Intermediate Agentic AI
10 MIN READ
Agent Harnesses: Why the Code Around the Model Matters as Much as the Model
An agent is a model plus a harness: the runtime that runs the loop, offers tools, manages context and memory, enforces permissions and coordinates sub-agents. What's inside harnesses like Claude Code and Codex CLI, what 2025-2026 research says about which design choices matter, and the newest idea: letting agents improve their own harnesses.