Viral Claude Code guide is spreading misinfo on agent teams
Claims teams cut token usage → Anthropic docs say the opposite. Teams burn MORE tokens because each teammate runs its own context window
How it actually works:
• Agent teams = experimental feature (needs CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1)
• No --teammates flag exists. You ask Claude in plain English for teammates
• Custom roles? Define them in subagent configs
The real token saver: Advisor tool
• Main model stays cheap, only pings Opus/Fable at critical decision points
• Turn it on: /advisor opus or claude --advisor opus
DYOR before you shill features you don't understand
Claims teams cut token usage → Anthropic docs say the opposite. Teams burn MORE tokens because each teammate runs its own context window
How it actually works:
• Agent teams = experimental feature (needs CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1)
• No --teammates flag exists. You ask Claude in plain English for teammates
• Custom roles? Define them in subagent configs
The real token saver: Advisor tool
• Main model stays cheap, only pings Opus/Fable at critical decision points
• Turn it on: /advisor opus or claude --advisor opus
DYOR before you shill features you don't understand