Skip to content

zhao1bo-source/harness-engineering

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

40 Commits
 
 
 
 

Repository files navigation

Harness Engineering Notes

Practical writing on Context Engineering, Agent Architecture, and Human-AI Collaboration Systems — from real-world experiments building AI-native products and teams.

"The model is public infrastructure. Your context is the moat."


A model is only as good as the system around it.

Harness Engineering is the discipline of designing everything that sits between a frontier model and a real outcome: context pipelines, tool orchestration, memory architecture, agent loops, skill systems, organizational workflows, and the human-AI collaboration patterns that make it all work.

Most teams obsess over which model to use. The teams that actually win obsess over their harness.

This repository collects my ongoing thinking and experiments at the intersection of:

  • Context Engineering — how to make implicit knowledge explicit, testable, and evolvable
  • Agent Architecture — how to design systems where humans and agents form effective working units
  • Organizational Harness — how to restructure teams around agents as the unit of work, not people

Articles

Title Core Idea Read
It's Not the Model. It's the Context. Context failures are silent. Teams blame the model when they should debug the context. Introduces the Context Maturity Ladder (5 levels) and the case for eval-driven context development. articles/en/context-not-model.md
AI Coding's Harness Engineering Guide Guides (feedforward) raise the probability an agent gets it right the first time. Sensors (feedback) help it self-correct. The critical rule: if a constraint can be enforced computationally, never delegate it to an LLM. Maps the full harness lifecycle from pre-commit to continuous drift detection. AI Coding's Harness Engineering Guide
Organizational Practice and Reflections by Harness Field notes from restructuring a team at Meituan around agents as the primary work unit. Context is the organizational foundation — agents need the same information density as humans to participate in real thinking. The unexpected finding: AI doesn't make people's work easier. It makes people more expensive. Organizational Practice and Reflections by Harness

About

I'm Yibo Zhao — a product builder and AI practitioner based in Beijing.

Background: 6 years leading innovation businesses at Meituan (China's leading local services platform) + 5 years as a startup co-founder. For the past two years I've been running hands-on experiments with AI-native team structures, context engineering, and agent-centric organizational design.

About

Context Engineering, Agent Architecture & Human-AI Collaboration Systems — notes from real experiments.

Topics

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors