Learning Markdown-Based AI Orchestration: Skills, Ralph & Open PROSE

Описание: In this episode of the AI First Show, Leonardo Gonzalez and Stephen Barr explore three powerful markdown-based approaches to orchestrating AI agents. Starting with a viral Andrej Karpathy tweet about English becoming a programming language, we dive deep into how structured markdown files can control AI workflows more efficiently than traditional frameworks.

*What you'll learn:*
*Agent Skills Standard (skill.md)* - A context-efficient way to load instructions on-demand, avoiding context window bloat from loading dozens of MCP servers
*Ralph* - An iterative bash loop approach that feeds one task at a time to LLMs with fresh context, enabling autonomous coding sessions that run for hours
*Open PROSE* - A revolutionary way to define flow control and multi-agent orchestration entirely within markdown, creating a "virtual machine" inside the LLM conversation

We cover practical demos, discuss context window management, and explain why "marrying the harness and dating the model" is the key philosophy for AI-first development.

🔗 *Resources:*
Agent Skills Standard: https://skill.md / https://agentskills.io
PROSE: https://prose.md
Agents.md: https://agents.md

💬 Join our Discord community   / discord  

---

*Chapters:*

00:00 Intro & Welcome to AI First
00:34 The Three Pillars: Using Markdown to Orchestrate AI
01:47 English as a Programming Language (Andrej Karpathy Tweet)
04:08 Pillar 1: Agent Skills Standard (skill.md)
06:15 Pillar 2: Ralph (Iterative Bash Loop)
08:50 Pillar 3: Open PROSE (Workflow Orchestration)
12:08 Deep Dive: How Agent Skills Work
14:15 Context Window Efficiency & Compression
18:18 Deep Dive: Ralph & The PRD Spec
21:09 Analyzing the Ralph Bash Script
22:15 Converting PRD to JSON & Verifiable Tests
24:22 Managing Context: agents.md vs progress.txt
30:18 Demo: Running the Ralph Loop
34:50 Deep Dive: Open PROSE (prose.md)
39:20 Demo: Creating a "Pirate Product Pitch" PROSE Program
41:16 Running the PROSE Virtual Machine
43:51 The Shift: AI Models Becoming Commodities
46:28 Demo: Prioritizing Linear Issues with PROSE
49:43 Summary: Key Principles of AI Orchestration
53:01 Q&A: Recommended Coding Models (GLM 4.7, GPT 5.2, Opus 4.5)
55:20 The Rise of Cheap & Efficient Models (MiniMax, Nvidia, Debra)
58:55 "Marry the Harness, Date the Model"
01:00:14 Outro & Discord Invite

---

#AIFirst #AIAgents #Markdown #PROSE #Ralph #AgentSkills #LLM #ContextEngineering #AndrejKarpathy #AICoding #PromptEngineering