Your AI Agents Can Now Delegate to Each Other

You have three AI coding agents installed. Claude is great at architecture. Codex is fast at one-shot edits. Gemini knows your codebase from last week. But using all three means copying prompts between terminal windows, comparing outputs by scrolling back and forth, and deciding which answer to keep — all manually.

That workflow just changed. 1DevTool v1.21.0 lets your AI agents talk to each other, gives you a single dashboard for every open project, and makes invisible tool calls visible.

Your AI agents can now delegate to each other

Type @Codex write unit tests for auth.ts in a Claude terminal, and Claude spawns Codex headlessly, waits for it to finish, reads the output, and reasons over it — all without you switching terminals. This works with every major coding CLI: Claude, Codex, Gemini, Amp, OpenCode, Qwen, and Aider.

The orchestrator is built on two new MCP tools — run_agent and list_agents — exposed to every AI terminal through the MCP bridge. Your current agent calls run_agent with a target agent name and a prompt. The target runs in headless mode, and its text output flows back as the tool result.

Multi-agent workflows happen in natural language. Ask "have Codex write jokes and score them until 8/10" and the orchestrating agent handles the retry loop on its own. Mention multiple agents and the AI decides whether to run them in parallel or sequence based on your prompt.

You manage the orchestrator from Settings → MCP, where you can toggle it on or off and see both tools with their descriptions and current status.

How multi-agent orchestration works in practice

Delegate a single task

Open a Claude terminal and type something like @Gemini review this PR for security issues. Claude calls run_agent targeting Gemini, passes the prompt, and waits. When Gemini finishes, Claude receives the full response and can summarize, critique, or act on it.

Chain agents for iterative work

Ask Claude to "have Codex generate a migration, then ask Gemini to review it." Claude runs Codex first, reads the output, then passes the migration to Gemini for review — two agent calls in sequence, zero manual handoff.

Run agents in parallel for second opinions

Type "ask both Codex and Gemini to refactor this function and show me which approach is cleaner." The orchestrator spawns both simultaneously, collects their outputs, and presents a comparison. Each agent runs with a configurable timeout (default 120 seconds, max 600).

Mission Control: see every project at a glance

Press the grid icon in the toolbar and Mission Control opens as a full-screen overlay. Every project appears as a card showing what's actually running: terminal count by agent type (3 Claude, 1 Codex), open browser tabs, active HTTP requests, and database connections.

Active projects float to the top. Inactive ones are grouped below with a subtle divider. Custom commands are parsed and badged — if a terminal runs claude --model opus, the card shows a colored Claude badge with the flags spelled out.

Navigate with arrow keys, press Enter to switch to a project, Escape to close. Spring-animated transitions make the overlay feel responsive: cards shrink in from full size on open, and the selected card zooms back out on close.

MCP tool activity badges: see what your AI is doing

When an AI agent calls an MCP tool — a database query, an HTTP request, or an orchestrator delegation — a live badge now appears in the terminal area. A pulsing dot shows while the tool is running, replaced by a checkmark or error icon when it finishes, with an elapsed time counter throughout.

Click any badge to open a detail panel showing the full result. Database queries render as formatted tables. HTTP responses show status, headers, and body. Orchestrator delegations display the target agent's complete response.

Badges are color-coded by category: green for database, blue for HTTP, amber for orchestrator. They stay visible until you dismiss them individually or clear all at once — so you can review results at your own pace instead of watching them scroll past in the terminal output.

Before vs After

Workflow	Before v1.21.0	After v1.21.0
Get a second opinion from another AI	Copy prompt → switch terminal → paste → wait → manually compare	Type @Agent do X → orchestrator handles it → results appear inline
Check what's running across projects	Click through each project tab → count terminals manually	Press grid icon → Mission Control shows everything on one screen
See what an MCP tool returned	Scroll through terminal output → find the response buried in text	Click the colored badge → full result in a formatted detail panel
Run parallel AI tasks	Open separate terminals → manage timing yourself	Mention multiple agents → orchestrator runs them simultaneously

Who benefits most

Teams running multiple AI agents. If you switch between Claude for planning and Codex for execution, orchestration eliminates the copy-paste handoff. Your planning agent delegates execution directly.

Developers juggling many projects. Mission Control replaces the mental overhead of remembering what's running where. One screen, all projects, real activity counts.

Anyone debugging AI-assisted workflows. MCP badges make tool calls transparent. Instead of guessing why your AI made a certain decision, you see every database query and HTTP request it triggered.

Try it

Download 1DevTool v1.21.0 and open Settings → MCP to enable the Agent Orchestrator. Then type @ followed by any installed agent name in your terminal — your AI agents will handle the rest.