Harness Engineering: What Separates Top Agentic Engineers Right Now
Everyone says "harness engineering" is the most important skill in AI coding for 2026, but almost nobody can tell you what it actually is. A coding agent is the model plus the harness around it, and the harness is the part that's more important! One teardown of Claude Code found roughly 98% of it is the harness, not the model. It has two levels. The first is the AI layer around a single session: your rules, skills, hooks, subagents, etc. And here it's all about using every LLM mistake as an opportunity to evolve your AI layer (your harness). The second, where the real leverage is, is orchestrating multiple coding agent sessions into one workflow, each with a focused job. This makes it possible to reliably take on larger tasks and is how you really scale coding agents. ~~~~~~~~~~~~~~~~~~~~~~~~~~ - Check out the new Google Cloud Agents CLI! I've been using their new Agents CLI to scaffold and deploy production agents straight from my terminal with the model-agnostic ADK, no cloud console required: https://fandf.co/3PsygXP ~~~~~~~~~~~~~~~~~~~~~~~~~~ - The Dynamous Agentic Coding Course is now FULLY released - learn how to build reliable and repeatable systems for AI coding: https://dynamous.ai/agentic-coding-course - Companion repo (the harness, built): https://github.com/coleam00/harness-engineering-demo - Harness Engineering article 1: https://addyosmani.com/blog/agent-harness-engineering/ - Harness Engineering article 2: https://martinfowler.com/articles/harness-engineering.html ~~~~~~~~~~~~~~~~~~~~~~~~~~ 0:00 What is Harness Engineering? 0:47 Defining the Agent Wrapper 1:55 The Three Layers of AI 4:39 Harness vs. Context Engineering 5:46 The Harness Engineering Mindset 7:45 Sponsor: Google Cloud Agents CLI 9:25 Components of the AI Layer 12:52 Orchestrating Coding Agent Sessions 14:34 Automating with the Ralph Loop 16:34 Final Thoughts ~~~~~~~~~~~~~~~~~~~~~~~~~~ Join me as I push the limits of what is possible with AI. I'll be uploading videos weekly - at least every Wednesday at 7:00 PM CDT!
