Code review · three jobs · teammates' pull requests I reviewed

The review ratio.

BloombergAug 2018 – Mar 2021 · 31 months
pre-AI
PostHogMay 2021 – Jan 2025 · 45 months
pre-agent
TanoJan – Sep 2026 · 8 months
with AI agents
2 → 1.85 → 0.04teammates' PRs I reviewed for every PR of mine that merged.Bloomberg → PostHog → Tano
8 → 19 → 206of my PRs merged per month.about 25× from Bloomberg to Tano
16 → 35 → 8teammates' PRs I reviewed per month.it rose with my output, then fell
Per month · same scale in each chart

My output went up. The reviewing I did for others went down

Monthly averages over each job. Hover a bar for its value.

My PRs merged per month

Teammates' PRs I reviewed per month

Quarter by quarter

Two reviews per merged PR for years, then almost none

Teammates' PRs I reviewed, divided by my merged PRs, for each quarter. The dashed line is my Bloomberg average; Bloomberg has no quarterly data. Hover a point for the counts.

Bloomberg averagePostHogTano

Swipe sideways to see the whole chart.

The numbers

Side by side

BloombergPostHogTano
My PRs merged~2508591,649
Teammates' PRs I reviewed or commented on500+1,58967
Reviews per merged PR~21.850.04
My PRs merged per month~8~19~206
Teammates' PRs reviewed per month~16~35~8
Busiest quarter, my PRs60+94770
Busiest quarter, reviews100+14634
What this shows

Before AI, review was half of how I worked with other engineers, and it grew with my output: the more I shipped, the more I reviewed. At Tano, an AI reviewer reads every PR and engineers own their areas end to end, so review stopped being a person's job. My output rose tenfold while the reviewing I did for others fell by three quarters.

Method and caveats