Large Language Models & Coding Agents

Terminal-based presentation covering LLM internals, coding agents, prompt engineering, and harness engineering — from theory to practice.

Contents

1. Language Models — self-attention, transformers, chain-of-thought, GRPO, limitations
2. Vibe Coding and Coding Agents — what agents are, the agent loop, tooling choices
3. Prompt Engineering — giving context, saving tokens, anti-patterns (over-obedience, drifting, no verification)
4. Harness Engineering — five principles, architecture, formalism
5. Tool Surface, Skills, MCP — how agents interact with the world
6. Live Demo: Stabilizer Tableau Simulator — incremental build applying harness principles

Prerequisites

• Coding agent: Claude Code (npm install -g @anthropic-ai/claude-code) or Codex (npm install -g @openai/codex)
• Terminal: Ghostty — split into two panes (slideshow + coding agent)
• IDE: Zed or VS Code open on the side
• Optional: voice mode — macOS Dictation, Superwhisper, or your agent's built-in voice mode

Usage

# Install presenterm (https://mfontanini.github.io/presenterm/introduction.html)
cargo install presenterm

# Run the presentation
presenterm slides.md

Skills

Managed with ion. Run ion add to install skills from Ion.toml.