Tomorrow I will attend the Grok Bot community event, where we will present new ways of working with AI, such as delegating work and how to validate it.
Tomorrow, September 29, 2026, I’m heading to Madrid for the Grok Bot community event. The official excuse is the presentation of new ways of working with AI; the real one is to understand firsthand what it feels like when you stop asking an AI for things and start actually delegating work to it.
What I’m bringing here is the reading I do before going: what Grok Bot is, what changes compared to the chatbot I’m already used to, and why I think this kind of tool is starting to touch the way I organize my day.
What Grok Bot Is
Grok Bot is the autonomous AI agent from xAI (now integrated into SpaceXAI), launched in beta on August 11, 2026. The difference with the Grok many people already know isn’t about branding, it’s about category: Grok responds, Grok Bot executes.
Instead of asking it “explain how to do X”, you give it a high-level task:
“research these 20 LinkedIn accounts and leave me a spreadsheet with the summary”
“review my inbox, classify what’s urgent and reply to whatever can be replied to”
“open this bug in GitHub, find the commit that introduced it and propose the fix”
and it takes care of it.
It breaks the instruction down into subtasks, browses, clicks, runs code, moves files and notifies you when it’s done.
The technical detail that matters most to me: it doesn’t only use APIs. Every bot lives in a Linux virtual machine in the cloud, with its own browser, terminal, and file system. That means it can get into any tool I would use with a mouse and keyboard, including the old undocumented CRM, the internal intranet, the government portal. It works with everything that doesn’t have an API and, by extension, with everything else.
How It Fits into the Stack
- It is downloaded, authenticated and billed from Cursor (the AI-powered code editor), not from xAI. The Cursor account is the Grok Bot account.
- There are desktop apps for macOS, Windows and Linux, and companion apps for iOS and Android.
- Models available behind it: the Grok 4 family, trained from the start to use tools, with context windows of 128k–256k tokens.
What Tasks It Can Actually Do
- Programming: open issues, read code, run tests, propose PRs.
- Research: browse, cross-check sources, summarize and leave the material organized.
- Sales and ops: research prospects, draft emails, update the CRM, schedule follow-ups.
- Office work: classify email, fill out forms, move data between sheets and tools.
And you can run multiple bots in parallel, with different roles, collaborating with each other while you sleep. The example that clicked for me: a support bot that learns a routine by demonstration, another that validates results, a third that escalates the unusual stuff to a person. This is no longer “a chat that knows a lot”, it’s a team that keeps working when I shut my laptop.
What It Costs and What You’re Signing Up For
| Plan | Price | Notes |
|---|---|---|
| Free (Grok chatbot) | €0 | Does not include Bot |
| Cursor Pro | ~$20/mo | Bot included from here |
| SuperGrok | $30/mo | Chatbot only |
| SuperGrok Plus | $100/mo | More usage and priority |
| SuperGrok Heavy | $300/mo | Full access to Bot |
Honest heads-up: there is no permanent free plan for the Bot, only an initial trial credit. The beta is still a beta and there’s no fixed date for general availability.
Why This Event Matters to Me
I’m going with two concrete questions, which I think are the ones that will really define how I use this over the coming months:
- Delegating work, not asking for answers. The qualitative leap isn’t that it “knows more”, but that the model does things with permission. This changes the basic unit of my day: I move from “ask and review” to “define the goal, let it run, validate”.
- How to validate what the bot delivers. If I give it the keys to my email and my repo, I need to know what patterns serious teams use to audit what the agent does. The official documentation warns, plainly, that “don’t use separate bots as a security boundary”: if I open a Gmail session for one bot, the other bots see that session. That forces you to have a permissions and review protocol before putting it into production with real data.
If the event answers those two questions with real cases not demos, then the early-morning bus is worth it.
How This Kind of Tool Affects Our Day to Day
I’ve spent weeks reading and playing with agents, and the honest reading is this: generative AI saved us typing, agentic AI saves us from deciding the small things. And that changes the day to day more than it seems.
What’s Already Happening
- Low-decision work is being automated. The Associated Press produces 40% more standardized content without growing its staff; Aviva France went from resolving 1% to 25% of its insurance claims without human intervention. These aren’t lab cases: they’re production in 2025–2026.
- Time is recovered, jobs aren’t lost (yet). McKinsey estimates that companies with well-deployed agents recover between 40 and 60 minutes per employee per day. What’s reassigned is the low-decision work: reporting, classification, searching, formatting.
- Control shifts to the human at a different point. You’re no longer on top of the click, you’re on top of the goal: defining the task well, setting limits, reviewing the result. The skill shifts from “execute” to “define and audit”.
How It’s Specifically Changing Me
- In the morning I no longer open 8 tabs to “see what’s out there”. I ask an agent to sweep the inbox, the feed, the mentions and return only what’s actionable.
- In code, I delegate the mechanical tasks (boilerplate, tests, boring migrations) and keep design and review for myself.
- In content, I spend more time thinking about what I want to say and less fighting with the editor.
- The rule that works for me: if a task bores me, it’s a candidate to be delegated. If a task matters to me, it’s a candidate to be validated twice.
What Still Worries Me
- Real isolation vs. marketing isolation. Grok Bot shares a single computer among all your bots. That’s fine for prototypes; in production, with clients or sensitive data, you need to separate well who does what.
- Prompt injection. If the bot browses, a malicious website can slip instructions into it. It’s the same risk as a human clicking links, but at 24/7 scale.
- No budgetary “kill switch”. An always-on bot consumes resources. If you don’t keep an eye on the bill, you’ll get surprises.
- Dependency on Cursor. The account, billing, privacy and data retention all run through Cursor, not xAI. Before uploading anything serious, you have to read both policies.
What I’m Going to Look for at the Event
I’m not going for fireworks. I’m going for three things:
- Real delegation cases, not scripted demos. I want to see what tasks — and which ones — people are giving to an agent today.
- Validation patterns. How they audit what the bot delivers: review by another bot, human review, tests, snapshots. Whatever, but something I can copy on Monday.
- Failure stories. Demos always show what works. What really teaches you are the cases where an agent went off the rails. If someone shares one, I’ll know the trust they’re asking for is deserved.
TL;DR
- Grok Bot is the autonomous agent from xAI/SpaceXAI that lives in a Linux cloud, browses for you and executes multi-step tasks without APIs.
- The cheapest plan is Cursor Pro ($20/mo); the full one is SuperGrok Heavy ($300/mo). Beta since August 2026.
- The real change for our day to day: we move from asking for answers to delegating work and validating results. The 21st-century skill is no longer execution, it’s defining the task well and auditing what’s delivered.
- Tomorrow I validate it in person. If I come back with something better than this, I’ll update it.
If you want to follow the event with me or have specific questions you’d like me to ask, drop them in the comments.
