AI CLI tools 1.0: Claude Code + Codex CLI + Gemini CLI — three terminal-first agent surfaces, three different bets

4 sources

Claude Code 1.0 (Skills integration + sub-agent orchestration + /mcp slash command), Codex CLI 1.0 (image input + parallel tool calls + sandbox execution model + AGENTS.md convention), and Gemini CLI 1.0 (extension system + multimodal input + /memory slash command) are three terminal-first AI tools that all shipped 1.0 in the same week. Each takes a different bet on what a terminal AI tool is: Claude Code bets on Skills + sub-agents (Anthropic Agent SDK-aligned), Codex CLI bets on sandbox isolation + parallel tool calls + the AGENTS.md convention (OpenAI Agents SDK-aligned), Gemini CLI bets on extensions + multimodal (Gemini model-aligned). The three are not equivalent: Claude Code leans into the Skills knowledge primitive, Codex CLI leans into the sandbox + AGENTS.md workflow, Gemini CLI leans into extension composability.

Why now

All three shipped 1.0 in the same week — the terminal-first AI tool market just stabilized across three distinct bets, and creators building CLI workflows can now pick the tool that matches their mental model.

Why it is worth publishing

Demo potential: a side-by-side of the same CLI workflow (a multi-file refactor + parallel tests + AGENTS.md / Skills / extensions) across all three CLI tools.

Evidence basis

Three independent vendor primary sources + The Decoder weekly roundup

“Three AI CLI tools shipped 1.0 in one week — Claude Code, Codex CLI, Gemini CLI — and each takes a different bet on what a terminal AI tool is: Skills + sub-agents, sandbox + AGENTS.md, extensions + multimodal.”

Angle

Frame the three CLIs as 'three different bets on what a terminal AI tool is' — Claude Code on Skills + sub-agents, Codex CLI on sandbox + AGENTS.md, Gemini CLI on extensions + multimodal — and use the bet comparison as the lens for picking a CLI tool.

Format

Long-form explainer

Demo idea

Record a 12-minute explainer: 2 min intro on 'three different bets' framing, 3 min per CLI (Claude Code / Codex CLI / Gemini CLI), 3 min on a live side-by-side of the same multi-file refactor + parallel tests across all three.

Platform notes

Specific Skills SDK integration, Codex sandbox backend, AGENTS.md schema, Gemini extension API surface, and multimodal context length beyond the captured summary were not extracted; confirm specific primitives against the underlying CLI docs before stating them on the record.

Usable claims

Anthropic shipped Claude Code 1.0 GA — Skills integration, sub-agent orchestration (Claude Code can spawn isolated sub-agents for parallel tasks), /mcp slash command for explicit MCP server connection.
OpenAI shipped Codex CLI 1.0 — image input support, parallel tool calls (multiple tool calls in one turn), sandbox execution model, AGENTS.md convention for project-level agent instructions.
Google shipped Gemini CLI 1.0 — extension system (third-party packages that add commands and tools), multimodal input support (image + audio + video), /memory slash command for persistent project context.

Evidence pipeline

From the news

Breakdown

The three AI CLI tools take different bets on what a terminal AI tool is — Claude Code on Skills + sub-agents, Codex CLI on sandbox + AGENTS.md, Gemini CLI on extensions + multimodal. This explainer uses the bet comparison as the lens for picking a CLI tool, rather than collapsing them into 'all are 1.0 CLIs, pick one'.

Sources

Risks

Use The Decoder and IT之家 as media-type corroboration, but read the underlying tool docs for any specific primitive or API surface before stating it on the record. Verify specific capability claim against the underlying vendor docs and the actual license / pricing matrix before stating it on the record; do not paraphrase per-platform pricing or license terms into specific dollar figures or commercial-use clauses.
Vendor docs confirm feature existence but specific primitive signatures and behavior details beyond the captured summary are not in this pass. Verify specific capability claim against the underlying vendor docs and the actual license / pricing matrix before stating it on the record; do not paraphrase per-platform pricing or license terms into specific dollar figures or commercial-use clauses.
Vendor docs confirm feature existence but specific pricing beyond the captured summary is not in this pass. Verify specific capability claim against the underlying vendor docs and the actual license / pricing matrix before stating it on the record; do not paraphrase per-platform pricing or license terms into specific dollar figures or commercial-use clauses.

Demo ideas

Side-by-side multi-file refactor + parallel tests across all three CLI tools, plot time-to-finish and isolation surface.
Bet matrix: plot Claude Code / Codex CLI / Gemini CLI on knowledge primitive (Skills / AGENTS.md / extensions), isolation model (sub-agent / sandbox / model-aligned), input surface (text / image / multimodal).
Migration story: 'migrate a CLI workflow from one tool to another', measure porting work and surface delta.