2025-02-17 Mon

#Nice
- Just saw this great description of MCP from MCP/Web/Intro:
  - Think of #MCP like a USB-C port for AI/Applications. Just as USB-C provides a standardized way to connect your devices to various peripherals and accessories, MCP provides a standardized way to connect AI models to different data sources and tools.
#Filed
- #Agents
  - Altera/YouTube/1000 AI NPCs simulate a CIVILIZATION in Minecraft - YouTube
- #Art/Digital, #AI/Model/Diffusion, #ComfyUI,
  - Person/comfyanonymous/GitHub/ComfyUI_examples
- #Agile
  - Backlog Grooming
- #Article
  - LangChain/Blog/25/02/Benchmarking Single Agent Performance
    - Key Insight
      Both more context and more tools degrade agent performance
      Agents that require longer trajectories degrade more quickly
      #o1, #o3-mini, and #Claude 3.5 Sonnet are comparable and in a different league than #4o and Meta/AI/Model/llama/3.3-70B
      #o3-mini performs as well as #o1 and #Claude 3.5 Sonnet with smaller context, but sees a steeper performance drop as context grows
- #Art
  - Art/Work/The Sudarium of Saint Veronica
- #Book
  - Book/Building a Second Brain
  - Thinking, Fast and Slow
- #Idea
  - CursorAI/Idea/Proj Rules GitHub CLI AI SLDC/MCP/Github Action Reasoning Model Planner
    - Idea - develop a GitHub Action which uses MCP to interact with GitHub
      As a result of the above, we could develop a GitHub Action which is itself an MCP client that could utilize a #GitHub MCP server in the action itself to develop more advanced planning abilities. For #Example -
      an execution plan for each issue
      filing issues or sub-tasks
      organizing the project backlog to sequence work
      An API key with access to a reasoning model could be used to execute planning in the issues.
      A Github Scheduled Workflow #cron could "wake up" and do Backlog Grooming on the project
      When a git commit is pushed, it could trigger a workflow that would execute a "plan" with an LLM Reasoning Model against the Best Next Action, storing an Intermediate Work Product somewhere in the GitHub Project, such as in an Issue Comment
      The local CursorAI model could use its own access to MCP to access the Intermediate Work Product left in the project to start from the instructions the LLM Reasoning Model left.
    - The difference between System 1 vs System 2 Thinking may be mirrored in fast, intuitive Local Cursor usage vs slow, CICD-initiated use of LLM Reasoning Models to conduct advanced Backlog Grooming and initiate CursorAI/Technique/Frontload Design
      Inference Time Compute cost and latency grows with the length of the Chain of Thought. It may eventually be structurally important to bifurcate when and where we use reasoning models.
      The cost structure of reasoning models encourage Cursor and likely also other AI Coding Tools to develop licensing models that restricts the use of reasoning models in their agent mode.
      But there remains an important need to develop a well-reasoned plan on what to do. CICD may have a role in doing some System/2 Thinking pre-work and sequencing to help the System/1 Thinking CursorAI/Agent make the most of its resources.
- #PKM
  - Intermediate Work Product
  - #Application
    - ReflectApp
- #People
- #Travel
  - Person/Kevin Kelly/Blog/25/02/50 Years of Travel Tips

2025-02-17 Mon

Idea - develop a GitHub Action which uses MCP to interact with GitHub

The difference between System 1 vs System 2 Thinking may be mirrored in fast, intuitive Local Cursor usage vs slow, CICD-initiated use of LLM Reasoning Models to conduct advanced Backlog Grooming and initiate CursorAI/Technique/Frontload Design

Unlinked References