The 7x Agent-Team Figure Applies to Plan Mode: Measure Your Own Ratio
The slug of this note says agent teams cost seven times more. The documented claim is narrower, and the difference matters when you plan compute or explain a budget.
Anthropic's cost guide says agent teams use approximately 7x more tokens than standard sessions when teammates run in plan mode, because each teammate keeps its own context window and runs as a separate Claude instance. Outside that case, the guide says token usage is roughly proportional to team size. It also notes that agent teams are experimental and disabled by default.
The payoff has a ceiling you can derive. If a fraction p of a task can run in parallel across k agents, Amdahl's law bounds the speedup at 1 / ((1 − p) + p/k). Reproducing a paper's ablations across three independent config files might be mostly parallel. Debugging one diverging training script is not. With p = 0.5 and k = 3:
1 / (0.5 + 0.5/3) = 1 / 0.667 = 1.5x
That is the best case, before anyone integrates the pieces. If the team spends roughly three times the tokens for at most 1.5x the speed, the trade only pays when your time is the scarce resource.
Run the comparison yourself. Pick five bounded tasks from your own repo. Do each once in a single session and once with a team of two or three. For every session, record the tokens /usage reports, the wall-clock time to a result you accepted, and the fixes needed to integrate the pieces. Report two ratios per task, tokens and time, rather than one average that hides the spread. This course has not run this comparison; the numbers are yours to produce.
Chapter 18 works through when a second agent pays for itself.
