Harness Engineering: A Field Guide to Building Agents That Hold Up
Every agent demo works. Day two is where they fall over. The five layers of harness engineering, the system around the model, that decide whether an agent survives production.
Field notes
Notes from building agents that hold up.
Harness engineering, agent evaluation, and what changes in an R&D org when agents write the first draft. Written for senior engineers.
14 notesRSS feed →
Archive
2026
Designing Tools AI Agents Can Actually Use
Most agents are bad at using tools because the tools were built for programmers, not for a model reading a one-line description under load. How to design tools, and MCP servers, that agents get right.
Agent Memory Is Not a Vector Database
Teams reach for a vector DB and stop thinking. Real agents need layered memory: a working set, durable preferences, retrieval, and state that survives a restart. A practical breakdown.
Evaluating Agents When the Demo Always Passes
Checking an agent's final answer tells you almost nothing. You have to score the whole trajectory: tool choice, arguments, step count, cost, policy. How to build evals that catch real failures.
When to Reach for Multiple Agents (and When Not To)
Multi-agent is the current default answer, and it is usually the wrong one. When orchestration earns its overhead, when a single agent wins, and the patterns that hold up in production.
Ralph Mode: Why AI Agents Should Forget
The technique behind viral AI agent loops: treat context like managed memory. State lives in files, not conversation. Progress persists. Failures evaporate.
The "Untrusted Actor" Pattern: My Take on Ralph Mode
Standard Ralph gates on tests passing. My version adds an adversarial intelligence that must explicitly sign off before progress is allowed.
2025
The "Context Hygiene" Problem: Why I Rewrote My Claude Code Workflows
I was losing 40k tokens on startup. So I built Flow, a plugin that uses command stubs, progressive disclosure, and adversarial review loops.
Announcing gmickel-bench: Real-World Evals
Most AI benchmarks tell you if a model can solve a LeetCode puzzle. They don't tell you if it can ship a product.
Claude Opus 4.5 Unlocks the "No Restart" Workflow
Practical tips from two weeks of shipping with the best agentic model yet.
The Merchants of Complexity: Why AI Finally Delivers What Agile Promised
Why modern AI delivery patterns finally unlock what agile promised - and how to structure teams to ship safely.
The Imperative of Smarter Clinical Information Systems
How AI agents and openEHR can lift healthcare out of its purgatory by structuring every input and embedding it into durable clinical knowledge.
2023
Unlimited Chatbot Context & History in Under 100 Lines with LangChain + Node.js
Store everything, retrieve only what matters: long-context chat in ~100 lines with LangChain and Node.
The Importance of Context in AI: Garbage In, Garbage Out
Context quality sets the ceiling for AI outcomes; better retrieval and cleaner signals beat model hype.