GrokBot

Source — Riley Brown

Walkthrough by Riley Brown (@rileybrown) — Agent Native / agents in business; podcast @agentnative_.

Original post: x.com/rileybrown/status/2087235887012749383 (11 Aug 2026) — “GrokBot Just Launched… Cursor/SpaceX new general agent platform for Desktop and iOS.” Chapters: create agents + connections → plugins → skills/descriptions → cloud computer → agents talking → triggers/automations → what’s missing.

Framing in the video: product from the Cursor lineage, branded GrokBot in the SpaceX / Elon era story Riley tells; positions against Claude Cowork and ChatGPT Work as a full agent platform for knowledge work (“feels like messaging a team of AI agents”).

This page: structured notes from Riley’s tour + a concrete how Mike should use it map (Grok Build, this blog, Para Living, theflysales.com, agent portfolio).

Riley follow-up: group chats do work — he missed them on first pass (correction tweet).

One-sentence TL;DR

GrokBot is a desktop + iOS multi-agent workspace: you create named specialists (EA, content, developer…), connect tools via chat or a plugins tab, give each agent a cloud computer, let agents message each other, and run time-based routines + event triggers (Slack, Git, Linear, etc.) — synced so phone and desktop stay one system.

What you get (feature map)

Walkthrough highlights

1. Create a bot + connect tools in chat

Plus → new bot animation → “what do you mainly want me around for?” Example: email + calendar manager, also Slack + Linear. Agent pulls plugins; you authorize in-product. Then: “At 9am, morning digest: Linear, Slack, email + calendar…” → becomes a routine in the side panel (chat / computer / routines / settings).

2. Description = the real differentiator

Name + title + description is the custom agent prompt. Content agent vs developer agent only behave differently because of those strings — not because plugins are siloed. Implication: invest in descriptions the way you invest in soul.md / AGENTS.md / company brain briefs.

3. Cloud computer for the agent (not for you to browse ESPN)

Full remote desktop: agent builds a landing page, runs it on localhost:3001, screenshots progress. iOS can open the same computer. Point is execution environment for the agent — same spirit as exe.dev VMs and ChatGPT Work cloud computer, productized per agent.

4. Agents talk to each other

Developer agent voice-prompted from phone: talk to content agent, use scraped Instagram transcripts → app idea → build iOS/Android prototype → deploy (Revel-style simulator). Human gets pings from both; can open the full agent↔agent thread. Animations show which agents are “working.”

5. Triggers = Zapier with an agent brain

Time routines and event triggers. Demo: Slack channel short-form-videos → agent summarizes. At capture, trigger apps were limited (Slack, Git, Teams, Linear, Sentry, PagerDuty); Riley expects email + more soon.

What’s missing (Riley’s list) / likely soon

• GrokBot as a native Slack bot you @ in channels — Cursor bot already works in Slack; Riley expects GrokBot bots to follow (platform lineage).
• Files tab — create files in chat, not reliably showing in Files UI yet.
• More triggers — email and broader app graph.
• Groups — UI clues + agent claims; Riley later confirmed group chats work.

Riley comparison: feels like an easier cloud Buzz (Jack Dorsey’s agent-friendly Slack-like) where every agent brings a cloud computer.

GrokBot vs Grok Build (how we already work)

You like Grok and Grok Build (terminal TUI, multi-thread SOP, HEOS-style repos — /grokbuild, /grokthreads, /grok-mobile). They are not the same product:

How you should use GrokBot (Mike-specific)

Matches how you already think: one front door + specialists, 5 roles, company brain, multi-company portfolio (Para Living, GFAVIP, Headless Domains, PowerLobster, this site).

Starter fleet (create these first)

High-value automations to set in week one

1. 09:00 digest (EA) — calendar → urgent Slack → Linear/inbox themes.
2. Slack short-form / content channel trigger — summarize + draft X angle (content agent); human approves posts.
3. Weekly product brain (skincare or hero SKU) — content + product agents: competitor/listing/review pulse into a brief for theflysales.
4. Git trigger on this repo (if available) — developer agent summarizes PR risk; real merge stays human + Grok Build.
5. A2A handoff — content → developer: “turn this field note into a landing prototype on your computer” then you decide if it becomes a real Astro page in Grok Build.

Plugins / skills to prioritize

• Slack, Linear, Gmail/calendar — EA + ops skeleton
• Vercel / Cloudflare — demo deploys
• Scrape Creators–class skill — content corpus (Riley’s favorite pattern)
• Remotion / video stack only if you’re actively shipping video automation
• Avoid dumping bank / production admin into every agent — shared plugins mean description discipline + least privilege (see gymhack)

Guardrails (same lessons as our other agent pages)

• Shared plugins = shared blast radius. One Gmail plugin is available to all bots. Use descriptions + human approval for send/delete/spend.
• Cloud computer ≠ production laptop. Prototype there; ship production code through git + Grok Build / CI.
• Triggers can spam. Start with one Slack channel and a summary-only instruction.
• Don’t abandon Grok Build. GrokBot is the fleet UI; Grok Build is still the best place for deep, multi-thread engineering on your real repos.

30-day experiment plan

Related on this site

• ChatGPT Work (Riley) · Work + Codex part 2
• God agent vs team — same multi-agent philosophy
• Grok Build · threads SOP · mobile SSH
• Buzz — agent-native chat room comparison
• exe.dev — VMs when you need real Linux isolation
• Product brain · SEO loop · Scrape Creators
• Chief of staff · Claude Cowork vs Eigent

Primary source: Riley Brown on X — GrokBot launch walkthrough