One of Japan's largest directories x find the right AI in as little as a minute

▶︎ For those who want to list their service

  1. AI BEST SEARCH
  2. AI Tool How-Tos & Use Cases
  3. What Grok Bot Is [September 2026]: The AI Teammate That Came Down to $20 a Month — Pricing, What It Does, How to Start

What Grok Bot Is [September 2026]: The AI Teammate That Came Down to $20 a Month — Pricing, What It Does, How to Start

Grok Bot launched on 11 August 2026, and the reason it suddenly got loud in September is not that the product changed — it is that the door came down from $200–300 upper tiers to Cursor Pro at $20 a month. This article walks through what Grok Bot actually does, following the official screens: it signs in to Salesforce, pulls 52 accounts, queues 36 drafts and holds them at "0 sent" waiting for approval, then saves the approved exchange as a routine. It covers the three routes in and eight plans, how Grok Bot differs from Manus, Devin and Genspark AI at the same $20 mark (what each one operates, and whether you hand over your accounts), and the three things to settle before handing anything over — which account it signs in with, how far it may act unattended, and the one shared cloud computer per user — all as published in September 2026.

The Grok Bot desktop app. Chief, Sales Outbound and Inbox Manager sit in the left rail; in the thread on the right the Bot has pulled 52 accounts from Salesforce and queued 36 drafts, holding them at "0 sent" while it waits for approval

Source: Grok Bot official site (all screenshots in this article come from the same source)

Grok Bot launched on 11 August 2026, and suddenly, in September, everyone started saying its name. The product has not changed since August. What changed is the price of the door.

At launch, eligibility was limited to Cursor Ultra ($200/mo), SuperGrok Heavy ($300/mo) and Cursor Premium Teams ($120/seat/mo). As of September 2026 the official pricing page includes it in Cursor Pro, at $20 a month. For an individual, the door is a tenth of the price.

This article covers what Grok Bot actually does, how it differs from the other agents now sitting at the same $20 mark, and what a team should settle before trying it — all from primary sources.


The short version

Grok Bot is not an AI that returns answers; it is an AI that operates applications and finishes the job. From its own cloud computer it signs in to Gmail or Salesforce for real, updates the CRM, writes the drafts, and stops to ask when approval is needed.

It is now included from Cursor Pro at $20 a month. You cannot buy it on its own: it comes bundled with Cursor (Pro, Pro+, Ultra), SuperGrok (standard, Plus, Heavy) or Cursor Teams.

Right now it suits an individual or a small team. The UI is English only, enterprise access is still a waitlist, and every Bot shares one computer. Rolling it out across an organisation can wait until those three are resolved.


What changed: from $200 a month to $20

The official pricing page lists three routes in. The cheapest plan on each already carries the core of Grok Bot.

RouteCheapest planMonthlyUpper tiers
CursorPro$20Pro+ $60 / Ultra $200
GrokSuperGrok$30Plus $100 / Heavy $300
Cursor TeamsStandard$40 per seatPremium $120 per seat

What matters is what the $20 Cursor Pro tier lists. In the official wording: "Grok Bot's own computer", "Signs into your tools", "Routines on a schedule", "Work anywhere: desktop, mobile, and more" and "Weekly Grok Bot usage included" — five items. This is not a stripped-down edition; it is the same substance as the upper tiers.

The official FAQ likewise puts eligibility at "Cursor Pro, Pro+, and Ultra, SuperGrok, SuperGrok Plus, and Heavy, and Cursor Standard and Premium Teams subscribers".

What the upper tiers add is not Grok Bot features but usage allowance, and the other things included in the same subscription. Each subscription carries a weekly Grok Bot allowance, and usage beyond it is billed on token cost. Run heavy work daily and an upper tier can end up cheaper.


What Grok Bot does: it drafts, then stops

The image at the top of this article is the behaviour itself. Read in order:

STEP 1

Sign in and gather the material

  • 52 accounts pulled from Salesforce
  • Context gathered from Hex and LinkedIn too
STEP 2

Produce the work

  • 36 drafts queued
  • 4 skipped on its own as recently contacted
STEP 3

Stop without sending, and wait

It reports at "0 sent". Only when a person says "the top 10 look good, send it" does anything go out — and that exchange is saved as a routine

Step 3 is the point. Grok Bot does not say "all done"; it says "this far — shall I send?" The first worry anyone has about handing work to an agent — that it will send something on its own — is answered in the product's default behaviour.

And the moment a person approves, the whole sequence is saved as a routine. That is the "Created routine / Overnight outbound" line in the image. From the following week it runs without being asked.

You show it the steps rather than writing them down

There are two ways to make a routine. One is the above: ask, let it finish, then put it on a schedule. The other is to let it watch you work.

Two Grok Bot features. On the left, "Grok Bot works where you work": the Bot asks to be signed in to Zendesk from its own computer. On the right, "Show a Bot how it's done": the Bot records and learns while a person builds a report

The official description: "Ask a Bot to follow along as you complete a workflow once. It saves it as a routine and runs it on its own next time." No procedure document, no workflow to build.

This is what makes it useful for work that is hard to put into words — open that admin screen, filter on this, export the CSV, keep only these columns. Faster to show than to explain.

Eight role templates

The official site lists ready-made roles: Sales Outbound, Talent Scout, Paid Media, Expense Manager, Product Performance, Bug Reproduction, Account Health and Chief of Staff.

The line-up says a lot. This is not aimed at development itself, but at the routine work that comes round every week beside it.


Comparing at $20: the other agents at this price

Now that Grok Bot has come down to $20, it has company. Here they are, sorted by what they operate.

ToolWhat $20-ish buysWhat it operatesDo you hand over accounts?
Grok BotBundled with Cursor Pro, $20Your apps and websitesYes — it signs in for real
Manus$20 (4,000 credits)Browser and code inside a sandboxNo
DevinCore from $20 (ACU-based)Codebases and GitHubRepository access only
Genspark AISee the official siteDecks, sheets, AI phone callsNo

Same label, entirely different things handed over. Miss this distinction and the permission conversation stalls the rollout after you have already bought.

Grok BotGrok Bot | Its own computer, signing in to your apps

Each user gets one persistent cloud computer, and the Bots work inside it with a browser and files. Because they sign in to Gmail or Zendesk exactly as a person would, they reach internal tools and dated admin screens that were never given an API. They keep working with your laptop closed.

It fits when the time is going not into the research but into the data entry and follow-ups after it.

See the full Grok Bot page

ManusManus | Keep your accounts, take the output

Give it a goal and it plans, executes and verifies inside a cloud sandbox, returning research, documents, websites and apps as artefacts. Browser control and code execution stay inside that sandbox, so you never hand over your accounts — the key difference from Grok Bot.

Pricing is $20 / $40 / $200 a month for 4,000 / 8,000 / 40,000 monthly credits, 17% off annually. Its strength is that you can start today, without waiting on a permission design.

See the full Manus page

DevinDevin | When the target is a codebase

Cognition's coding agent. From a natural-language instruction it plans, implements, tests and opens the pull request inside its own environment. Pricing is Core (from $20 a month, ACU-based), Team ($500 a month) and Enterprise.

Because the scope stops at code, review works as the safety valve you already have. It does not compete with Grok Bot; it covers a different area.

See the full Devin page

Genspark AIGenspark AI | Research through to finished documents, in one place

A workspace that switches between several AI models and builds the work into finished form — slides, sheets, documents. Note that the pricing page requires a sign-in, so the figures are not publicly visible.

See the full Genspark AI page


Three things to settle before you hand anything over

Grok Bot asks for more than the others do. Decide these three before you start and you will not have to walk anything back.

01Which account it signs in with

Handing over your own admin account is the most dangerous way to begin.

  • Create a dedicated account — a work account with narrowed permissions, and connect only that
  • Avoid shared accounts — you lose the ability to tell who did what, and it falls out of the audit trail

02How far it may act unattended

Sending, paying and deleting are the three to settle in advance.

  • Use Auto Review — the official FAQ states that "Sensitive actions can go through Auto Review before they run"
  • Start at the draft — leaving the send button to a person keeps early mistakes cheap

03Whether one shared computer is acceptable

This constraint is specific to Grok Bot.

  • Isolation is per user — the official FAQ says "Every Grok Bot shares one persistent cloud computer" and "Isolation is per user, not per Grok Bot"
  • It is also the point — sharing files, browser and logins is exactly what lets Bots hand work between themselves. The flip side: access granted to one Bot reaches the others

On data, the official FAQ states training opt-out and encryption in transit and at rest. DLP, certificates, proxies and network controls are described as configurable at boot for enterprise admins — but enterprise access itself is currently a waitlist.


Try it now, or wait?

Fine to try now

  • You already pay for Cursor or SuperGrok — no extra cost to try
  • An individual or small team, deciding its own permissions
  • An English UI is not a blocker

Better to wait

  • You want an org-wide rollout — enterprise is still a waitlist
  • You need a Japanese UI — nothing has been announced
  • Per-Bot isolation is a requirement — one computer per user is the design

Even if you land in the right-hand column, there is no need to stop. Try Manus or Genspark AI, which never ask for your accounts, or look at Cloudflare OS, where the permissions and logs stay yours.

Cloudflare OSCloudflare OS | Self-hosted, handing out permissions from zero

Cloudflare's open-source (Apache 2.0) agent workspace. You run it in your own Cloudflare account, and every agent starts from zero permissions. A per-service Gatekeeper hands over only a limited API, and cost is visible and cappable per person, per team and per app.

For an IT function that wants to hold the permissions and the logs, it is the practical option while Grok Bot's enterprise edition is still coming.

See the full Cloudflare OS page


FAQ

Can I subscribe to Grok Bot alone?
No. It comes with Cursor (Pro, Pro+, Ultra), SuperGrok (standard, Plus, Heavy) or Cursor Teams (Standard, Premium)
Is the $20 tier limited?
Its own computer, signing in to your tools and scheduled routines all appear in the lowest tier's feature list. What the upper tiers mainly add is usage allowance
What happens past the allowance?
Each subscription includes a weekly allowance; beyond it, usage is billed based on token cost
Can it be used in Japanese?
Grok itself handles Japanese, but the app UI is English and no Japanese support has been announced (as of September 2026)
Does it work on a phone?
Desktop (macOS and Windows) and the iOS app, with the same thread continuing on either
Is there an enterprise contract?
Described as opening "in the coming weeks"; for now it is a waitlist. The official FAQ points enterprise customers to their Cursor account teams for early access
Does it work on tools without an API?
Yes. It works from the screen, which is exactly why it reaches internal systems and dated admin screens that were never given an integration
What should we try first?
Pick one task you repeat weekly and let it go as far as the draft. If that works, make it a routine

In summary

  • Grok Bot is the kind of agent that signs in to your apps and finishes the work. What comes back is the work, not an answer
  • As of September 2026 it is included from Cursor Pro at $20 a month. There is no standalone plan; it is bundled into existing ones
  • By default it stops short of sending. An approved exchange is saved as a routine and runs itself next time
  • Three constraints: English UI, enterprise on a waitlist, and one shared computer per user
  • Worth trying for an individual or small team. For an organisation, keep Manus and Cloudflare OS in view alongside it

Related reading

The information here reflects what was published on the official site as of 4 September 2026. Specifications and pricing change; check the official site for the current terms.

Share this article

AI tools featured in this article

Related articles

Autonomous AI Agents Compared: 7 Picks for August 2026 — Grok Bot, Manus, Devin and How to Choose

You tried an AI agent and it ended at "wow, impressive" — because what it returned was an answer, not finished work. That changed in August 2026, when products that leave the work inside your actual tools arrived all at once. This article compares seven of them — Grok Bot, Hey Noah, Manus, Genspark AI, Devin, Kilo Code, and Cloudflare OS — along a single axis: where the output lands. It covers the unit of isolation, how much access you hand over, the three layers of cost, and where to start, all from primary sources.

Autonomous AI Agents Compared: 7 Picks for August 2026 — Grok Bot, Manus, Devin and How to Choose

7 AI Simultaneous Interpretation Tools Compared (August 2026): Is the Built-in Feature in Teams and Zoom Enough?

Search for AI interpretation tools and you get lists of 25 apps mixing consumer travel tools with enterprise meeting software. The real question is whether the built-in feature in the Teams or Zoom plan you already pay for is enough — and if not, what is missing. With Microsoft Teams shipping its Interpreter agent in January 2026 and Google Meet making speech translation generally available in February, that question now has an answer. This article pins down what all three platforms can and cannot do, then compares seven dedicated tools — Sentio, VoicePing, kotoba, DeepL Voice, Wordly, Interprefy, and KUDO — by how they capture audio, what they output, and how they charge.

7 AI Simultaneous Interpretation Tools Compared (August 2026): Is the Built-in Feature in Teams and Zoom Enough?

[2026 Edition] 10 AI Tools to Accelerate Executive Decision-Making | Mid-Term Plans, Investment Analysis, Learning, and Presentation Decks

[Latest 2026] Speed up mid-term business plan drafting, investment decision reports, executive learning, and presentation creation. A deep dive into 9 AI tools that support leadership decision-making — covering features, benefits, selection criteria, and real-world adoption examples.

[2026 Edition] 10 AI Tools to Accelerate Executive Decision-Making | Mid-Term Plans, Investment Analysis, Learning, and Presentation Decks

What Is Cloudflare OS? A Deep Dive into the Open-Source AI Agent Workspace (Architecture, Gatekeepers, Self-Hosting, Pricing)

Cloudflare open-sourced Cloudflare OS, an AI agent workspace, on August 5, 2026. This article covers its zero-permission Gatekeeper security, model selection and cost control via AI Gateway, Gadgets (small personal apps), how to deploy it into your own Cloudflare account, and pricing — based on the official blog and GitHub.

What Is Cloudflare OS? A Deep Dive into the Open-Source AI Agent Workspace (Architecture, Gatekeepers, Self-Hosting, Pricing)

[2026 Edition] The Complete Guide to Claude Code: From Installation to Advanced Techniques

A comprehensive guide covering Claude Code installation, basic operations, practical tips, and pricing plans. Packed with best practices you can apply immediately to real projects — suitable for beginners through intermediate users.

[2026 Edition] The Complete Guide to Claude Code: From Installation to Advanced Techniques

[2026 Edition] 20 AI Tools Transforming Developer Workflows | Coding Assistance, Automated Code Review, No-Code Web Development, and Workflow Automation

A curated look at the best AI tools for developers in 2026. Covers 18 tools across coding assistance, automated code review, no-code development, and workflow automation — all aimed at improving productivity.

[2026 Edition] 20 AI Tools Transforming Developer Workflows | Coding Assistance, Automated Code Review, No-Code Web Development, and Workflow Automation