Understanding Fantastic
Coding Agents

A deep technical exploration of how modern AI coding agents work — from architecture to implementation details.

Codex CLI

OpenAI · Rust · Kernel Sandbox

0

Rust Crates

0+

Tools

Claude Code

Anthropic · TypeScript · LLM Classifier

0+

Tools

0+

Modules

The Agent Loop

Hover on a step to see what happens underneath

Codex CLI

User Input

↓

TurnContext

↓

Build Prompt

↓

Stream API

↓

Tool Futures

↓

RwLock Exec

↓

Drain

↓

Continue?

Claude Code

User Input

↓

Auto-Compact

↓

System Prompt

↓

Stream API

↓

Streaming Exec

↓

Batch Tools

↓

Yield Results

↓

Continue?

Click a cell to see implementation details

Capability	Codex CLI	Claude Code
File Search	Shell (rg) Model runs rg via shell command	GlobTool + GrepTool Built-in ripgrep wrappers with rich params
File Editing	apply_patch Unified diff/patch format	FileEdit String replacement (old→new)
Shell Execution	shell Execute in kernel sandbox	Bash Execute with permission checks
Web Access	Shell (curl) Via shell commands	WebFetch + WebSearch Dedicated tools with AI processing
Multi-Agent	spawn_agents_on_csv In-process agents with forked history	AgentTool Separate child processes
Code Intelligence	—	LSPTool Language Server Protocol integration
Permissions	request_permissions Model can request new permissions	Permission Dialog Interactive approval + LLM classifier