Shipping at AI speed.
Senior engineers directing AI coding agents, measured the plain way: pull requests merged each month, next to what one developer merges in LinearB's benchmarks.
The numbers
Pull requests merged on any branch across the repositories our team builds and maintains. Counted from GitHub, America/New_York (ET) months. Updated weekly; the current month runs to the time of the last update, Oct 30, 2026 (ET).
Our month vs one developer log scale
Twelve-month trend
- 1Nov 2025
- 37Dec 2025
- 23Jan 2026
- 42Feb 2026
- 34Mar 2026
- 39Apr 2026
- 294May 2026
- 332Jun 2026
- 371Jul 2026
- 492Aug 2026
- 692Sep 2026
- 964Oct 2026, through Oct 30, 2026 (ET)
The last bar is the current month to date.
Developers to match
964 merged PRs in Oct 2026 through Oct 30, 2026 (ET), divided by each band's midpoint and rounded up. Elite has no upper bound, so its rows divide by the band floor and read as at most.
| Benchmark | PRs per developer per month | Divided by | Developers to match | Monthly cost |
|---|---|---|---|---|
| Elite + 26% Copilot upliftIllustration, not published | 11.0 (Elite x 1.26) | 11.0 illustration (Elite floor x 1.26) | at most 88 | at most $2.3M/mo$25,577/dev/mo |
| EliteMore than 2 PRs a week | > 8.7 | 8.7 (floor) | at most 111 | at most $2.8M/mo$25,558/dev/mo |
| Good1.2 to 2 PRs a week | 5.2–8.7 | 6.95 (midpoint) | 139 | about $2.8M/mo$20,476/dev/mo |
| Fair0.66 to 1.2 PRs a week | 2.9–5.2 | 4.05 (midpoint) | 239 | about $3.9M/mo$16,190/dev/mo |
How the monthly cost is calculated
- Band-to-pay mapping is our assumption: Fair = median, Good = 75th percentile, Elite = 90th percentile.
- Wages: BLS OEWS, software developers, May 2025: $135,980 median / $171,980 75th / $214,670 90th percentile (BLS OEWS).
- Loaded multiplier = 1 / wage share of total compensation, BLS ECEC private industry, June 2026: $46.89 total compensation / $32.82 wages = about 1.43x (wages are 70.0% of compensation) (BLS ECEC table 4).
- Plus $19/seat per month for GitHub Copilot Business on the Copilot row (GitHub Copilot plans).
- Office overhead is excluded, so totals run low.
- Rows are alternatives, not additive.
LinearB publishes weekly bands; monthly figures are weekly x 4.33. Method: bounded bands divide by the band midpoint (Good (5.2 + 8.7) / 2 = 6.95, Fair (2.9 + 5.2) / 2 = 4.05), so the count is a typical developer, not the slowest one in the band. Elite is open-ended (more than 8.7), so dividing by its floor gives the most developers it could take: at most. The Copilot uplift row applies a 26% gain from field experiments to the elite band as an illustration; it is not a published benchmark.
Cost estimates use U.S. Bureau of Labor Statistics wages for software developers, May 2025 (Fair = median, Good = 75th percentile, Elite = 90th percentile), loaded about 1.43x for benefits and payroll taxes (BLS ECEC, June 2026), plus $19/seat for GitHub Copilot Business. Performance bands are LinearB throughput benchmarks; the pay mapping is our assumption. Rows are alternatives, not additive. Office overhead excluded.
How we work
Senior engineers direct AI coding agents end to end.
The engineer owns the plan, the architecture and the call on what ships. Agents do the typing, the tests and the first pass at review.
Every product change goes through a pull request and review.
Checks run and an engineer reviews before it merges. That is why merged pull requests are the number we count.
A fair reading
A PR is not a fixed unit of work. Ours are small and focused: one change, one review. LinearB reports that AI-generated PRs run larger and merge less often, so a like-for-like count would not flatter us the way a raw ratio might suggest. Read these figures as order-of-magnitude throughput, not a precise headcount. The current month is partial, through Oct 30, 2026 (ET). The count covers every human-authored pull request merged on any branch, including any by client engineers working in the same repositories. A pull request merged into a feature branch counts, and so does the one that later merges that branch. Bot PRs (any GitHub App or bot account, such as Dependabot, Renovate and GitHub Actions) are excluded.
Sources
- LinearBEngineering benchmarks 2026
- Management ScienceThe Effects of Generative AI on High-Skilled Work: three field experiments
- InfoQStudy shows AI coding assistant improves developer productivity
- U.S. Bureau of Labor StatisticsOccupational Employment and Wages: Software Developers (15-1252)
- U.S. Bureau of Labor StatisticsEmployer Costs for Employee Compensation, Table 4
- GitHubPlans for GitHub Copilot
Stats generated Oct 30, 2026 (ET). Want this pace on your project? Talk to us.