x.comthreadZazenCodes (@ZazenCodes) on X
ZazenCodes' X post defines an agent harness with a brain/body model: LLM is the brain, the harness is the body, and together they form the agent. It says Claude Code and Codex contain a harness plus an interface rather than being only the harness. Codex Core is open source in Rust (openai/codex codex-rs/core), while Claude Code's harness is closed and the open source Claude Agent SDK exposes the same runtime.
TL;DR Defines agent harness as the software running the agent loop, with Codex Core and Claude Agent SDK as examples.
Organisations named here
- OpenAI
- Anthropic
Technologies named here
- Codex
- Claude Code
- Claude Agent SDK
- Rust
- GitHub
- LLM
- Codex Core
Techniques named here
- Agent loop
Concepts named here
- Agent
- Agent harness
- ZazenCodes
Related in agents
