Video summary

What AI Should You Use? My Honest Answer

Main summary

Key takeaways

Technology

Technological concept: Using an “AI harness” instead of choosing a single chatbot

The speaker argues that instead of picking one “go-to” AI (e.g., Claude, ChatGPT, Gemini, Grok, Copilot), he primarily uses an AI harness—a software layer that:

  • Routes between different AI models
  • Connects to multiple data sources and tools
  • Can take actions on the user’s behalf using the user’s credentials

Example harness/product: Mindstone Rebel

He says his main workflow runs through Mindstone Rebel. Key capabilities include:

  • Model selection and routing
    • Choose which connected model(s) drive the intelligence, rather than being locked to one provider.
  • Connect multiple subscriptions/accounts
    • Use different provider accounts depending on the task (e.g., ChatGPT Pro or Claude Pro).
  • Large ecosystem of integrations (examples named)
    • Gamma (presentations)
    • Google Workspace (Docs, calendar, email)
    • Notion
    • Outlook
    • Fathom (meeting-related)
    • HeyGen (video creation)
    • Kit / ConvertKit (email/newsletter)
    • Nano Banana (image generation)
    • Plaud (meeting recording)
    • Plus actions involving ThriveCart (payments/subscriptions)
  • Parallel task execution
    • Can run multiple actions/conversations at once (e.g., fetching email summaries while retrieving other information).

Automation examples (what he has Rebel do)

  • Query subscriber activity from Kit (e.g., “tell me about my recent subscriber activity”).
  • Retrieve “most important” missed emails and summarize/flag them.
  • Search a community/shared space for relevant community content and make pitches/recommendations based on likely relevance.
  • Handle operational tasks:
    • Tag users in ConvertKit
    • Add/manage subscriptions/memberships in ThriveCart
  • Pull down transcripts from content he recorded (e.g., keynote/video) to support outputs like blog posts.

Differentiation vs provider-built harnesses (e.g., Claude Code)

He highlights several comparisons:

  • Model flexibility
    • Claude Code reportedly leverages only Anthropic models.
    • Rebel can work with multiple model providers and choose what’s best based on task routing.
  • Integration/connector breadth
    • Claude Code has “far more limited” MCP connectors.
    • Mindstone claims it has built custom MCP connectors to make integrations easier.
  • User-biased instead of provider-biased
    • Provider harnesses are portrayed as biased toward the LLM provider.
    • Rebel is described as biased toward the user’s connections, preferences, and goals.
  • Community memory / shared file access
    • Rebel can access shared file systems between people, then read/edit/write and contextualize to the user’s environment.

“Community skills” tutorial-style workflow (AI Junto example)

The speaker describes a learning community (“AI Junto”) with a shared skills space (shared uploaded “skills”). Rebel can:

  1. Read skill files from the shared community folder
  2. Evaluate them against the user’s likely use cases
  3. Recommend relevant skills
  4. Execute the chosen skill on user-provided sources (like transcripts)

The specific recommended skill

  • “Review Junto call / missed Junto call into personal intelligence” (as described in the transcript)
  • Claimed value:
    • Not only summarizing, but filtering the transcript through the user’s projects, producing “four buckets”
    • Making missed calls compound into outputs (e.g., book/newsletter)

The action he performs

  • He asks Rebel to download that community skill and apply it to the transcript of a missed Junto meeting on July 31st, with the transcript stored in Google Drive.
  • He explains that Rebel will rewrite/adapt the shared skill to fit his specifications/preferences and then run it on the retrieved transcript.
  • He provides a scoped, step-by-step process Rebel follows:
    • Read the Junto shared space
    • Inspect relevant tool protocol schemas
    • Search for the July 31 meeting transcript
    • Retrieve transcript content
    • Apply the review skill
    • Deliver a final chat response with source attribution

Claimed outcome/analysis: productivity and “recall” enhancement

  • He asserts he runs many prompts daily (e.g., “30 or 40 a day”).
  • Benefit described:
    • Builds cross-meeting and cross-interaction “recall” that a human brain can’t easily maintain.
    • Converts knowledge into downstream content: books, newsletters, podcasts, videos, etc.
  • The harness is framed as making him feel “superhuman” by directing him to relevant sources and doing multi-step reasoning/action.

Main speakers / sources

  • Speaker: Jeremy (referred to as “Jeremy” throughout the subtitles; his name appears to be “Jeremy”)

Product/company references (sources)

  • Mindstone Rebel (Mindstone)
  • Other model providers/tools mentioned:
    • Claude Code (Anthropic)
    • Codex (OpenAI)
    • Gemini
    • Grok
    • Copilot (Microsoft)
  • Meeting/transcription/publishing tools mentioned:
    • Fathom, Plaud, Gamma, HeyGen, Kit/ConvertKit, ThriveCart, Nano Banana

Community source

  • AI Junto (shared skills space)

Original video