Can One Developer Do the Work of Five? Exploring GitHub Copilot CLI’s /fleet Parallel Agent Feature

/fleet Overview

The /fleet command adds an "AI foreman" (orchestrator) on top of Copilot, allowing you to break a big requirement into smaller tasks, identify which can run in parallel, dispatch multiple "sub‑agents" to work on different files, and finally collect and package the results.

How It Works

Define the overall goal and let the orchestrator split it into sub‑tasks.

Mark tasks that can run concurrently and those that must be queued.

Each sub‑agent receives its own context window but shares the same codebase; they do not chat directly.

The orchestrator coordinates execution, resolves dependencies, and merges the outcomes.

Because agents do not share chat history, prompts must be self‑contained. The orchestrator prevents conflicts by ensuring only one agent modifies a given file.

Basic Usage

/fleet refactor auth module, update tests, and fix docs/auth/*

Adding --no-ask-user runs the command non‑interactively, suitable for scripts.

Three Prompt‑Writing Tips

Specify the deliverables : vague prompts like /fleet write docs are ignored; detailed prompts such as

/fleet create API docs: auth.md write login flow, endpoints.md write examples, errors.md write error codes

let the orchestrator parallelise the three docs.

Define boundaries : tell each agent which directory it may touch (e.g., src/api/ vs src/ui/) to avoid overlapping edits.

State dependencies : explicitly order dependent steps, e.g.,

first run SQL migration → then update ORM models → finally update API and tests (the last two can run in parallel)

. The orchestrator will schedule accordingly.

Personal Assessment

⚠️ Agents do not retain chat history; prompts must contain all necessary context.

⚠️ Parallel agents are not always faster; simple tasks may be more lightweight in single‑mode.

⚠️ File conflicts must be prevented manually; the AI does not auto‑merge changes.

Best suited for cross‑file refactoring, multi‑module documentation generation, and full‑stack feature development. Not ideal for one‑line config changes or isolated bug fixes.

Quick Monitoring Tricks

Run /tasks to view sub‑agent progress.

If parallelism seems missing, pause and let the orchestrator re‑split before continuing.

Place custom agents in .github/agents/ (e.g., Claude for docs, GPT for code) to leverage each model’s strengths.

Even the strongest tool requires a skilled commander; /fleet is a lever that upgrades you from executor to orchestrator.