I Don’t Use One AI Model for Everything. I Built an Agent Harness Instead.
How I keep context, tools, state, and evals in one place while choosing the right model for each job
The AI Architect
Field-tested AI agent workflows for engineering leaders and builders. I test Claude Code, Codex, local models, and multi-agent systems so you can ship faster without wasting money.
Live Last read · last published · next check
Latest posts
I Built an AI Scout for the Ideas My X Feed Buried
I built an AI Scout that searches a focused feed, challenges the claims, and preserves only the ideas worth investigating.
Can Kimi K3 Build a Storefront That Feels Like a Real Product?
One storefront brief produced three distinct artifacts. The useful result was not a leaderboard.
My “Cheaper” AI Coding Handoff Cost 4× as Much
The first-edit switch wasn’t enough. Here’s what the full technique adds, and what you need before using it.
AI Is Cheap Until You Have to Do the Work Twice
I built five checks for the hidden failures that better prompts and better models don't catch.
How to Give Any AI Harness Access to the Live Web
A practical BrowserOS + MCP setup for researching dynamic websites, current prices, and real availability from Claude Code, Codex, Pi, or another compatible AI harness.
The AI Chatbot Era Is Ending. Teams Are Optimizing the Wrong Layer.
The next AI skill is not prompting. It is knowing what work to delegate, what authority to give, and what evidence proves the agent finished.
Your AI Agent Doesn’t Need Better Prompts. It Needs a Goal.
The new AI skill is not writing better instructions. It is building a harness that remembers the objective, checks the work, and keeps going until the job is actually done.
I Stopped Chasing Better Prompts. These 5 Loops Cut My AI Babysitting.
The reusable systems that made drafts, code reviews, and research passes repeatable without steering every turn by hand.
Before You Trust an AI Agent, Make It Fail Locally
The Boring Agent Test: a Microsoft Agent Framework walkthrough for developers who want traces, artifacts, and a workflow they can actually debug.