addresses it, rather than adopting a tool because it's novel.
Glossary
Terms worth knowing.
Agent computer
A persistent virtual computer each Grok Bot agent gets, which it can browse and operate, and which the human can also open and drive directly.
Teachable skill
A repeatable task Grok Bot turns into a reusable skill after watching a human demonstrate it once, so the agent can run it again on its own.
Routine
A scheduled action an agent runs on a set cadence, such as posting a morning briefing every weekday at a fixed time.
Trigger
An event, such as a new Slack message or a GitHub event, that fires an agent action immediately instead of waiting for a scheduled check.
Connector / plugin
A linked third-party account, like Gmail or Slack, that becomes available to every agent in the workspace once any one agent connects it.
Resources
Things they pointed at.
01:35productCursor Ultra plan
02:12toolGitHub
04:30toolGmail
04:32toolGoogle Calendar
07:34toolSlack
14:50toolClaude Code
14:51toolCodex
15:02productHermes agent
Quotables
Lines you could clip.
00:32
“It kinda feels like you've got Cloud Code, Codex, and Hermes agent all in your pocket.”
compresses the whole pitch into one comparison line→ TikTok hook↗ Tweet quote
07:10
“Without me even asking it to, after it ran the skill, it basically gave itself feedback on what worked and what didn't, and then it fixed itself.”
the self-correcting skill moment is the clearest 'wow' beat in the video→ IG reel cold open↗ Tweet quote
13:16
“They're not gonna be able to keep things secret. They're gonna snitch on each other if they're, you know, being weird.”
funny, memorable framing of agent-to-agent transparency→ newsletter pull-quote↗ Tweet quote
14:36
“If I'm sitting down at my desk and I am working on knowledge work, if I'm trying to build things, I will not be using this.”
an unusually blunt limitation admission from someone demoing the product→ IG reel cold open↗ Tweet quote
19:30
“Really think about what is constraining your business, what are the pain points that you're feeling on the day to day, and does this tool solve some of those pain points?”
closing line works as a standalone piece of advice about any new tool, not just this one→ newsletter pull-quote↗ Tweet quote
The Script
Word for word.
Read-along
Don't just watch it. Burn it in.
See every word as it's spoken — crank it to 2× and still catch all of it. The same dual-channel trick behind Amazon's Kindle + Audible.
17px
metaphor
Okay. So we just got GrockBot, which is probably the easiest way that I found to be able to spin up teams of different AI agents that you can use from your phone, on your computer, everything stays synced, and they're always on. It kinda feels like you've got Cloud Code, Codex, and Hermes agent all in your pocket.
It's super easy to set up, so let's just jump right in. So what you're gonna wanna do is go to Google, type in Grokbot, and then click on this link, and download this for Windows, for Apple, and for your iOS device so everything stays synced together. Now this is super cool.
This is what it's gonna look like. You're gonna have a ton of different agents on the side, and you can talk to each one individually, and you can have them actually talking to each other, which is super cool.
Each of the bots have their own computer, which is really cool. You can sign in to a session, and it will keep that login. You can record yourself doing something on the bot's computer, and then it will be able to just replicate it like a skill.
And there's so many other things. They get smarter over time. It's super cool.
Now in order to try this out, you are going to have to be on the cursor ultra plan. So get the ultra plan, then you can download Grok bot and play around. Okay.
So when you open the app, this is kind of what it looks like. It's gonna first walk you through like, hey. This is Grok bot.
Create your first bot. You can, you know, give it a name and a color and a shape and all this. And then once you're actually in here and you've made a few, it's gonna look like almost like an iMessage channel or like a Telegram channel.
You have different bots that you can switch between. On the right hand side, you're gonna have settings. You can give them a name, a title, a description, and then you can also look at their computers.
So right here, you can see that Klaus, my agent right here, is inside of my school community logged in as me, and it's going through and it's liking posts that I wanted it to like. So it's controlling this browser, and I also can too, though. If I wanted to come in here and I wanted to, for example, go to YouTube, I could log in as myself if I wanted to, and then it would stay logged in.
It's even making me do this little Capscha thing. Wow. This thing is really not letting me in.
Okay. There we go. Now I can finally get through to YouTube.
You can see we can also open up right here a file manager. We could open up the terminal, and we have this teach a task button, which is where I could record myself doing something. And then Grokbaw would automatically turn that into a skill, which is exactly what I did right here with this AIS skill that it created.
And now it has this one called, like, seven day challenge post on school. It went ahead and ran it. And I've also got this thing set up with my GitHub repo, my Herc two, as you can see, because that's where I've been building my AI operating system.
So if you guys are building your own second brains and AI operating systems, you can bring that into GrokBot, and all of your agents that you build can automatically have that context so you're not starting over. But if you don't have an operating system, that's completely fine because when you create a new agent, it basically onboards you.
It asks you some questions, it says, hey. What do you want me to do for you? So right here, can see I've got Klaus.
He is my executive assistant. I've got this dev who is an AI engineer. And what's cool is if I want Klaus to help me, like, build a product, Klaus will actually delegate work to my dev because it knows what they do.
So, basically, when you have an agent, you give it a name, a title, and a description. And the description's basically almost like the same way agents use skills. They say, okay.
Cool. I have to do this task. Let me look at all the other agents and see if there's any agent whose description matches task.
And if so, I will just shoot them a message, and then they will help me out with that instead. So each agent that you build can be very specialized to do very specific things. If I come here to click on new chat, I can start a chat with one of the bots.
Or if I wanna go ahead and create my own new bot, now I have the ability to set this thing up however I want. You can see right here it says, hey. Good to meet you.
What do you want me around for? Research and writing, something specific, day to day work, whatever you want. So let's just make this agent like my morning briefer.
So I just want you to help me every morning, plan my day, and look at what I have to do. And you can see it says, okay. Cool.
I'm checking what's already connected so I know where to pull tasks from. And the only thing that I've connected so far, if I go to the plug ins, is GitHub. I need to connect things like Gmail, Google Calendar, Slack, Google Drive, ClickUp.
All of these other things that I use, I need to connect, and it's just as simple as a little sign in. But what's cool about that is if you connect GitHub for one agent like my Klaus or my dev, then all of your different agents will be able to use that connection. So they share those plugins.
It asked me where my to dos and calendar live, and I just went ahead and say Google Calendar and and Gmail. And now it's gonna have me actually connect those plugins. It's asking me if I wanna connect these.
It says you'll get a quick sign in, and then it's a one time setup. So I'll click yes, connect both. And so it just pops up like this, and all you have to do is click authorize, and then you just do your sign in as you would in lots of other apps.
Okay. So we just connected Gmail and calendar, and now it's asking when do we actually run these morning routine things. So it's weekday mornings by default.
I'll pull today's calendar and anything important in Gmail. I'll just go ahead and say 7AM. Now what does this mean?
This means that your agents can actually create routines. And because all of this is happening on the cloud, this will run even if my computer's off, even if my phone's off. You can see right here, created routine called morning day plan, which shows up right here in the agent section.
So these are the instructions. Build Nate's morning plan and send it in the chat. Look up current MCP tools, calendar, Gmail, pull today's calendar events, pull actual Gmail items, write a short description, blah blah blah.
And, obviously, we could tweak this instruction if we wanted to. It asks if we want a sample. Yeah.
I'll say run one right now. So this is so cool because it makes it so easy to build automations, to have different agents.
And as you can see in here, this one's still called new bot. If I click on it right here, I can give it a name. So I'm just gonna call this one coffee because it's a morning skill.
I will just say my morning planner. And then you wanna give this agent a super specific one liner of what it actually does so that if Klaus or if my dev or other agents needed to use it, it would know to use this one.
This agent looks in my Gmail and my calendar to help me plan my day every single weekday. And then if you also want to play with the way it looks, you can change the shape and the color in here. You could also generate it by describing it, or you could upload some sort of image.
And here's the cool part. Remember how earlier I showed you guys how Klaus has its own computer, and right now it's open on school as you can see or YouTube, I guess? Coffee has its own screen, and so does dev.
So every time you're building things with different agents, they have their own computers, and they can do their own things. It's very cool. Okay.
So here is my sample for tomorrow, what I'm gonna get in the morning. It's showing my day. It's showing the meetings I have.
It says here are things worth acting on from email. So I've got this thing that I need to get back to. I've got this call on Thursday that I have to respond to.
I have got this chat invite. Very cool. And then it's suggesting things like use seven to nine plan day to clear this item that you have to skim this notes and prepare for that call.
Awesome. Your afternoon is pretty stacked, so pick one of these things, shorten your lunch and gym, or move the gym block. Cool.
It even told me to protect my evening because I have a dinner reservation tomorrow night. So that's super cool because think about this. I didn't tell it anything.
All I said was help me plan my day. And once I iterate on this and say, hey. By the way, here are some things I liked about this.
Here are some things I didn't like. It's going to get better and better and smarter over time. And think about how quick that whole skill and whole routine was for me to set up.
Alright. So I'm gonna jump back over to Klaus real quick. Take a look at this.
I had it do the skill where it went through my free school community and liked posts that were for my seven day challenge. And here's what it did after one run. It said, okay.
Here's the dry run. Here are the three posts that I liked. Shout out to you three community members.
You guys are awesome. And then it said, hey. Here's something I noticed.
It's not very clear. I can't tell very clearly if they're liked or not already. And then it said, okay.
Cool. I went ahead and I put open the post before liking into the skill. So without me even asking it to, after it ran the skill, it basically gave itself feedback on what worked and what didn't, and then it fixed itself.
So over time, it's going to get smarter every time it runs the skills, and it builds more context and more memory over time as you talk to it more and as you do more things. Now you guys saw how quickly and easy it was for me to create this routine over here with coffee, and what I did here is I ran this based on a schedule.
So I ran this, you know, weekdays at 7AM. But what else you can do is you can run things here on actual triggers.
So on a Slack message, on a GitHub event, on a Teams message. These are currently the triggers right now just the same way in n n n, you had all these different, you know, Gmail triggers and things like that. I'm assuming that these are going to become much, much more broad.
Like, there's gonna be hundreds and hundreds of triggers in here, but right now, we're starting off with just these six. So watch this. I'm going to create a new agent real quick, and I'm gonna call this one sort of like my media agent.
And I'm gonna connect this agent to Slack and so that it can basically notify me when there's things going on in the Slack channels that have to do with sponsorships or new videos or new opportunities, anything to do with the media side that I deal with in Slack with a few different teams. You're gonna be helping me manage media, sponsorships, things like that all within Slack.
And so you notice how I'm spinning up agents to do very, very specific things. Even on their website, on GrockBot, it's basically showing you that agents should be doing one specific job. Sales outbound, talent scout, paid media, expense manager.
If you try to make one mega agent do a ton of different things, it's just gonna get confused, and all those there's gonna be all these routines. It won't be as good. But if you have each agent do one specific job very well, and because you have good descriptions and because you have the ability for agents to talk to each other, the jobs are going to still get done.
It's just the same way in Cloud Code or Codex. You like to have your main orchestrator, and you like to have them basically delegate work to a bunch of different little sub agents. So here's the Slack card that popped up.
I'm gonna go ahead and authorize into Slack real quick. It's asking what Slack channels it should treat as home base. I'm gonna ask it what channels can you actually see and read.
Because if you're like me in Slack, you might be in a ton of different workspaces and channels and things like that, so it's just worth seeing what it actually knows. Now look at this. It basically was only able to see some of the stuff inside one workspace, but I have another email with some other channels I'm in.
Right? So I said, hey. I wanna connect another one, and it just lets you add another one.
So here we have a default connection, but now I also have one that I'm calling work, and I can sign in with a different account here so that I can use different channels. There we go. Now I have authenticated another Slack account, and it should be able to see everything that I'm doing in there now, which is great.
Now my Grock bot can see into all of these public channels as well as these private channels in DMs, which is just great. So what I want you to do is help me set up a routine that's on a Slack trigger. So inside of the default Slack workspace, what I want you to do is when you get a message from the YouTube testing Nate channel, I want that to trigger you to basically give me a notification on what happened.
So let's see if it's able to understand what I'm saying and create us a routine inside of its settings that's based on a trigger rather than a time based routine. Okay. Cool.
So it created a routine called YouTube testing Nate alerts. Perfect. You can see right here, it says when to run on new messages in the YouTube testing Nate channel.
So it founded the right channel. It created the right trigger, and it set up the actual instruction. So before this works, it says one setup step so it actually fires is in that channel.
You have to run slash invite at cursor. So let me just copy this command. The channel listeners only hear channels the Cursor Slack app is in.
So let me just go ahead and try that real quick. I'm inviting at cursor. There we go.
It was just added. And now in here, let's say, hey.
A new sponsorship deal has come in for 1,000 million billion dollars for, um, a YouTube video. The company is called Chipotle.
Okay. So I shot that off. Let's see what's going on inside of Grock bot.
Okay. So we're seeing that it's having the little new bot is working animation. It says new bot because I haven't named this yet.
I haven't touched anything besides sending that message in Slack. Boom. New in YouTube testing, Nate.
You posted that a Chipotle YouTube sponsorship deal came in for this much money. Cool. So that's how you can see that that routine is working, and that got set up so quickly.
With n n n, that might have taken me maybe fifteen minutes. With Cloudco, that may have taken me, like, seven minutes.
With this, this took, like, two minutes and one prompt. It's pretty insane. And remember, you just wanna give your bots a name.
I'll call this guy money. The title will be Slack media, and the description is basically, this agent is in charge of managing my Slack communication for media and sponsorships, things like that.
Now if you wanted to reference some of the skills or routines that this agent uses and giving it more context for other agents, that's perfectly fine. And it even acknowledged that I named it money, and it said money works. I'll answer to that from here.
Okay. So what about skills? How do skills work in here?
Well, if you do slash, you can run skills like slash commands. So the only one that we've currently built, I believe, was that school one. So if I type school, it says like seven day challenge posts on school, and I could go ahead and use that skill if I want to.
So skills work across all the agents that you have. Same thing with plug ins, but the computer sessions are different, and the descriptions of your agents are different. Okay.
Now watch this. I'm gonna go to Klaus, and let's pretend, you know, because this is kind of like my executive assistant. This is what I'm gonna be assuming as, like, my router to different agents.
I'm just gonna be like, hey. I kinda forgot what's going on with one of our most recent YouTube videos, which you might be able to find in the Slack channels.
Maybe you should just check-in with Money to see if he has any information about, you know, what's going on with our anthropic reel or just what's been going on with that. So what this should do is it should basically tell Klaus that, you know, I'm just an executive assistant, but I can see that Money is a Slack media agent.
So I'm gonna go ahead and shoot a message off to Money to take care of that and then come back to me, and I'll answer to Nate in this main session. There we go. It said pinging Money about the Anthropic Reel.
It said message Money. And if I click into this, we can actually see the conversation. It's a view only, but we can see what Klaus said to money, and we're gonna be able to see money responding to Klaus right here.
Can you share the latest status from Slack or whatever you have? Any blockers, next steps, blah blah blah. Keep it tight.
I'll relay it to him. So there you go. You can see Money just responded to Klaus.
And then if I go back to the main Klaus, and if I go back to my conversation with Klaus, it just responded right here. And what else is cool is if I go to my conversation with Money, it says, hey.
By the way, Klaus asked me about this. So they're not gonna be able to keep things secret. They're gonna snitch on each other if they're, you know, being weird, but that's how that works, and I think that that is a super cool feature.
And I think that that really encourages you to make sure that each agent is very specialized because they're able to easily talk to each other. So at this point, I've built out these four agents. And what if I wanted to turn off my computer and go for a walk, or maybe I'm even going on a vacation?
Well, guess what? On my phone, I've got this exact same stuff. So if I open up Klaus on my phone, I have the exact same messages, and I can just say, hey there.
And what's gonna happen is that's gonna come through there. I just sent it off. It comes through in real time.
So everything is completely synced because this is all in the cloud. You don't even have to worry about spinning up your own VPS or doing anything locally. It all is just syncing perfectly.
The connectors still work. All of their desktops still work. So if I go in here and I go to the browser, remember how Klaus was over here on YouTube, I can come into here on my phone, and I can also see what's going on in here.
And I can control this with my phone. So in right here on my phone, I'm typing in, um, Nate Herc. I'm gonna go ahead and search for Nate Herc videos on YouTube, and I'm able to do this on my phone.
So I'm controlling a cloud computer from my phone, which is kind of being managed over here by Klaus. It's just very, very cool that this is being set up so quickly for you. Okay.
So let's talk about what else we've got in here and what are some actual limitations. So the first thing is, how much am I gonna use this? Will I actually use this?
Well, if I'm sitting down at my desk and I am working on knowledge work, if I'm trying to build things, I will not be using this. I will be using my codex, my cloud code. That's what I like to drive with when I'm sitting down at my desk.
Now if I'm traveling, if I'm in the back of an Uber, if I'm on a vacation, if I'm sitting at a restaurant, this is what I'll be using because my phone syncs to it right away, and it just keeps things super clean and super easy. That's kind of the way that I distinguish between Hermes agent right now and Claude Code or Codex is I just like to use my Hermes when I'm on the go.
And I think that using this on the go is honestly much better and much easier. Now, obviously, this on the back end is using a Grok model.
Right? It's not using my Codex subscription or know, Anthropix APIs. It's not using those models.
So if you really, really like the other models way more and if you notice that you don't like the way that the Grok models interact, then maybe you won't wanna use this as much. But the way that I'm thinking about this is I'm not gonna be building crazy software, crazy apps from here. I'm more so gonna be doing things like checking in on things, communicating to the team, reviewing things, setting reminders, things like that.
Just very general knowledge work. And, obviously, I'm gonna save my deep dive, like my deep I I need to be in flow state for this kind of stuff. I'm gonna be saving that for when I'm sitting down at my desk.
Okay. So let's see what else we have when we think about, like, other settings and things. But real quick, I'm just gonna send off this message to Klaus.
I'm asking him to basically spin up a waiting list page, and I told him to work with our AI engineer.
So it's looping in dev. They're gonna get to work. Hopefully, we're able to see something on a local host on their cloud computer in a bit.
While they're working on that, let's go over here to our settings. So I'm gonna click on my name. I can see that I have weekly usage.
So I've used 2%. This resets in seven days, and we can also, you know, spend money.
We talked about getting this on iOS. And if I go to the settings, let's see what else we have. Your theme, your time zone, execution on local computer.
You can change the permissions to be ask every time or never allow. You can set rules so when Grock bot wants to do certain things, you should have it ask first or automatically allow. Wow.
That prototype took, like, no time at all. Let's see if we can open this up. So if I click on local host, we have it right here.
Get certified in a automation without the overwhelm. So this is interesting because it has good data, but it didn't correctly use, like, our brand guidelines and our logos. So I wonder what the issue is there.
I wonder if it's because it couldn't find it. I noticed that you didn't really use our brand guidelines as far as color scheme and logo.
So just wanted to know why you didn't do that. And you know what's interesting about this? It created this waitlist page in our local host because this is on my Google Chrome.
Whereas if I wanted it to do it on its own localhost, that's different. If it wanted it to do it on its computer, it would have to build that in a different way. So as you guys know how localhost works, that's only gonna work on one local device.
And that's kinda cool that it created it here locally for me on my actual desktop rather than doing it on their cloud computer. It could do that, but we just would have had to specify. And then Claus said, yeah.
That's on me. When I briefed dev, I said to pick a sensible default if branding wasn't handy, and they assumed that AIS ish was navy and indigo. I didn't point them at brand assets or HERC two packaging.
Do you want me to do that again? I'm just going to say yes, and we'll take a look at that version. But, anyways, that's basically it.
It's really just a matter of getting in here, connecting in some of your business context, and then just building out a bunch of essentially different skills, would say. Different skills, different sub agents, and then just seeing how you can have them work together and just spin up your routines as quick as you can because they're super quick and easy to set up.
If you wanna take a look at all the other plugins, you just click on plugin. You can scroll through and search for them. You wanna connect certain custom ones, you certainly can, and you can also see all of your actual skills in here.
These are skills that I have made. I've only made one so far. This is where we can update the name, the description, the instructions, and you can also go ahead and delete them.
Now look how cool this is. Klaus found the brand guidelines, and then it messaged Dev.
And here is the new message and the chain between Klaus and Dev. They talked about the prototype.
Dev delivered it. And then Klaus came back and said, hey. Here's the actual brand guidelines.
Here's the logo and the brand guideline doc. And now Dev is saying, hey. I got this from Klaus.
I have to restyle the wait list, blah blah blah. So I think you guys get the point. They're talking to each other.
We have routine set up. We can keep building these things really easily, and think that this is just super cool that I'm able to continue my work on my phone super easily.
Now one thing that I think when you sort of look at tools like this is you don't wanna get carried away or caught up in the hype of these tools because, yes, they're cool and they're fun to play around with, But don't just start forcing all these agents and routines that you don't actually need. Really think about what is constraining your business, what are the pain points that you're feeling on the day to day, and does this tool solve some of those pain points?
For me, I'm gonna try to set up all of the Hermes agents that I have right in here, and think it's gonna feel more cohesive because they can talk to each other more easily than the way I have them setting up talking to each other in Hermes. But if you actually don't have a need, if you don't have a problem to solve with this tool, then it might just be a waste of time if you spent, like, all week playing with it.
You know, I think it's cool and it's fun, but if you really wanna talk about, like, if you need it, it's good to sort of think about that element. Right? So this should be done.
I'm gonna go into the wait list, and I'm gonna hard refresh. Boom. Okay.
This looks way more on brand. There's still some design things that I don't love about it, but all I said was, hey. Build me a super simple wait list form.
I don't even know if this actually leads anywhere. We would have to, like, keep iterating if we wanted the dev to actually make this go to ClickUp or something like that. We definitely could.
But on a first pass, you can see that it was able to look inside of my repo. I had had it had all the context of what's going on with this program, and it built it out for me super quick. But, anyways, that is gonna do it for today.
I hope that this showed you what's possible with this tool. As they release new updates, I'm gonna be excited to cover them. Hopefully, we get some cool stuff.
Hopefully, we get the automations and the routines to become easier and easier to build with different triggers. I'm super excited to see how this tool evolves. So if you enjoyed or you learned something new, please give it a like.
It helps me out a ton. And as always, I appreciate you guys made it to the end of the video, and I'll see you on the next one. Thanks, everyone.
The Hook
The bait, then the rug-pull.
A creator opens with a bold claim: this is the easiest way he's found to run a whole team of AI agents from his phone. What follows is a hands-on tour of what that team can actually do, and an unusually honest read on where he says it still falls short.
Frameworks
Named ideas worth stealing.
08:14concept
One Agent, One Job
Give each agent a narrow, specific job and a one-line description. That description is what lets other agents find it and delegate work to it correctly, instead of one overloaded agent trying to do everything.
Steal forstructuring any team of AI agents, or Claude Code sub-agents, so delegation resolves correctly instead of routing to an overloaded generalist
CTA Breakdown
How they asked for the click.
VERBAL ASK
20:08subscribe
“if you enjoyed or you learned something new, please give it a like”
soft, low-pressure ask stacked with genuine thanks, placed after the full teardown rather than before it
FROM THE DESCRIPTION
PRIMARY CTAWhere the creator wants you to go next.
Nate Herk breaks down Boris Cherny's YC interview on cutting 80% of Claude Code's system prompt, then tests deleting his own skills to see what actually changes.
Nate Herk breaks down the four ways an AI operating system's context quietly goes wrong, then walks through the five habits that keep a growing second brain accurate instead of confidently wrong.
00:23
00:38
00:53
01:20
01:40
01:53
02:10
02:18
02:41
02:57
03:27
03:43
03:58
04:14
04:29
04:44
05:00
05:15
05:31
05:46
06:01
06:17
06:32
06:48
07:03
07:34
07:49
08:05
08:20
08:28
08:53
09:06
09:22
09:37
09:52
10:08
10:23
10:39
10:54
11:13
11:40
11:56
12:11
12:26
12:42
12:57
13:13
13:43
13:59
14:13
14:45
15:00
15:17
15:31
15:47
16:02
16:17
16:27
16:48
17:04
17:14
17:50
18:05
18:21
18:36
18:56
19:07
19:22
19:53
20:07
Video of the Day1:08:12
16:27
11:56
Video of the Day30:53
25:03
05:23