My workflow to spin up 3 coding agents at once: - type “ai” in terminal - launches Claude Code, Gemini CLI, & Codex CLI - fully synced windows - prompt w/ voice - hit send & watch Now Claude Opus 4, Gemini 2.5 Pro, and OpenAI o3 work on... task - you pick the winner. Guide below.show more

Full markdown guide below. Smells like the future in here, doesn’t it? — # AI CLI Automation with tmux This setup automates running the same prompt across three AI coding CLIs (Claude, Gemini, and Codex) simultaneously using tmux. It eliminates the need to manually open three terminal tabs and paste prompts into each one. ## Prerequisites - macOS with Homebrew installed - tmux: `brew install tmux` ## Installation ### What's Actually Happening? When you run these commands, you're: 1. Creating a configuration file that tells tmux how you want it to behave 2. Creating a shortcut command `ai` that does all the setup work for you 3. Enabling features like mouse clicking and keyboard shortcuts Don't worry about understanding the syntax - just copy and paste! ### 1. Set up tmux configuration Run this command to create your tmux config with all customizations: ```bash cat > ~/.tmux.conf << 'EOF' # Change prefix from Ctrl+b to Ctrl+a unbind C-b set-option -g prefix C-a bind-key C-a send-prefix # Enable mouse support (click to switch panes) set -g mouse on # Toggle sync with 's' after prefix (Ctrl+a then s) bind s setw synchronize-panes # Direct sync toggle with Option+s (no prefix needed) bind -n M-s setw synchronize-panes # Direct sync toggle with Ctrl+s (alternative) bind -n C-s setw synchronize-panes # Kill session with 'k' after prefix (Ctrl+a then k) bind k kill-session # Show sync status in status bar set -g status-right '#{?pane_synchronized,#[bg=red]SYNC ON,}' EOF # Reload the configuration tmux source-file ~/.tmux.conf ``` ### 2. Create the `ai` command alias Add this to your shell configuration: ```bash echo "alias ai='tmux kill-session -t ai_clis 2>/dev/null; tmux new-session -d -s ai_clis \; send-keys \"claude\" C-m \; split-window -h \; send-keys \"gemini\" C-m \; split-window -h \; send-keys \"codex\" C-m \; select-layout even-horizontal \; setw synchronize-panes on \; attach-session -t ai_clis'" >> ~/.zshrc && source ~/.zshrc ``` ## Usage ### Basic Workflow Here's what happens at each step: 1. **Start all CLIs**: Type `ai` in your terminal - This opens a split-screen view with Claude on the left, Gemini in the middle, and Codex on the right - All three AI tools start up automatically 2. **Enter your prompt**: Type your prompt (it appears in all three panes) - Whatever you type shows up in all three tools at once (like typing on three keyboards simultaneously) 3. **Submit prompt**: Press Enter (sends to all three CLIs) - All three AIs start working on your prompt at the same time 4. **Toggle sync off**: Press `Ctrl+s` or `Option+s` - This "unlocks" the panes so you can interact with each AI individually - You'll see "SYNC ON" disappear from the status bar 5. **Work individually**: Click on any pane to interact with that CLI - Now you can ask follow-up questions to just one AI, or give different prompts to each 6. **Kill session**: Press `Ctrl+a` then `k` when done - This closes everything at once (like closing all three terminal tabs)

This workflow is very customizable. The one I demo is great for getting different takes on solution to complex tasks from different models. But you can also imagine spinning up another workflow where maybe Gemini plans and then Claude codes that plan. Endless possibilities.

turns out a use case for apple vision pro is ultrawide screen mirroring for managing a bunch of agents together on the same screen lol

Scan any documents, convert images into text, PDF files, etc. 👍

Must get expensive to run all three constantly

Cheaper than you’d think: - Claude Code Max sub - Gemini CLI is basically free rn - o3 5x price reduction made it affordable

What about combining the 3 cli in a single process and have them ask each other for help when they need?

So I actually prefer that each works on its own so that I get 3 distinct takes/approaches

Try git worktrees to work in the same project

:)

so dope now if there was a way to critque each others' work

The paradigm of have 1 agent work and then have another review it works pretty well - especially if the models are different