Grok Bot or Cursor cloud agents: which one does the job

Last checked: 21 Oct 2026. Every vendor fact below comes from the Grok Bot and cloud agent documentation on cursor.com and docs.x.ai, the Grok Bot terms dated 12 Oct 2026, and the vendor's pricing, plan and data governance pages, each linked where it is used.

Grok Bot and Cursor's cloud agents both come from SpaceX Super Intelligence (SpaceXSI), both run on cloud computers and both keep working after you close your laptop. They are built for different work. Grok Bot takes on jobs across business apps and websites and stops for a person's approval before risky actions, while a cloud agent changes code in a repository and hands back a draft pull request for review. Pick by the work you want done and by where the human check needs to sit.

What each product is for

Grok Bot gives you named, persistent AI teammates: in the vendor's words, "AI teammates with names, jobs, and context that compounds over time", which take on multi-step work across apps and websites and come back when something needs approval. Most example roles in the Grok Bot use cases are business operations, such as sales outbound, expense management and account health. Two sit closer to engineering: a product performance investigation and a role that reproduces reported bugs in staging. We explained what Grok Bot is, and how it differs from Grok and the Grok API, in our Grok API pricing post.

Cursor's cloud agents are coding agents that "run in isolated VMs in the cloud with full development environments instead of on your local machine". They were formerly called background agents, a name you will still see in searches. A cloud agent clones your repository from GitHub, GitLab, Azure DevOps Services or Bitbucket Cloud, works on a separate branch, and opens a draft pull request for a person to review before anything merges. Our post on Cursor cloud agents and automations covers setup and triggers.

The overlap is narrow: a Bot can reproduce a bug or hand a coding task to a cloud agent, and a cloud agent can drive a desktop and browser to test what it builds (see its capabilities page).

The two side by side

We found no vendor page that compares the two products. The table is SpotDev's comparison, built from the vendor documentation linked throughout this post.

Where each one runs, and what that means for UK data

Each Grok Bot user gets one persistent cloud computer, with a browser, filesystem and terminal, and every Bot that user runs shares its files, browser sessions and app logins, according to the Grok Bot overview. The Grok Bot terms say a user's Bots "must not be treated as separate security boundaries". Anything one of your Bots can reach, your other Bots can reach too. The Grok Bot security page says the computers are vendor-hosted only, with no on-premises or bring-your-own-image option, and run in the United States today. Cursor's US-only data residency scheme does not apply to Grok Bot by default, and the page tells customers who need a written residency commitment to contact their account team.

Cloud agents get one virtual machine each. The cloud agent security page says each agent "runs in its own VM boundary, not a shared process sandbox", in a cloud account separate from the rest of Cursor's production infrastructure. No page we read states the region. Cursor's privacy and data governance page offers Enterprise customers US-only residency, enabled per team, and EU and UK coverage on request for inference only. Storage and processing are not covered for the UK, so this is not UK data residency. If the execution must happen on your own machines, self-hosted machines move file edits and terminal commands onto hardware you run, while Cursor keeps the agent loop, inference and planning.

Both products need the current Privacy Mode, and accounts on Privacy Mode (Legacy) must switch first. With it on, both sets of documentation say your data or code is not used for training. Cloud agent conversation history is kept indefinitely by default, according to the cloud agent network and data page. For the wider UK picture across vendors, see where your data actually goes.

What admins can control on each plan

This is the sharpest difference for a business on a self-serve plan. The Grok Bot teams page says that on the Teams plan Grok Bot is enabled by default, every member has access without admin action, and there is no switch to turn it off. Teams admins can set team rules, the cloud agent delegation switch, template sharing and the local execution ceiling. The organisation-wide enable switch, network controls, enforced Auto Review, audit logs, action recording, the MCP allowlist and SCIM are Enterprise only. Without a network policy, Grok Bot defaults to allow-all.

Cloud agents give more control below Enterprise. Users, team admins and environment owners can set network access to allow-all or allowlist-only, according to the cloud agent settings page, and locking that setting is Enterprise only. Admins also decide whether team follow-ups are allowed. The same page warns these let one user steer an agent that runs with another user's secrets and credentials.

The human check works differently. Grok Bot asks for approval with Allow once, Always allow or Deny, and Enterprise teams can enforce Auto Review, an independent review model that checks risky Bot actions before they run, though not every side effect. Approvals for work that started without you, such as a routine, expire after about 10 minutes at the time of checking, according to the approvals documentation. Cloud agents run terminal commands without stopping for approval, so the controls sit around the run: network rules, runtime secrets, excluded files, draft pull requests and signed commits. Our post on approval design for AI workflows sets out who should sign off what.

How each one is billed

On The X Subscription, the new bundled plan, Grok Bot and Cursor share one weekly usage pool. The Grok Bot plans page says Grok Bot "draws from the same weekly pool as Cursor and the Grok apps, not a separate Grok Bot meter", and the models and pricing page puts cloud agents in the same pool. A heavy day of Bot work can use allowance your engineers were relying on. When the pool runs out, extra Grok Bot usage comes from prepaid credit. On older Cursor plans, Grok Bot stops at the end of its weekly allowance unless on-demand usage is on, and legacy Teams plans have on-demand on by default.

Cloud agents are charged at the API rate for the model you select, and a larger context window can increase cost. Cursor asks you to set a spend limit at first use. Our reading of the two pages together is that cloud agent usage is metered at the model's rate against the shared pool and then credit, though the vendor does not spell out the order.

Grok Bot has no spend cap of its own today, only account-level controls, and a limit does not stop a run already in progress. Broad routine triggers "consume usage", according to the Grok Bot work guide, and Bots messaging each other adds turns. Prices on the pricing page exclude VAT. Our post on Cursor spend controls and model choice covers the controls in detail.

Both products start at the same tier of the new plans: Premium and above include cloud agents and Grok Bot, and Plus does not. On legacy Cursor plans, Grok Bot comes with every paid individual plan and with Cursor Teams. Enterprise access to Grok Bot goes through your account team. Plan names are changing as customers move to the new subscription, so check the plan page before you buy.

Risk, and who carries it

The Grok Bot terms, as at 12 Oct 2026, make the customer solely responsible for the charges, commitments, communications and other consequences of actions Bots take on its behalf. They describe approval and automated review as aids that may not catch every harmful action and do not reverse completed ones, and they expect human review before Bots send external communications, make payments, delete data, change permissions or touch production systems. The provider's liability for those actions is excluded except where its gross negligence or wilful misconduct directly caused them. We did not read a separate cloud agent supplement, so we draw no contract comparison.

Neither product claims to solve prompt injection. The Grok Bot security page says its controls "reduce, but don't eliminate" risk from malicious content, and the cloud agent security page says planted instructions could try to make an agent send code to an external host. For a UK contract review, note that the terms of service and bundled subscription terms we read choose Texas law (Enterprise order forms, which we did not read, can differ), and that more than one company can be on the paper depending on the plan and how it was bought. Cursor's data processing addendum covers UK GDPR transfers through the UK International Data Transfer Addendum. The ICO's call for evidence on agentic AI, opened on 8 Oct 2026, runs to 20 Nov 2026.

Running both: when a Bot hands work to a cloud agent

The Grok Bot teams page says a Bot can delegate coding tasks to Cursor cloud agents, which run on separate computers under your existing cloud agent controls. The switch applies to the whole team and is on by default, and the vendor's recommended baseline is to disable spawning if you don't need delegation. No page we read says how delegated runs are billed, so treat delegation as a new feature and decide on purpose whether your team needs it.

A decision guide in four questions

1. What is the unit of work? A job across business apps and websites, such as account research or expense follow-up, is the work Grok Bot is documented for. A code change that should end as a pull request is what cloud agents do.
2. Who approves it, and when? Grok Bot puts a person in front of each risky action. Cloud agents run freely inside their machine and put the review at the pull request.
3. Where must the computer live? If United States hosting with no residency commitment by default is a problem, Grok Bot offers no other hosting option today. Cloud agents offer US-only residency on Enterprise and self-hosted machines.
4. What can your plan control today? On a self-serve Teams plan, admins cannot switch Grok Bot off, set a network allowlist for it or read a Grok Bot audit log. Cloud agents allow allowlist-only network access below Enterprise.

A business can run both, because the work is different. Before switching either on, move off Privacy Mode (Legacy), check whether Teams members already have Grok Bot, decide on delegation, name an approver for each Bot role, read what the Grok Bot terms make you responsible for, and set a cloud agent spend limit. Outcomes in vendor material are vendor-reported: SpaceXSI's launch post says its own teams used Bots for sales outbound, marketing campaigns, office operations and bug fixes, and we found no independent evidence on either product's results.

Common questions

Is Grok Bot the same thing as a Cursor cloud agent?

No. Grok Bot gives you named, persistent Bots that take on multi-step jobs across business apps and websites and ask for approval before risky actions. A Cursor cloud agent is a coding agent in its own virtual machine that works on a branch of your repository and hands back a draft pull request for review. Both come from SpaceXSI and both run on cloud computers.

Where does each one run, and what does that mean for UK data?

Grok Bot computers run in the United States today, and Cursor's US-only data residency scheme does not apply to Grok Bot by default. Cloud agents run in isolated virtual machines in the vendor's cloud. Enterprise customers can enrol in US-only residency, and EU and UK coverage is available on request for inference only, which is not UK data residency.

How is each one billed, and can we cap spend?

On the new plans, Grok Bot and cloud agents draw from one shared weekly pool, then from prepaid credit. Cloud agents are charged at the selected model's API rate, and Cursor asks for a spend limit at first use. Grok Bot has no spend cap of its own today, and a limit does not stop a run already in progress.

Can a Grok Bot hand coding work to a cloud agent?

Yes. The vendor documents delegation from Bots to Cursor cloud agents, which run under your existing cloud agent controls. The setting is on by default and admins can switch it off. As at 21 Oct 2026, no vendor page we read explains how delegated runs are billed.

What can our admins control on a Teams plan compared with Enterprise?

On the Teams plan, Grok Bot is on for every member with no off switch, and network controls, enforced Auto Review and audit logs are Enterprise only. For cloud agents, users and team admins can set allowlist-only network access, and locking that setting across the organisation is Enterprise only.

If you are weighing up either product, we can review the work you want to hand over and the controls your plan gives you. Our Grok Bot rollout sets Grok Bot up with those controls in place, and the wider picture is in our guide to SpaceX Super Intelligence for UK business. To talk it through, Request a Quote through our SpaceXSI implementation page.

John Kelleher holds the SpaceX Super Intelligence Partner Network Accreditation: Forward Deployed Engineer.