Coding harness landscape
Published Updated
This is the Learning Atlas from Orbit. Get your invite
AI agents and harnesses
-
Step 1: Codex CLI
Why this matters here
Start with a terminal-native harness because it highlights direct project editing and command execution as the core workflow.
In short
An official terminal-based AI coding agent from OpenAI that reads local repositories, edits files, and executes commands autonomously.
-
Step 2: Claude Code
Why this matters here
Claude Code provides a nearby comparison point, helping learners separate the coding-agent pattern from one provider's model choices.
In short
An autonomous agentic coding assistant built by Anthropic that operates directly within the terminal to read codebases, edit files, and execute commands.
-
Step 3: Aider
Why this matters here
Aider belongs next because it shows a Git-centered approach where the harness is tightly connected to patch review and commits.
In short
A highly popular open-source AI pair programming tool that runs in your terminal, connecting multiple LLMs directly to your local codebase and Git repository.
-
Step 4: Cline
Why this matters here
Cline moves the same agent pattern into the editor, so learners see how tool access changes when work happens inside an IDE.
In short
An open-source AI coding agent that runs as an IDE extension or CLI, managing complex tasks with a strict human-in-the-loop approval system for ultimate safety.
-
Step 5: Cursor
Why this matters here
Cursor broadens the view from one agent extension to an AI-first editor where chat, edits, and codebase search are integrated.
In short
An AI-first code editor built as a fork of VS Code, famous for pioneering deep codebase understanding and real-time inline AI code generation.
-
Step 6: Windsurf
Why this matters here
Windsurf is useful here because it sits close to Cursor while emphasizing more autonomous flow and project-level awareness.
In short
An AI-native IDE by Codeium that combines a highly collaborative local agent with background cloud agents to autonomously write, test, and deploy code.
-
Step 7: Devin
Why this matters here
Devin is the endpoint because it represents the more autonomous software-engineer-agent ambition beyond editor assistance.
In short
An autonomous, cloud-based software engineering agent designed to plan, execute, and resolve complex, long-running coding tasks entirely in the background.