May 11
Loading video player...
How AI Coding CLIs Actually Work: Codex, Claude Code, Gemini CLI This video is an architecture deep dive into the new generation of AI coding CLIs. Instead of doing a shallow product comparison, it breaks down the shared system design behind tools like Codex, Claude Code, and Gemini CLI: - layered instructions and repo-local guidance - context assembly and transcript compaction - tool runtimes, MCP, approvals, and sandboxing - plan state, checkpoints, and subagents - the reference architecture you would build if you were creating one from scratch Sources are official docs current as of March 11, 2026. Primary sources: - OpenAI Codex: https://developers.openai.com/codex/cli/ - OpenAI AGENTS.md: https://developers.openai.com/codex/cli/agents-md - OpenAI Skills: https://developers.openai.com/codex/cli/skills - OpenAI MCP: https://developers.openai.com/codex/cli/mcp - OpenAI Tasks: https://developers.openai.com/codex/cli/tasks - OpenAI Multi-agent: https://developers.openai.com/codex/cli/multi-agent - Anthropic Claude Code overview: https://docs.anthropic.com/en/docs/claude-code/overview - Anthropic Hooks: https://docs.anthropic.com/en/docs/claude-code/hooks - Anthropic Subagents: https://docs.anthropic.com/en/docs/claude-code/subagents - Anthropic MCP: https://docs.anthropic.com/en/docs/claude-code/mcp - Gemini CLI docs: https://geminicli.com/docs/cli/ - Gemini CLI Architecture Overview: https://geminicli.com/docs/architecture/ - GEMINI.md: https://geminicli.com/docs/core/gemini-md/ - Gemini CLI Plan Mode: https://geminicli.com/docs/core/cli/plan-mode/ - Gemini CLI Trusted Folders: https://geminicli.com/docs/trusted-folders/ - Gemini CLI Checkpointing: https://geminicli.com/docs/checkpointing/ If you are building agent tooling, the main lesson is that the terminal UI is the thin shell. The real product is the control plane behind it. #ai #cli #codex #claudecode #geminicli #softwarearchitecture Chapters 0:00 How AI Coding CLIs Actually Work 0:40 Why Terminal Agents Won 0:54 The Shared System Diagram 2:18 Instructions And Context 3:40 Tool Runtime, Approvals, And Sandboxing 5:05 Planning And Subagents 5:44 Codex, Claude Code, And Gemini CLI 8:15 Reference Architecture 9:35 Failure Modes That Break Trust 10:13 The Real Product