# Introducing Grok Bot: Your AI Agent for iPhone and Mac > Grok Bot, the new AI agent from SpaceXAI and Cursor, offers always-on digital workers for smooth task management across devices. > By Dave · 2026-08-12 > Source: https://otf-kit.dev/blog/grok-bot-iphone-mac ## The always-on worker just moved from chat into your inbox A bot that signs into your tools, finishes a multi-step job while you grab coffee, and asks you only when the decision actually needs you — that's what SpaceXAI and Cursor shipped this week with Grok Bot, an AI agent for iPhone and Mac (plus Windows and Linux, with Android listed as coming soon). It's in beta for a limited group of paying subscribers, and access requires a SuperGrok Heavy, Cursor Ultra, or Cursor Teams Premium seat. The point worth sitting with is the shape of the thing: each bot runs on its own cloud computer, so it doesn't tap out after one prompt, and it learns your preferences by watching you work. That's a real step forward. The rest of this post is the how — how it actually works, how to get on the beta, and where the durable layer sits when the agent underneath keeps changing. ## What Grok Bot actually is Grok Bot is an always-on AI agent from SpaceXAI and Cursor — the two companies are moving toward becoming a single company, per the launch coverage. You talk to a bot like a coworker, hand it a task, and come back later when the work is done or when the bot needs your call on something it can't decide alone. The bot signs into the apps, inboxes, websites, and services you give it access to, and it keeps going between steps instead of waiting for you to press enter. SpaceXAI's own teams have been using bots internally for sales outreach, marketing work, office operations, invoice processing, recruiting, bug reporting, and software fixes. That's a useful signal — it's not a demo, it's the vendor dogfooding the same product on its own back office. ## How it works under the hood Three mechanics matter, and they're worth naming: **Isolation.** Every bot gets its own cloud computer, so multiple bots can run in parallel without stepping on each other's logins or in-progress work. A recruiting bot, an invoicing bot, and a bug-triage bot can all run at the same time, each with its own browser sessions and credentials. **Continuity.** The bot completes multi-step jobs without stopping after every action. You message it with a goal, it executes, and pings you when there's a real fork in the road — not after every click. **Coordination.** Bots share context, talk to each other in group chats, hand work off, and bring you in only when something genuinely needs a human decision. That's the line between "an assistant that helps with one task" and "a small team that runs while you're in meetings." ![one cloud computer per bot, persistent logins, multi-step continuity, multi-bot coordinati](https://cdn.otf-kit.dev/blog/grok-bot-iphone-mac/inline-1.png) ## How a bot learns your preferences Two paths, both worth knowing: 1. **Conversation memory.** Bots remember past conversations and learn how you prefer tasks to be handled. The more you use one, the less you have to re-explain your conventions. 2. **Workflow observation.** A bot can watch you complete a workflow once, save the steps, and replay them later without the same instructions. This is the bigger enable — the bot moves from "responds to prompts" to "replaces a routine." The replay loop is what turns the bot from a chat partner into a colleague. You teach it once, and the next time that job shows up in your inbox, it just handles it. ## How to actually use Grok Bot today This part matters more than the launch hype. Here's the concrete path: **1. Get on the right subscription.** Grok Bot is in beta. Access requires one of: SuperGrok Heavy, Cursor Ultra, or Cursor Teams Premium. Enterprise customers can join a waitlist for future access — there's no public self-serve onboarding yet. **2. Install on the surfaces you work on.** The beta runs on macOS, iPhone, Windows, and Linux. An Android version is listed as coming soon. Install the app on whichever device you want to delegate from — most people start on desktop for the keyboard, then add iPhone for "assign while walking" workflows. **3. Spin up your first bot.** Give it a name, point it at the first task you actually want off your plate, and grant the tool access it asks for. Treat the first week like onboarding a new hire — small, well-scoped tasks, clear "done means X" criteria. **4. Teach it a workflow by doing it.** Run the routine once with the bot watching. Confirm the saved steps. Hand the same routine over to the bot the next time it appears. **5. Add a second bot once the first one is boring.** The multi-bot coordination — shared context, group chats, work handoffs — is where the platform gets interesting. Don't rush this step; one well-trained bot is worth five half-trained ones. ![one bot doing one task vs a small team of bots sharing context and handing off work](https://cdn.otf-kit.dev/blog/grok-bot-iphone-mac/inline-2.png) ## How this differs from the assistants already on your phone The AI helpers baked into your existing OS are designed for one-shot, in-the-moment requests: set a timer, open this app, answer this question. They live in the chat layer. They don't sign into your tools with persistent credentials, they don't keep state between actions, and they don't learn a routine by watching you do it once. Grok Bot sits in a different category. It's a long-running worker with its own computer, persistent logins, and the ability to take a goal and break it down into steps it executes on its own clock. The closest analogue isn't a voice assistant — it's more like having a remote teammate who only pings you when something genuinely needs your judgment. That distinction matters for builders, not just end users: the products your bot will operate are the products you ship. If those products behave differently on web, iOS, and Android, your bot has to learn three different shapes — and so does your user. ## The part that doesn't change when the bot does Here's where the durable layer comes in. Agents like Grok Bot are going to keep getting cheaper, faster, and better at multi-step work. The model under the bot will churn — today's Grok Bot won't be next year's Grok Bot, and the bot you onboard this quarter will look quaint in six months. What doesn't churn is the surface the bot touches. The web app, the iOS screen, the Android screen — those are the artifacts the bot has to recognize, click through, and reason about. If those surfaces are three different codebases with three different button labels and three different navigation patterns, every new agent you plug in has to relearn them from scratch, and your users have to relearn them too. This is the layer that holds steady while the tooling above it rotates. One coherent product surface across web, iOS, and Android — same component, same behavior, same mental model for both the human and the agent driving it — means a new bot ships, and your product already speaks its language. You don't re-onboard your product every time a better agent shows up. ![the surface the bot touches is the part that doesn't change when the model does](https://cdn.otf-kit.dev/blog/grok-bot-iphone-mac/inline-3.png) Use Grok Bot. Put it on the work that drains your week. And keep an eye on the part of your stack that has to outlive every model release — because the agents will keep changing, and that part shouldn't have to.