↓Skip to main content
Zeitgeist — a spike by Chris Gathercole
  1. Creators/

Cole Medin — AI Agents & Agentic Coding

About #

Software developer, AI educator, and founder of Dynamous AI, known for practical, hands-on YouTube tutorials on building AI agents, RAG pipelines, MCP servers, and local AI deployments. Creator of Archon, an open-source “operating system” for AI-assisted coding that manages knowledge, context, and tasks across human/agent workflows.

2026-07-09 — I Love the Karpathy LLM Wiki but it Doesn’t Scale. Here’s What Does. #

YouTube · YouTube

  • Examines why the personal markdown “LLM wiki” pattern (popularized by Karpathy) breaks down as agent memory needs grow
  • Proposes Redis Iris as a retrieval-backed memory layer to replace flat markdown files
  • Addresses multi-user/team scenarios where a single personal wiki can’t serve as shared context
  • Takeaway: for production or multi-user agent memory, move to a retrieval-backed store rather than scaling markdown files by hand

2026-07-10 — Pydantic AI 2.0: The New Best Way to Build AI Agents is Composing Capabilities #

YouTube · YouTube

  • Covers Pydantic AI 2.0’s shift to a capability-based architecture for agent construction
  • Agents are built by composing reusable “capabilities” rather than writing monolithic agent definitions
  • Aims to simplify production deployment and reduce duplicated logic across agents
  • Takeaway: structure new agent projects around composable capability modules rather than one-off agent classes to ease reuse

2026-07-12 — I Turned Claude Code Into a Complete Video Generation System (with Archon) #

YouTube · YouTube

  • Demonstrates using Archon plus Claude Code to automate end-to-end video-ad creation from a product catalog
  • Archon manages knowledge, context, and tasks across the multi-stage pipeline
  • Shows a coding-agent harness repurposed for creative/media generation, not just software tasks
  • Takeaway: coding agent harnesses can be extended to orchestrate non-coding, multi-stage creative automation workflows

2026-07-16 — This Completely Changes the Way We Build Production AI Agents (Vercel Eve) #

YouTube · YouTube

  • Introduces Vercel’s open-source “Eve” framework for building production AI agents
  • Agents are configured via markdown rather than code, with built-in reliability features
  • Positioned by Cole as a significant shift in how production agents get built
  • Takeaway: markdown-based agent configuration may lower the barrier to reliable production agents — worth evaluating as an alternative to code-first frameworks

2026-07-23 — How to Actually Run Your Coding Agent Safely (And Avoid the Horror Stories) #

YouTube · YouTube

  • Practical guide to securing autonomous coding agents using Docker sandboxes
  • Addresses risks of agents with unrestricted filesystem and database access
  • Motivated by real “horror stories” of agents causing damage when run without containment
  • Takeaway: sandbox any coding agent with broad filesystem/DB access in Docker before granting it autonomous execution

2026-07-24 — Is Kimi K3 Really That Good?! (Don’t Just Believe The Hype) #

YouTube · YouTube

  • Benchmarks Kimi K3 against Claude Opus 4.8 through hands-on engineering tasks rather than published scores
  • Examines the reliability gap between open-weight and closed models
  • Stress-tests hype claims with real-world testing, consistent with his benchmark-skeptical stance
  • Takeaway: don’t take open-weight model hype at face value — evaluate reliability on actual engineering tasks, not just benchmarks