ShawhinT

Subagents VS Teams — Development skill for Claude Code

Development community

A repository for comparing the performance of Claude Code's subagents and (experimental) agent teams features.

How to install Subagents VS Teams

This entry records only its repository, not the path inside it, so there is no exact command to give. Open ShawhinT/subagents-vs-teams and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

What Subagents VS Teams does

A repository for comparing the performance of Claude Code's subagents and (experimental) agent teams features.

Alternatives in Development

Prioritize Features — Prioritize backlog by impact, effort, risk, and strategic alignment 7.8k ★
Notebooklm Skill — Use this skill to enable Claude Code to communicate directly with your Google NotebookLM notebooks 5.9k ★
Expand Project — Expand an existing project with new features 1.8k ★

README

Subagents vs Agent Teams

Benchmarking Claude Code's **subagent** approach (single session using the Task tool) against **agent teams** (multi-session via `CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS`) on identical tasks. Compares token usage, compute time, and output quality.

Resources:

YouTube Explainer

Results

**Model**: Claude Sonnet 4.6 (Claude Max)

Task	Metric	Subagent	Agent Team
Task 1: Lead List	Run time	32.6 min	24.4 min
	Est. unique tokens	~232k	~265k
	Output quality	50/50 emails	8/50 emails
Task 2: Web App	Run time	13.4 min	10.9 min
	Est. unique tokens	~99k	~111k
	Output quality	Better design and features	Worse design, no progress bar
Task 3: Landing Page	Run time	6.3 min	21.1 min
	Est. unique tokens	~110k	~218k
	Output quality	Better design	Better copy

How to Run

Each task follows the same pattern:

Subagent run — Open Claude Code in task-N/subagent/ and paste prompt.txt. Claude completes the task in a single session, using the Task tool as it sees fit.
Agent team run — Open Claude Code in task-N/team/ and paste prompt.txt. The settings.local.json in that directory enables the experimental agent teams feature.
Evaluate — Use evaluate.md to parse JSONL session logs from ~/.claude/projects/ and produce task-N/README.md.