The Model Isn't the Product. The Harness Is.
After auditing a publicly posted Claude Code source snapshot and reading OpenAI's harness essays, I came away with one conclusion: the model is only the brain. The product is the runtime around it.
Inside Agent Systems
6 parts · Mar 4, 2026 – Apr 6, 2026
A Coding Agent Is Not a Prompt. It's a Stateful Query Loop.
The most important file in the Claude Code snapshot isn't a prompt file or a tool definition. It's a state machine. Once you see that, a lot of agent design decisions start making sense.
Prompt Cache Is Architecture, Not an Optimization
The most novel thing in the Claude Code snapshot isn't a tool or a prompt. It's how aggressively the runtime is built around preserving prompt-cache identity across forks, compaction, and long-running sessions.
The Chattiness Problem: Teaching an LLM When to Shut Up
LLMs love to talk. For a companion app, that's a problem — real people don't write essays when you text them. Here's how I built a hybrid system to control response length without killing personality.
Memory That Feels Human: A Three-Phase Roadmap
An AI companion that forgets your name after 200 messages isn't a companion — it's a chatbot with amnesia. Here's how I'm rebuilding Mio's memory from a dumb key-value store into something that feels like actual recall.
Cortex: What If Your LLM Was a Wiki Compiler?
I built a personal knowledge compiler that treats the LLM as a programmer and your wiki as a codebase. 15 source files, 5 dependencies, zero databases.