xAI shipped Grok Bot on August 11, 2026, and the cheapest tier costs $120 per seat per month. That is the price of a team plan, not a personal subscription, and it tells you almost everything you need to know about who the product is built for. Within a week, the public conversation had already split three ways: people calling it the iPhone Moment for AI, builders posting screenshots of bots that ran all night, and a third group pointing out that one of those bots had its memory wiped by an update. The product is real. The price gate is real. The mood is genuinely split, and nobody outside xAI has a clear answer about which side of the split a normal user should land on.
The argument I want to make here is simple. Whether Grok Bot works as advertised is the wrong question to start with. The right question is what its existence implies: xAI bought Cursor, rebranded it into SpaceXAI, and now ships a “Bot” that gets its own persistent cloud computer, signs into your existing tools with its own logins, and keeps working after you close the app. That is not a chatbot. That is a distribution bet, and the bet is that agents-as-coworkers are a shipping product category now, not a research demo. Everything else, including the iPhone comparison, is downstream of that bet.
What a Bot actually is
A Grok Bot is not a chat window with a personality. Each Bot gets its own persistent cloud computer — a real browser, a real filesystem, a real terminal — and signs into the tools it needs using its own logins. You assign it a role (Chief of Staff, Paper, Future Lori, anything you want), and it picks its own underlying model automatically. There is no model picker, no advanced mode, no “use Opus 4 for this” toggle. You tell the Bot what you want done, and it figures out which model to invoke to do it.
That last point matters. If you have used Cursor or Claude Code, you have probably developed opinions about which model to route which task through. Grok Bot removes that lever entirely. The Bot is supposed to pick, and you are supposed to trust the pick. Builders who want control are going to find this frustrating. Builders who want to delegate are going to find it useful. Both reactions are showing up on X.
The always-on runtime is the other half of the architecture. A Bot does not stop when you close the app. It keeps working in its own computer, finishes the job, and reports back. That sounds like marketing, but the early demo videos show a Bot running a multi-step web workflow for forty minutes without supervision. The pattern works. The open question is how often it works on the second attempt, on the tenth, on the hundredth.
The broader shipping footprint matters here too: Grok Bot launched in beta on macOS, iOS, Windows, and Linux, with Android flagged as coming soon (x.ai). Four desktop platforms on day one is not a soft launch. It is a deliberate cross-platform push aimed at the people who already live inside Cursor’s IDE.
What it costs, and who is actually buying
There is no standalone Grok Bot subscription. You cannot pay xAI $30 a month for one Bot and call it a day. The product is bundled into three existing plans:
| Plan | Price | What you get | |—|—|—| | SuperGrok Heavy | $300/mo | Highest model access, Grok Bot included | | Cursor Ultra | $200/mo | The rebranded Cursor top tier with Bots | | Cursor Teams Premium | $120/seat/mo | The cheapest way in |
The $120 floor is the number to anchor on. That is team pricing, and it assumes a small company is splitting the seat cost across multiple users, or that an individual is paying enterprise rates to evaluate a product that has been in beta for four days. Nobody is doing a casual $9.99 trial here. The realistic buyer is either a Cursor household already paying $20/month for the editor and willing to upgrade, or a team lead who needs to put an agent in front of a client this quarter.
@SelukMatt captured the market reaction cleanly: “After Grok 4.6 and Grok Bot, if Anthropic is seeking 2T valuation, doesn’t that mean that buying SpaceXAi stock is like getting a rocket company for free?” (🔗 @SelukMatt on X) That is the second-order pricing question that nobody is asking out loud but everybody is doing the math on. Grok Bot is not a product launch. It is a valuation event with a download link attached.
Why every plugin says Cursor
SpaceX bought Cursor outright in the weeks before the launch, and the entire Cursor team is now inside SpaceXAI. The Grok Bot plugin system carries Cursor branding throughout the documentation, the SDKs, and the in-app marketplace. If you have used Cursor in the last two years, the muscle memory transfers directly. If you have not, the onboarding assumes you already know what a “command palette” is and what “agent mode” means.
@MarioNawfal put the move plainly: “SpaceX just bought Cursor outright. The entire Cursor team is now inside SpaceXAI, tasked with making Grok faster, sharper, and more dangerous.” (🔗 @MarioNawfal on X) That is the distribution story in one sentence. xAI did not have to build a developer ecosystem from scratch. They bought one that was already paying users, already had plugin authors, and already had a community of builders who knew how to ship agents. The Cursor acquisition is not a tech story. It is a go-to-market story, and it is the reason Grok Bot has a marketplace on day one while every other “AI agent” framework from the last two years is still begging people to install their SDK.
The cynical read is that Grok Bot is “Cursor with compute and money.” @lurkylearning posted exactly that line on launch day (🔗 @lurkylearning on X), and it landed. The generous read is that the acquisition gave xAI a developer funnel that no other lab can match without spending billions. Both reads are true at the same time. The question is which one ages better.
The demo mode nobody is talking about
Teach-a-task mode is the feature that did not get any of the launch-day hype but probably matters the most. The workflow is: you perform the task on your own screen, the Bot watches, and it generalizes a reusable skill from what you did. Show it how you file an expense report once, and it files every expense report after that. Show it how you triage your inbox, and it triages the inbox tomorrow, and the day after, and the day after that.
This is the part of the product that pulls in non-developers. If you have ever tried to explain to a colleague what Zapier is, you know that the gap between “I can do this task” and “I can teach a computer to do this task” is the actual barrier to automation for ninety percent of office workers. Teach-a-task collapses that gap. You do not write a script. You do not draw a flowchart. You click through the task once, and the Bot watches.
The early demos are uneven. Some skill extractions generalize cleanly. Others overfit to the exact clicks you made and break when the UI shifts by a pixel. The promise is real; the reliability is not there yet. But this is the part of the product that a non-developer will actually use, and it is the part that nobody on launch day was posting about.
What X is actually saying
The launch-week discourse on X split into three clean waves, and the cleanest way to see them is through the engagement numbers.
The hype wave is led by @BrianRoemmele, whose tweet calling Grok Bot “the ‘iPhone Moment’ for AI” hit 1084 likes and 130 reposts — the most-shared single post of the week (🔗 @BrianRoemmele on X). It is a strong claim, and it is the line that got picked up by tech podcasts and Substacks that would not have covered a “Cursor with compute” launch. The release-week celebration ran wider than just Grok Bot: @MiaAI_lab grouped it alongside Grok 4.6, GLM-5.3, and Qwen 3.8-Max as part of a “massive AI release week” (🔗 @MiaAI_lab on X), and that bundling is part of why the top-of-feed energy was so loud.
The builder wave is where the most useful signal lives. @NickADobos posted, “I’ve been using grok bot for a few days. It’s starting to click. Bro they leapfrogged codex app and Claude code entirely. Super app 2.0. Grok bot’s marketing is completely wrong and misleading.” (🔗 @NickADobos on X) That is the 929-like post. Dobos is a known quantity in the iOS automation space, and “the marketing is wrong” is a specific complaint: the normie-friendly framing hides a serious multi-agent product. He is telling other builders to skip the onboarding videos and go straight to setting up named roles.
@paturious posted the screenshot that every builder has now seen: a roster of running bots — Chief of Staff, Paper, Future Lori, Lori — all live, all configured, all reporting status (🔗 @paturious on X). That is what a serious user looks like on day four. It is also a preview of what $120/seat looks like in practice: not a chat thread, but a small team of named workers with their own computers.
The third wave is operational complaint, and it is the one that nobody pricing the stock is talking about. @Michael_Fenech_ reported that a Grok Bot update wiped his agent roster: “Each agent was renamed ‘New Agent’. I didn’t know who was who and my chief of staff was gone.” (🔗 @Michael_Fenech_ on X) That is not a UI bug. That is a data-loss bug on a product that stores your team’s identity. If you cannot trust that the bots you set up yesterday will exist tomorrow, the $120/seat pitch collapses. The fact that this hit X within four days of launch is not a good sign for the first enterprise customer who hits the same issue on a Friday afternoon.
What Grok Bot does not do
There is no model picker. There is no advanced mode. The Bot chooses which underlying model to use for each subtask, and you do not get to override it. For builders who route tasks manually in Cursor or Claude Code, this feels like a downgrade. For users who want to delegate and not micromanage, it is the entire point.
The trade-off is clear and it is honest. Grok Bot is optimized for delegation, not for control. If you want to hand a Bot a goal and trust it to pick the right model for the job, this product is built for you. If you want to specify “use the cheap model for this, use the strong model for that,” you will need to wait for an advanced mode, or stay on Cursor proper, or stay on Claude Code. None of those options ship the always-on runtime that Grok Bot ships. That is the bet.
The distribution play is the story
The piece I want to leave you with is the Cursor acquisition. xAI did not build Grok Bot in a vacuum. They bought a developer ecosystem that already had paying users, plugin authors, and a community that knew how to ship agents, and they bolted an always-on agent runtime onto the top of it. The iPhone Moment framing is premature. The “cursor with compute” framing is reductive. The real story is that agents-as-coworkers are now a shipping product category, and Grok Bot is the first product in that category with a real distribution channel behind it.
Whether the product holds up under sustained use is the question the next three months will answer. @Michael_Fenech_’s wiped-roster incident is a real warning sign. The lack of a model picker is a real limitation. The $120 floor is a real barrier for anyone who is not already paying for Cursor. But the product exists, it runs, and the bots keep working after you close the app. That is new. Everything else is a matter of price, patience, and which side of the X split you trust.