More AI output just moves your bottleneck.
By Michael Mancuso · October 29, 2026 · Originally published on LinkedIn
Making one step faster rarely makes the work faster.
It just moves the line.
Anthropic merged about 2,500 code changes a day in May.
By September: over 30,000.
Writing code stopped being the hard part. Source control, builds and CI, built for human scale, became the bottleneck.
So 7 people built new source control and code review, and migrated in 5 months.
Most teams won't need their own version control. Anthropic says so.
Every team needs the question.
This week, take one change an agent finished.
Time each wait until it's live:
Review ___
Tests ___
Merge ___
Deploy ___
The longest wait is your next constraint. Fix it before you add more agents.
AI-native organizations don't measure what AI writes. They measure where it waits.
#AI #Leadership
Read slide text
Slide 1
More AI output just moves your bottleneck.
Find the next one before it breaks.
Diagram: AI output narrows into Review, merge, CI.
Slide 2
Anthropic’s merges grew more than 10x.
30,000
Merged changes a day by September.
About 2,500 a day in May.
Slide 3
Writing code stopped being the hard part.
The tools it flows through were built for human scale.
Diagram: Write → Source control, builds, CI.
Slide 4
So Anthropic rebuilt that layer.
7 people
5 months
New source control and code review.
Slide 5
Time one change from done to live.
Review ___
Tests ___
Merge ___
Deploy ___
Slide 6
Don’t measure what AI writes.
Measure where it waits.
Michael Mancuso
I design AI-native organizations
Sources cited in the slides: Anthropic, “Redesigning source control at Anthropic,” October 2026.