harness-engineering
Read this series in order.
Each part builds on the one before it. Reading time, level, and the concepts covered are shown for every part below.
SeniorAdvanced
13 articles
296 minutes
32 concepts
Parts
Articles in this series
- Harness Engineering — What This Series Is, and Why You Should Read It in Order The harness is the OS layer models cannot give you.
- What a Harness Actually Is (and What It Is Not) Separate the model CPU from the operating layer.
- The Four Primitives Every Working Agent System Has Five systems converged on the same four primitives.
- The Reasoning Sandwich: Why More Thinking Made My Agent Worse More reasoning in the wrong place made the agent worse.
- Coordinator Mode: A Working Multi-Agent System, From the Source Claude Code's coordinator is a real multi-agent blueprint.
- Replay Safety: The Bug That Breaks Every HITL Workflow Replay turns approval workflows into duplicate side effects.
- Skills as Information Architecture, Not Features The right markdown at the right time changed the pass rate.
- Prompt Cache Is Architecture: Designing Around the 50K-Token Mistake Cache stability is a systems boundary, not an optimization.
- The Session-Memory Feedback Loop (ACE + Codified Context) Sessions need a memory loop, not a longer prompt.
- The Org-Harness Thesis: Why Context Does Not Transfer The durable moat is workflow context competitors cannot copy.
- The Numbers That Killed the 'Wait for Better Models' Excuse The receipts say the harness moved the score, not the model.
- Build Your Own Harness: A 6-Week Plan for a 3-Person Team A six-week plan for shipping the harness without boiling the platform.
- The Ten Pitfalls (and How to See Them Coming) Every pitfall has a symptom before it has a postmortem.