The Cost-Benefit Framework

Agent teams are worth it when the quality improvement or time savings justify 5-10x the cost of sub-agents. That is the decision framework in one sentence. The challenge is quantifying both sides.

On the cost side: count agents, estimate tokens per agent, multiply by your API rate. A 5-agent team running for 10 minutes on Sonnet might cost $3-5. The same team with Opus agents and 3 debate rounds might cost $30-50. On the benefit side: what is the value of the output? A security audit that catches a vulnerability before production is worth thousands. A competitive analysis that informs a $500K deal is worth the $20 team cost.

1. High-Stakes Tasks: Use Teams Security audits, architecture reviews, competitive analysis for major deals, compliance reviews. When getting it wrong costs more than the token spend, invest in the debate and critique patterns that agent teams enable. The $50 team cost is noise against a $500K deal risk.
2. Routine Tasks: Use Sub-Agents Codebase searches, file analysis, parallel research on independent topics, test generation. These are independent tasks that do not need inter-agent communication. Sub-agents handle them at 5-10x lower cost with comparable quality.
3. Learning Tasks: Start Simple, Scale Up When exploring a new domain or pattern, start with a single agent. Graduate to sub-agents when you need parallelism. Graduate to teams only when you need debate or iteration. Each tier adds cost and complexity — only escalate when the lower tier is genuinely insufficient.