← Back to all playbooks

Ashe

ASHE MAGALHAES: Personal AI suite
Start: 00:59:56
Sketch note for Ashe's session
TL;DR

Build a persistent, multi-agent personal workspace in Slack where specialized bots maintain long-term context across all projects, public streams, and relationship networks.

Key Insights

Video as the 'Proof of Human' Layer

01:08:44

As AI-generated text and images become 'slop,' video remains the primary medium for authentic connection because it preserves human imperfections like stutters and physical state (e.g., being sick).

Cross-Model Debugging Orchestration

01:23:44

Use Claude Opus to orchestrate and prompt OpenAI Codex/o1 specifically for debugging loops; Opus handles the high-level reasoning while Codex/o1 provides the precision required for technical fixes.

Relational Intelligence via Co-occurrence Networks

01:06:16

Move beyond contact tables to co-occurrence networks to visualize the 'optimization landscape' of relationships, identifying who you are connected to and why based on shared context.

Systems

Ambient Thought Stream

  1. Send a text or Slack message to a dedicated agent webhook.
  2. Agent processes the input using OpenAI tool calling.
  3. A Next.js cron job updates a public-facing 'mood board' or stream page (e.g., ash.ai/stream).
  4. The entry is tagged with metadata like location or specific project context.
Tools: Slack Webhooks, Next.js, OpenAI API, Vercel

Video-as-Code Explainer Workflow

  1. Define a visual style in a TSX/JS template using Remotion.
  2. Use Claude Code with the Remotion skill to programmatically generate video frames.
  3. Iterate by feeding the Remotion Studio output back into the LLM to fix layout or animation timing.
  4. Link the final render to a personal dashboard for public accountability.
Tools: Remotion, Claude Code, TypeScript

Tools

Ash AIRemotionClaude CodeCursorOpenAI Codex 5.2Google GDELTNext.jsSlack