2.75×
more efficient than native plan mode
Same task, same session — 130k tokens saved
Graph-powered discovery plus a structured blueprint template beats native plan mode on the identical task: 74k tokens against 204k, 43 tool calls against 133 — and the quality gap closes with one template change.
The same feature was planned two ways in the same session: native plan mode, then a Composure blueprint. Plan mode rediscovers the codebase by reading and searching; the blueprint queries a graph that already knows the structure, then forces per-file specs that prevent mid-build course corrections.
The blueprint did it in 74k tokens and 43 tool calls against plan mode's 204k and 133 — 130,000 tokens saved on one task. A single template change then lifted blueprint quality from 28/40 to 39/40 without giving back the efficiency.
- 74k tokens vs 204k — 130k saved on one task
- 43 tool calls vs 133
- One template change closed the quality gap, 28/40 → 39/40
This is what Pro delivers.
Not features for their own sake — measurable leverage on every session.