The whole idea in one line: you hand over a job, you leave, the work gets done, you approve it.
§ 1I ────── what actually launched
Grok Bot is a team, not an assistant.
It's a desktop and iPhone app where you create AI teammates — each one with its own name, its own job, and its own computer in the cloud.
What you're watching: xAI's own launch page. On the left is the bot roster — Chief, Sales Outbound, Inbox Manager, Talent Scout, Expense Manager. On the right, one of them working.
Tweet 1 · the launch
xAI put it in one sentence
Bots sign in to your tools, use them the way you would, and come back with finished work.
10.4 million views in a day tells you how many people want that.
§ 2II ────── each bot gets its own computer
This is the part that changes everything.
Every bot runs on a real computer in the cloud that never shuts off — so closing your laptop doesn't kill the job.
The reason agents died all last year: they lived in your session. Give one its own machine and the job outlives you closing the lid.
Thinking it?"There's no way I can set something like that up."
There is no code anywhere in this. You create a bot, you type "handle my inbox", and it asks you what you want.
If you can send a text message, you can run one.
§ 3III ────── you message them like people
Say the job. Then go to the gym.
You message a bot the way you'd message someone on your team — "clean up my inbox", "research these leads and draft outreach in my voice".
Tweet 2 · close your laptop
The bit that makes it feel like a teammate
Give them a task, shut your computer, and reach them from anywhere — phone included.
Your day stops. The work doesn't.
Your business used to only move when you were at your desk.
§ 4IV ────── the overnight sales run
52 accounts in. 36 drafts out.
Straight from xAI's launch demo: the sales bot pulled the account list, researched every company, checked the CRM, and wrote outreach overnight.
What you're watching: the bot's own checklist ticking off — Salesforce list pulled, 52 accounts · LinkedIn, 4 profiles skipped because they were recently contacted · 36 drafts queued, 0 sent.
Every number here is xAI's, from their own demo. The one I'd tattoo on the wall is the last one.
§ 5V ────── the leash
Nothing goes out until you say so.
The bot queues the work as drafts and waits. In the demo, 36 messages sat there until a human said "the top 10 look good, send it".
The design choice that makes this usable in a real business: draft, queue, approve. Never send-and-pray.
Thinking it?"I can't trust an AI agent with my real business."
Fair. So look at how it's actually built — 36 messages drafted, zero sent, and the bot even flags what it skipped and why.
You're not handing over the keys. You're handing over the typing.
§ 6VI ────── what people are using them for
Vendors, support tickets, and a CRM that stays clean.
xAI says beta users already have bots negotiating with vendors in their voice, running support for online stores, and keeping the CRM up to date — their claims, but they match what the product clearly does.
01Negotiating with vendors
In your voice, in your tone, on your terms.
02Running store support
Tickets answered, hard ones escalated to you.
03Keeping the CRM clean
The job nobody in your business wants to do.
§ 7VII ────── show it once
Do the job once while it watches.
This is the feature that made me stop scrolling: you do the task yourself, the bot follows along, and it saves what you did as a routine it runs on its own.
What you're watching: "Weekly Reporting is watching and learning" on xAI's page — the bot recording a workflow while a human does it, then running it on schedule.
§ 8VIII ────── they talk to each other
Put a few in one thread and they pass work between themselves.
In the demo the sales bot pulled account notes from the account manager bot and priority flags from the chief of staff bot — with the human touching nothing.
What you're watching: "Connect the Bots" on the launch page — one bot looping in another, live, while you watch instead of approving every step.
So that's Grok Bot. Now — does it destroy Hermes Agent?
§ 9IX ────── the honest reaction
People with early access are hooked.
That's worth saying out loud before I compare anything — the polish here is real.
Tweet 3 · early access
The comparison everyone landed on
Lenny Rachitsky got in early and called it the most exciting AI product in a while.
His framing matters for this video: like OpenClaw, but easier and less scary to use.
§ 10X ────── the framework
The Ghost Team: four things, or it isn't one.
A Ghost Team is a team that works while you're not there — so your business moves at 2 AM, at the gym, on a Sunday. Every one needs these four.
Judge any agent tool with these four and the marketing falls away in about ten seconds.
§ 11XI ────── grok bot's scorecard
Grok Bot checks all four, out of the box.
That's the honest verdict — nothing here is missing.
HandsBots sign into your email, your CRM, your LinkedIn — on their own cloud computer.
MemoryBots keep context on how you work and get smarter the more you use them.
ScheduleRoutines like a morning briefing at 8 AM, or inbox cleanup every weekday at 6 PM.
LeashThe draft-and-approve queue you just watched.
§ 12XII ────── hermes' scorecard
Hermes checks the same four — a completely different way.
It's open source, from Nous Research, it's free, and it runs on your setup — so here it is answering on my own machine.
What you're watching: my real Hermes agent, live, being asked which job an AI agent should take over first. Its answer — inbox triage and email drafting — and it auto-saves the whole exchange to my own vault.
Thinking it?"Running an agent team must burn a fortune in tokens."
No — that's the biggest myth about it. The everyday 90% runs on a free local model on your own machine, free APIs slot in for more, and for the heavy work it drives the CLIs you already pay for — your Claude subscription already includes the Claude CLI.
It's a layer on top of what you already own, not a new meter. And inside the Boardroom there are full token-efficiency tutorials so you cut usage to the bone.
§ 13XIII ────── closed vs open
You rent one. You own the other.
This difference is the whole video.
Grok Bot · closedrented
xAI runs the computers
xAI runs the models — you can't swap them
Locked to their higher-tier paid plans
Limited early beta, invite-shaped
Your bots live inside their app
Your email, CRM and customer lists run on their machines, under their terms
Hermes Agent · openowned
Open source, from Nous Research
Free, and it runs on your setup
You pick the model behind it
You wire it into whatever tools you want
Shared memory with your other agents
If the rules change or the beta shuts, your setup doesn't even notice
Thinking it?"I already pay for ChatGPT and Claude — do I need another one?"
You don't replace them. The open setup orchestrates them — including the subscriptions you already pay for.
One dashboard, every model you already own, working together.
§ 14XIV ────── so who wins
It depends which of two people you are.
Most people watching this are one of them.
If you want zero setup
Grok Bot is genuinely impressive
Download an app, type a message, a bot starts working. The bots-with-their-own-computers design fixes the problem that has haunted agents for a year.
For a business owner who has never touched an agent, that is a very low wall to climb.
If you want ownership
Hermes wins, and it's not close
Your agents, your data, your rules, no company sitting between you and your team.
If the rules change, the wall goes up, or the beta shuts — your setup doesn't notice.
§ 15XV ────── the third difference
A Grok team only contains Grok bots.
Hermes can be one player on a bigger team — running next to OpenClaw, next to Claude, next to a free Claude Code setup, all sharing one memory.
What you're watching: my own dashboard — Free Claude Code, Hermes, the agent Kanban board, the shared memory vault, and the inbox-to-shipped pipeline. Different agents, one team, one memory.
§ 16XVI ────── the verdict
No — Grok Bot doesn't destroy Hermes Agent.
They're built for different people. But Grok Bot just made the easy path a lot easier, and that will pull a lot of first-timers into agents this month.
Agents stopped being chats. They became coworkers.
You're not copying answers out of a tool anymore — you're reviewing finished work, which flips your job from doer to approver.
The old way~2 hrs a day
Open an AI tool and type a request
Wait for the answer
Copy it out
Do the real work yourself in your real tools
Repeat for every single task
The AI sits next to your work
The new way~10 min of approving
The agent signs into your email itself
It clicks around your CRM itself
It drafts on LinkedIn itself
You wake up to a queue of finished drafts in your voice
Your only job is tapping approve on the good ones
The AI is inside your work
§ 18XVIII ────── shift two
One agent became many.
The design pushes you to build a roster — a chief of staff, sales outbound, an inbox manager, a talent scout, an expense manager, customer support — all working in parallel.
What you're watching: one person's roster on xAI's launch page — Chief, Sales Outbound, Inbox Manager, Account Manager, Talent Scout, Expense Manager — each reporting back in a line while the sales bot works on the right.
Inbox Manager0 unread5 drafts parked for tomorrow
Customer Support0 tickets resolved2 escalated to you
Invoice Collector0 invoices pulledstraight from vendor portals
Talent Scout0 intros draftedin your voice, held for your ok
These are the status lines from xAI's own demo. Each bot has one job, gets good at it, and reports back in a sentence.
Thinking it?"Will this actually work for my type of business?"
Every business on earth has an inbox, follow-ups that never get sent, and a report someone builds by hand every week.
Those are the three jobs a roster takes over first — agency, ecom, coaching, local services, it doesn't matter what you sell.
You just became a manager, whether you have employees or not.
§ 19XIX ────── shift three
They learn by watching — and from each other.
Teaching an agent used to mean writing long instructions and hoping; now you show one a workflow today and hand off the project by Friday.
What you're watching: a bot updating its own memory — it noticed the client only signs annual deals and that Dana is the one who approves, then told the rest of the team.
§ 20XX ────── three beliefs to drop
These three are keeping smart owners on the sidelines.
Each one used to be true. None of them are anymore.
Wrong: "I'm not technical enough for AI agents."
Right: There's no code anywhere in this — you type "handle my inbox" and it asks you what you want. Owners with zero background get the open setup running in a weekend with a roadmap and step-by-step videos.
Wrong: "I can't trust an AI agent with my real business."
Right: The whole flow is draft, queue, approve — 36 messages written, zero sent until a human said so. You hand over the typing, you keep the judgement, and on your own setup you decide exactly which tools it can touch.
Wrong: "I'll wait until this settles down and pick the winner."
Right: The skill isn't the tool. It's breaking your business into jobs an agent can do, writing the instructions, reviewing agent work fast, and building routines — and that transfers to every tool that ships next. The people who start now are building something that compounds.
Don't take my word for it
Members post their wins every day — agency owners, ecom founders, course creators, solo operators across 38 countries. Real businesses, real numbers, in their own words.
You need one of their top paid plans just to get in, it's Mac and iPhone first, and early users are already reporting rough edges.
What you're watching: the plans page and the download — macOS, Apple silicon, early beta. Most people watching this can't use it today even if they want to.
Tweet 4 · from Elon himself
They're fixing basics before opening the doors
The beta widens after they fix issues with the early build.
Normal for a launch — but it means this isn't a finished product yet.
Thinking it?"Then I'll just wait for it to open up."
You can. But the lock-in is the real cost — your bots, their routines, everything they learned about you lives inside their system, and leaving means starting over.
On an open setup that memory is yours: take it anywhere, swap models under it, keep building for years.
§ 22XXII ────── what to do this week
Pick one job. Not a strategy.
The real question isn't which bot wins — it's which jobs in your business should a bot be doing by the end of this month.
1Pick one job. Your inbox, your lead follow-up, or your weekly reporting. One only.
2Give it to one agent. Name it after the job so you always know who does what.
3Show it once. Do the task while it watches, or write the instruction the way you'd brief a person.
4Keep the leash on. Drafts only for the first week — you approve everything before it goes out.
5Turn it into a routine. Same time every day, so it runs without you starting it.
6Then add the next one. That's how you build a Ghost Team — one bot at a time.
What Grok Bot really destroys is the excuse.
The "it's too complicated" excuse.
The "agents aren't real yet" excuse.
When one of the biggest AI labs in the world ships a product where anyone can hire an AI sales team from their phone, watching from the sidelines is over.
The AI Profit Boardroom is where the Agent OS lives — Hermes, OpenClaw, Claude and free Claude Code in one dashboard with shared memory, plus the 30-day roadmap and four live calls a week.
