Deep dives into context architecture and enterprise efficiency
Commits up 56% on August, 504 releases to production, 33.65B tokens at 95% cache. Agents shipped 500 of the 504; a person shipped four, up from one. We publish the numbers that moved the wrong way too.
Volume fell 22% from July — and 353 of the 354 releases that reached production shipped with no human on them. The first month we published a dip.
Our biggest month — 3,401 commits across 26 active projects — and 490 publishes to production, up from 51 in June, as customers began shipping their own software on Fabric.
Our biggest month to that point — 2,493 commits across 20 active projects, 870K+ lines added — as we productized two years of AI development into Fabric and put our first customers live.
A breakout month — 1,741 commits across 16 active projects, more than double March, with an estimated 444K+ lines added as platform and client work scaled in parallel.
April engineering velocity — 520 commits across 8 active projects, ~177K lines added, led by iterative refinement and steady new-feature work.
A comprehensive look at March 2026: 295K+ lines added across 10 active projects, led by new-feature work and steady iterative refinement across the platform ecosystem.
105 billion tokens at 95% cache efficiency is the number we lead with. 500 of 504 — last month's production releases that shipped with no engineer in the loop — is the one that matters.
Read the deep dive →Every statistic on this site is real. Development figures are pulled live from the GitHub API; token & cache numbers from the Anthropic Admin Usage API. Nothing fabricated.