By Thomas Cohen, founder of Maestro
What our ×10 measures, and what it does not
We publish one figure: around 10 times less methodological context injected into agents. Here is the full measurement, including its limits.
When a team of AI agents works, each agent receives methodological instructions: who it is, what it must produce, in which format, and with which checks. This text takes up space in its working memory, or ‘context’. The more verbose the method, the less room remains for what matters: your project.
The protocol
We compared Maestro's method with BMAD-METHOD v6, the field's reference open method, over an identical cycle: from the idea to the first verified feature. On both sides, we reconstructed the instructions exactly as they are sent to the agent, phase by phase: scoping, specification, architecture, breakdown, construction and verification.
The result: around 6,400 method tokens for Maestro, and around 65,700 for BMAD-METHOD. A ratio of around 10, obtained through counting choices deliberately unfavourable to Maestro: BMAD's optional files are not counted, and the token estimate slightly penalises French.
Why restraint matters
A concise method lets the agent read your project rather than its manual. It also makes each stage less demanding to process, because the same work carries less text. And it ages better: our documents are short by design, a brief fits on one page and a work step is self-contained, so agents at later stages reread lightweight documents.
What the figure does not tell you
This benchmark measures the methodological text injected, not a project's final bill: that also depends on the code agents read and the responses they produce. Nor does it measure the quality of the documents produced. And matching phases between two methods that divide work differently involves some interpretation; we document the counting details so anyone can repeat the calculation.
We prefer an honest, bounded figure to a sweeping promise. If we ever measure billed costs from start to finish, that will be another article, with the same level of detail.