AI Personal Assistants: Muse, Grok Bot, Cue Usage & Comparisons

The author argues against writing custom instructions every session and proposes a fixed skill file that stores state locally, fails hard on missing variables and stops after two retries.

The poster says X Ultra offers 50x usage that no one else matches, but third-party models are excluded, while Opus 5.5 counts as a first-party model in Grok Bot.

Monitoring platform agentstatus.dev now lists iMessage as a host, so personal assistants living in iMessage can track which tools fail during conversations.

The poster asks what to call the category of personal assistants like SpaceX's bot, OpenAI's Dots and Manus's Cue, comparing it to the ADE (Agent Development Environment) label, and invites other products and a definition.

Grok Bot 0.70.2 lets bots claim their own email addresses to send and receive mail, search X without an account, embed Figma/Notion/Canva interfaces in chats, pay via X Money or Stripe Link, and adds approval checks before irreversible actions.

The official leaderboard shows Grok Bot at 38% and Grok at 37.5%, while notional volume from ChatGPT traders nearly quadrupled week over week.

The author estimates a roughly $140B revenue pool for personal AI assistants in 2030, and argues that split among 5+ players it does not justify Meta's current expectations.

The author shares his Grok Bot lineup and describes onboarding Muse, Codex and others onto Notion for cross-assistant collaboration, noting the lack of a Slack-style messaging layer.

A post with an image comparing OpenAI's ChatGPT Dot and Grok Bot for anyone choosing a personal assistant.

Someone asked Grok Bot to make a viral-figure fight video, and it delivered a Conor McGregor vs. Mr. Bean matchup whose contrast makes it hard to scroll past.

A demo shows Grok Bot with the open-source REA browser tools inspecting a TradingView-style terminal's page structure, requests and interaction logic, then writing and testing its own implementation from those findings.

The author names a chief of staff and five specialist agents; one overnight run produced 36 outreach drafts with none sent, and the morning brought a single Telegram report.