[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"article-grok-bot-turns-browser-chores-into-cloud-work-en":3,"article-related-grok-bot-turns-browser-chores-into-cloud-work-en":29,"series-ai-agent-add7325b-6922-4d4f-941d-f7c716a48217":75},{"id":4,"slug":5,"title":6,"content":7,"summary":8,"source":9,"source_url":10,"author":11,"image_url":12,"cover_image":12,"category":13,"language":14,"translated_content":11,"related_article_id":15,"keywords":16,"key_takeaways":22,"views":26,"created_at":27,"published_at":28,"topic_cluster_id":11},"add7325b-6922-4d4f-941d-f7c716a48217","grok-bot-turns-browser-chores-into-cloud-work-en","Grok Bot turns browser chores into cloud work","\u003Cp data-speakable=\"summary\">Grok Bot turns browser chores into always-on cloud work.\u003C\u002Fp>\u003Cp>I've been living inside \u003Ca href=\"\u002Ftag\u002Fagent\">agent\u003C\u002Fa> workflows for a while now, and honestly, most of them feel like they were built by someone who never had to use them all day. They look smart in a demo. Then you try to use them for real work and they fall apart the second your laptop sleeps, your Wi-Fi hiccups, or a login flow asks for one more annoying confirmation. That part drives me nuts. I don't want an assistant that only works while I'm babysitting it. I want something that can keep moving when I close the lid, switch tabs, or get pulled into something else.\u003C\u002Fp>\u003Cp>The thing that got me paying attention here was a Chinese write-up on \u003Ca href=\"https:\u002F\u002Fzhuanlan.zhihu.com\u002Fp\u002F2070798466993992023\">Zhihu\u003C\u002Fa> about Grok Bot, framed as \u003Ca href=\"\u002Ftag\u002Fxai\">xAI\u003C\u002Fa>'s early beta push. The post claims Grok Bot is an always-on AI teammate with its own cloud computer, and it ties that to \u003Ca href=\"\u002Ftag\u002Fcursor\">Cursor\u003C\u002Fa>'s ecosystem, cloud execution, and multi-bot workflows. I can't verify every claim in the post from the source alone, so I'm treating it as a breakdown of the author's read, not as a product spec sheet. But the core idea is clear enough to be useful: stop thinking of an agent as a chat window and start thinking of it as a worker with a machine.\u003C\u002Fp>\u003Ch2>What changes when the agent gets its own machine\u003C\u002Fh2>\u003Cblockquote>“Each Bot has an independent cloud Linux virtual machine. It has its own browser, file system, and terminal, and can act like a human: open pages, click buttons, fill forms, and take screenshots.”\u003C\u002Fblockquote>\u003Cp>What this actually means is simple: the agent is no longer trapped inside your browser session or your foreground tab. It has a place to live. That sounds boring, but it's the difference between a toy and something you can trust with repetitive work. If the bot owns the machine, then the machine can keep running after you leave. If the bot only borrows your laptop, then every interruption becomes a failure mode.\u003C\u002Fp>\n\u003Cfigure class=\"my-6\">\u003Cimg src=\"https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1786757604668-vcyt.png\" alt=\"Grok Bot turns browser chores into cloud work\" class=\"rounded-xl w-full\" loading=\"lazy\" \u002F>\u003C\u002Ffigure>\n\u003Cp>I ran into this exact wall with early browser agents. They were fine for one-off tasks like checking a page or drafting a reply. But the moment I wanted them to monitor a queue, continue after a timeout, or handle a form that took longer than expected, they got brittle. A cloud VM changes the whole shape of the problem. Now the agent can keep state, keep files, keep sessions, and keep going without asking me whether my battery is at 12%.\u003C\u002Fp>\u003Cp>How to apply it: if you're building or evaluating an agent, stop asking first about prompt quality and start asking about runtime ownership. Where does the agent live? What survives a browser crash? What survives a laptop close? If the answer is “nothing,” you're not looking at an agent platform yet. You're looking at a demo with a nicer wrapper.\u003C\u002Fp>\u003Cp>There's also a practical upside that gets overlooked: a dedicated machine gives you a clean boundary for logs, artifacts, and debugging. If the bot fills out a form wrong, you can inspect the VM state instead of trying to reconstruct what happened from a chat transcript. That's the kind of unglamorous detail that decides whether a tool becomes part of a workflow or gets abandoned after two tries.\u003C\u002Fp>\u003Ch2>Why “it keeps working when I close the laptop” matters\u003C\u002Fh2>\u003Cp>The post makes a big deal out of the bot being cloud-native from day one. I get why. That one detail kills a bunch of the usual annoyance. No need to keep your laptop awake. No need to babysit a local process. No need to panic because your machine went to sleep during a long task. For anyone who's ever tried to automate a day of browser work, that is not a small thing. It's the whole thing.\u003C\u002Fp>\u003Cp>What this actually means is that the agent can run like a background service instead of a screen-sharing session. That's closer to how I expect infrastructure to behave. I don't want to ask, “Is the bot still open?” I want to ask, “Is the job done?”\u003C\u002Fp>\u003Cp>I like this framing because it also changes the failure model. If a local agent dies, the task dies with it. If a cloud agent dies, at least you can restart from a known state. You can queue work, resume work, and inspect work. That matters more than clever reasoning when the work itself is repetitive and stateful.\u003C\u002Fp>\u003Cp>How to apply it: if you're designing an internal agent, separate the control plane from the work plane. Let the user decide what to do, but let the bot execute in a durable environment. Store credentials carefully, keep session state isolated, and make sure the user can hand off a task without staying glued to the machine. This is where a lot of “AI coworker” products quietly fail: they confuse a front-end with a worker.\u003C\u002Fp>\u003Cul>\u003Cli>Use a remote runtime for long tasks.\u003C\u002Fli>\u003Cli>Persist browser state and artifacts outside the chat UI.\u003C\u002Fli>\u003Cli>Give the user a handoff point for approvals and logins.\u003C\u002Fli>\u003C\u002Ful>\u003Ch2>Multi-bot setup is where the idea gets less silly\u003C\u002Fh2>\u003Cp>The post argues that one bot is an assistant, but multiple bots become a team. That's the part I actually buy. A single agent trying to do everything ends up context-switching itself into mush. Split the work, and the whole thing becomes more believable. One bot watches inboxes, one pulls reports, one drafts content, one checks competitors. That is much closer to how real teams work, and frankly, much closer to how I want automation to behave.\u003C\u002Fp>\n\u003Cfigure class=\"my-6\">\u003Cimg src=\"https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1786757605543-5ny4.png\" alt=\"Grok Bot turns browser chores into cloud work\" class=\"rounded-xl w-full\" loading=\"lazy\" \u002F>\u003C\u002Ffigure>\n\u003Cp>What this actually means is that coordination matters more than raw intelligence. If each bot has a clear lane, then the system can scale by parallelism instead of by asking one model to remember everything. The Zhihu post says bots can communicate and even call in another bot when they get stuck. That's not just a cute feature. It's the difference between a single-threaded helper and a small operations crew.\u003C\u002Fp>\u003Cp>I ran into this pattern while wiring internal automations for support and sales ops. The moment I tried to make one worker do inbox triage, CRM updates, and report generation, the workflow became fragile. Breaking it into roles fixed most of the mess. One worker classified, another enriched, another wrote back. The quality improved because each step had a narrower job.\u003C\u002Fp>\u003Cp>How to apply it: define bot roles before you define prompts. Give each bot one job, one input shape, and one output contract. If a task needs human approval, route that approval to a separate step instead of letting the main bot improvise. This is how you keep a multi-agent system from turning into a pile of self-referential notes.\u003C\u002Fp>\u003Cul>\u003Cli>Inbox bot: triage, tag, draft replies.\u003C\u002Fli>\u003Cli>Ops bot: update CRM, generate reports, flag anomalies.\u003C\u002Fli>\u003Cli>Research bot: collect sources, summarize changes, prepare briefs.\u003C\u002Fli>\u003Cli>Approval bot: pause and hand tasks back to a human when needed.\u003C\u002Fli>\u003C\u002Ful>\u003Ch2>The real trick is showing, not prompting\u003C\u002Fh2>\u003Cp>One detail in the post that I found genuinely useful is the “follow-along learning” idea. You do the task once the normal way, and the bot watches. It records the steps and reuses them later. That is much less annoying than forcing everyone to become prompt engineers or workflow designers just to automate a dumb recurring task.\u003C\u002Fp>\u003Cp>What this actually means is that the system is trying to learn by observation, not by perfect instructions. That's a better fit for most office work because most office work is messy. People don't operate with pristine APIs. They click around, copy values, switch tabs, and make judgment calls. A bot that can learn from a real run is closer to the actual job than a bot that only understands a carefully written prompt.\u003C\u002Fp>\u003Cp>I like this because it lowers the barrier for the first automation. The first run is manual. The second run is assisted. The third run is repeatable. That's a much saner path than asking someone to build a full workflow upfront. The article compares this to turning repetitive work into one-click automation by “showing it once.” That's the right mental model.\u003C\u002Fp>\u003Cp>How to apply it: record the first successful manual run of any repetitive task. Keep the steps short and observable. Save screenshots, page names, button labels, and decision points. Then convert those steps into a reusable runbook. If your system supports it, let the bot ask for clarification only at the branch points, not at every step. That's how you keep the automation from becoming a nag.\u003C\u002Fp>\u003Cp>There's a second benefit here: it makes hidden process knowledge visible. A lot of teams rely on one person who “just knows how to do it.” If the bot can watch that person once and capture the flow, you've turned tribal knowledge into something repeatable. That's not glamorous, but it saves real time.\u003C\u002Fp>\u003Ch2>Why the pricing and ecosystem matter more than the hype\u003C\u002Fh2>\u003Cp>The post also points out the cost. It mentions SuperGrok Heavy at $200-300 per month, Cursor Ultra at $200 per month, and Cursor Teams Premium at $120 per user per month. I can't validate those numbers from the Zhihu post alone, but the broader point is obvious: these tools are not priced like casual toys. They're priced like infrastructure for people who expect to get work out of them.\u003C\u002Fp>\u003Cp>What this actually means is that the product isn't being sold as a novelty. It's being sold as a replacement for chunks of labor. That is why the ecosystem matters. The post says Grok Bot inherits connectors from Cursor, including tools like \u003Ca href=\"https:\u002F\u002Fwww.notion.so\u002F\">Notion\u003C\u002Fa>, \u003Ca href=\"https:\u002F\u002Fslack.com\u002F\">Slack\u003C\u002Fa>, and \u003Ca href=\"https:\u002F\u002Fgithub.com\u002F\">GitHub\u003C\u002Fa>. If that's true in practice, then the onboarding friction drops hard. Nobody wants to rewire every integration from scratch just because a new bot showed up with a shinier name.\u003C\u002Fp>\u003Cp>I care about this because integrations are where agent products usually die. The demo has fake data and a clean flow. Real work has credentials, permissions, edge cases, and half-broken SaaS settings pages. If the bot can't plug into the tools people already use, then it becomes yet another place to copy and paste from. That's not automation. That's admin with extra steps.\u003C\u002Fp>\u003Cp>How to apply it: when you evaluate an agent platform, check three things before you look at model quality. First, what connectors already exist. Second, whether session state is inherited or rebuilt. Third, whether the bot can operate inside the tools your team already pays for. If you have to redesign your stack just to make the agent useful, the product is asking too much.\u003C\u002Fp>\u003Ch2>This is less about AI magic and more about workflow ownership\u003C\u002Fh2>\u003Cp>The strongest claim in the post is also the least flashy: the bot is not just a chat assistant, it's a coworker that logs into your accounts and completes end-to-end tasks. That's the real shift. Not “it writes text.” Not “it answers questions.” It owns a workflow from start to finish, with human handoff only where needed.\u003C\u002Fp>\u003Cp>What this actually means is that the unit of value changes. Instead of paying for tokens or replies, you're paying for completed work. That is much closer to how businesses think. A completed support reply, a cleaned-up CRM entry, a booked order, a research brief, a canceled subscription. Those are outcomes. If an agent can own outcomes, then it starts competing with actual labor, not just with another chatbot.\u003C\u002Fp>\u003Cp>I think that's why the post keeps comparing Grok Bot with \u003Ca href=\"\u002Ftag\u002Fclaude\">Claude\u003C\u002Fa>'s coworker-style approach and with Cursor's cloud agent direction. Once the runtime is remote, the model is just one piece. The bigger question is whether the system can hold state, ask for approval, and keep moving across apps without collapsing into a pile of prompts.\u003C\u002Fp>\u003Cp>How to apply it: define your agent by the work it completes, not the text it generates. Write down the exact workflow boundary: start event, tools used, approval points, completion criteria, and failure handling. If you can't describe that in one page, you probably don't have an agent workflow yet. You have a hope.\u003C\u002Fp>\u003Ch2>The template you can copy\u003C\u002Fh2>\u003Cpre>\u003Ccode># Cloud Agent Workflow Template\n\n## Bot name\n[Give the bot one job only]\n\n## Job\n[Describe the exact outcome the bot must complete]\n\n## Runtime\n- Runs in a cloud VM\n- Has its own browser, file system, and terminal\n- Stays active after the user closes their laptop\n\n## Inputs\n- Source app(s): [Gmail, Slack, Salesforce, Notion, GitHub, etc.]\n- Trigger: [new email, daily schedule, manual request, webhook]\n- Required context: [customer name, project ID, date range, account]\n\n## Allowed actions\n- Open pages\n- Click buttons\n- Fill forms\n- Download files\n- Upload files\n- Generate screenshots\n- Write drafts\n\n## Approval points\nPause and hand control back to the user for:\n- Login\n- SSO verification\n- Payment confirmation\n- Sending external emails\n- Deleting data\n- Any action with legal or financial impact\n\n## Workflow steps\n1. Read the trigger and load the task context.\n2. Open the required app in the cloud browser.\n3. Perform the task step by step.\n4. Save artifacts and logs.\n5. Stop at approval points.\n6. Resume after user confirmation.\n7. Mark the task complete.\n\n## Learning mode\n- Show the bot one successful manual run first\n- Record page names, button labels, and decision points\n- Reuse the captured steps on the next run\n- Update the workflow when the UI changes\n\n## Multi-bot roles\n- Research bot: gathers information\n- Ops bot: updates systems and reports\n- Inbox bot: triages and drafts replies\n- Approval bot: waits for human confirmation\n\n## Failure handling\n- Retry once on transient page errors\n- Save the current state before retrying\n- Escalate to a human if login, payment, or permissions fail\n- Keep a log of the last successful step\n\n## Success criteria\n[Write the observable result]\n- Example: CRM updated\n- Example: report exported\n- Example: email draft ready for review\n- Example: order placed and confirmation saved\n\n## Notes\n- Keep each bot narrow\n- Do not let one bot own every workflow\n- Treat browser state as disposable, but task state as persistent\n\u003C\u002Fcode>\u003C\u002Fpre>\u003Cp>I would use this template as a starting point for any browser-heavy automation work. It keeps the runtime, approval, and failure model explicit, which is where most agent projects get sloppy. If you want the bot to act like a coworker, you need to define the job like a coworker would understand it.\u003C\u002Fp>\u003Cp>One last thing: don't let the template seduce you into overbuilding. Start with one boring task. One inbox flow. One report. One order. If the bot can finish that reliably, then you can add a second role. That's how you avoid turning a useful agent into a science fair project.\u003C\u002Fp>\u003Cp>Source attribution: the original breakdown I referenced is the Zhihu post at \u003Ca href=\"https:\u002F\u002Fzhuanlan.zhihu.com\u002Fp\u002F2070798466993992023\">https:\u002F\u002Fzhuanlan.zhihu.com\u002Fp\u002F2070798466993992023\u003C\u002Fa>. My write-up is a developer-facing decomposition of that source, plus my own workflow advice and template.\u003C\u002Fp>","I break down Grok Bot’s cloud-agent workflow and give you a copy-ready template for multi-bot task automation.","zhuanlan.zhihu.com","https:\u002F\u002Fzhuanlan.zhihu.com\u002Fp\u002F2070798466993992023",null,"https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1786757604668-vcyt.png","ai-agent","en","71bf52af-c04d-4371-854c-75dea55aa88a",[17,18,19,20,21],"Grok Bot","AI agent","cloud VM","browser automation","multi-agent",[23,24,25],"Cloud-owned runtime beats laptop-bound agents for real work.","Multi-bot workflows are more useful when each bot has one job.","A reusable template matters more than the hype around the model.",2,"2026-08-15T01:33:02.708294+00:00","2026-08-15T01:33:02.697+00:00",{"tags":30,"relatedLang":34,"relatedPosts":38},[31,33],{"name":20,"slug":32},"browser-automation",{"name":18,"slug":13},{"id":15,"slug":35,"title":36,"language":37},"grok-bot-cloud-agent-copy-template-zh","Grok Bot 讓工作搬上雲端","zh",[39,45,51,57,63,69],{"id":40,"slug":41,"title":42,"cover_image":43,"image_url":43,"created_at":44,"category":13},"1ae1e1b4-ef77-411e-94d5-0f5fd70897d6","chatgpt-mac-computer-history-memory-en","ChatGPT now remembers your Mac work history","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1786777367544-sddp.png","2026-08-15T07:02:25.759069+00:00",{"id":46,"slug":47,"title":48,"cover_image":49,"image_url":49,"created_at":50,"category":13},"10a73d21-d035-474d-9fd9-1a1dd5caa863","agent-plugins-100-turns-skills-into-folders-en","Agent Plugins 1.0.0 turns skills into folders","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1786610024064-pe2a.png","2026-08-13T08:33:10.730688+00:00",{"id":52,"slug":53,"title":54,"cover_image":55,"image_url":55,"created_at":56,"category":13},"b77650c4-7664-4af1-8ff1-2e1b792cdd58","faitheyes-tool-faithful-vision-agents-en","FaithEyes lets you train tool-faithful vision agents","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1786258993834-8aqq.png","2026-08-09T07:02:49.995282+00:00",{"id":58,"slug":59,"title":60,"cover_image":61,"image_url":61,"created_at":62,"category":13},"3491c99c-621c-40d3-bbd8-b4a73702fb42","connect-qwen-3-8-max-cli-agents-en","Connect Qwen 3.8 Max to CLI Agents","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1786255373689-rrwa.png","2026-08-09T06:02:30.990794+00:00",{"id":64,"slug":65,"title":66,"cover_image":67,"image_url":67,"created_at":68,"category":13},"445ecce7-ea09-49f7-a7f4-88391eb1bbf3","fine-tune-small-llm-legal-labeling-en","Fine-tune a small LLM for legal labeling","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1786150968314-7umc.png","2026-08-08T01:02:27.269166+00:00",{"id":70,"slug":71,"title":72,"cover_image":73,"image_url":73,"created_at":74,"category":13},"306b4d13-3911-4fcb-9f53-861fa5e9b430","sala-boosts-long-context-edge-ai-en","SALA Boosts Long Context on Edge AI","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1785765793459-52io.png","2026-08-03T14:02:45.076761+00:00",[76,81,86,91,96,101,106,111,116,121],{"id":77,"slug":78,"title":79,"created_at":80},"03db8de8-8dc2-4ac1-9cf7-898782efbb1f","anthropic-claude-ai-agent-task-automation-en","Anthropic's Claude AI Agent: A New Era of Task Automation","2026-03-25T16:25:06.513026+00:00",{"id":82,"slug":83,"title":84,"created_at":85},"045d1abc-190d-4594-8c95-91e2a26f0c5a","googles-2026-ai-agent-report-decoded-en","Google’s 2026 AI Agent Report, Decoded","2026-03-26T11:15:23.046616+00:00",{"id":87,"slug":88,"title":89,"created_at":90},"e64aba21-254b-4f93-aa21-837484bb52ec","kimi-k25-review-stronger-still-not-legend-en","Kimi K2.5 review: stronger, still not a legend","2026-03-27T07:15:55.385951+00:00",{"id":92,"slug":93,"title":94,"created_at":95},"30dfb781-a1b2-4add-aebe-b3df40247c37","claude-code-controls-mac-desktop-en","Claude Code now controls your Mac desktop","2026-03-28T03:01:59.384091+00:00",{"id":97,"slug":98,"title":99,"created_at":100},"254405b6-7833-4800-8e13-f5196deefbe6","cloudflare-100x-faster-ai-agent-sandbox-en","Cloudflare’s 100x Faster AI Agent Sandbox","2026-03-28T03:09:44.356437+00:00",{"id":102,"slug":103,"title":104,"created_at":105},"04f29b7f-9b91-4306-89a7-97d725e6e1ba","openai-backs-isara-agent-swarm-bet-en","OpenAI backs Isara’s agent-swarm bet","2026-03-28T03:15:27.849766+00:00",{"id":107,"slug":108,"title":109,"created_at":110},"3b0bf479-e4ae-4703-9666-721a7e0cdb91","openai-plan-automated-ai-researcher-en","OpenAI’s plan for an automated AI researcher","2026-03-28T03:17:42.312819+00:00",{"id":112,"slug":113,"title":114,"created_at":115},"fe91bce0-b85d-4efa-a207-24ae9939c29f","harness-engineering-ai-agent-reliability-2026","Harness Engineering: From Bridle to Operating System, The Missing Link in AI Agent Reliability","2026-03-31T06:36:55.648751+00:00",{"id":117,"slug":118,"title":119,"created_at":120},"7a09007d-820f-43b3-8607-8ad1bfcb94c8","mcp-explained-from-prompts-to-production-en","MCP Explained: From Prompts to Production","2026-04-01T09:24:40.089177+00:00",{"id":122,"slug":123,"title":124,"created_at":125},"116d5ee9-a4f1-4b5a-aac5-5d035dd22bbe","amazon-bedrock-agents-multi-agent-workflows-en","Amazon Bedrock Agents Gets Multi-Agent Workflows","2026-04-01T09:30:30.197685+00:00"]