State of Merge Queues 2026 | MergifySkip to content🇫🇷We're hiring a Staff Software Engineer in France→×

" data-astro-cid-6carq7tq>

Copy Logo as SVG

" data-astro-cid-6carq7tq>

Copy Wordmark as SVG

Download Brand AssetsVisit Brand Guidelines

State of Merge Queues · 2026AI writes code that breaks main half as often.<br>What 200,000 merges reveal about the gap between "approved" and "merged". The stretch where a green PR can still break the branch, and almost nobody measures it because almost nobody has the data.<br>By Julien Danjou, Co-founder & CEO of MergifyUpdated July 27, 2026

1 in 7private merges are AI-assisted14.4%, and that is a floor<br>1 in 20private merges would have broken main40+ devs: 1 in 8<br>2.4×AI-assisted PRs break main LESS1.9% vs 4.4%, within-repo<br>200K+merges measured477 teams · 90 days

We pulled every pull request that went through Mergify's merge queue over the last three months: more than 200,000 merges across 477 engineering teams. The question was simple. What actually happens to a PR between the moment it's approved and the moment it lands on main? That stretch is where a green PR can still break the branch and where merges quietly pile up behind each other. Almost nobody measures it, because almost nobody has the data.<br>One result surprised us more than the rest. Pull requests written with AI assistance broke main about half as often as ones written without it (1.9% versus 4.4%). The common fear is that AI floods the queue with subtly broken code. In more than 100,000 merges we found the opposite, and it held up when we controlled for PR size and compared inside the same repositories.<br>The quieter finding matters more. The risk of a green PR breaking main is not constant. It climbs steeply with team size, from roughly one merge in 130 on small teams to one in eight once you pass forty engineers. A merge queue stops being optional somewhere on that curve.

What we found<br>→The broken-main rate scales about 16x with team size: 0.8% at 2–5 engineers, 12.5% at 40+.<br>→Private codebases hit it 4.5x more than open source (5.1% vs 1.1%), because interdependent code collides more often.<br>→AI assistance already shows up in 1 in 7 private merges, and that is a floor, since most tools leave no trace in the PR.<br>→AI-assisted PRs broke main half as often, even after controlling for PR size and repository.

A note on method before the numbers. The rates below are computed on the teams running a real queue (at least ten merges and two human authors in the window), about 160 teams and 153,000 merges, fully anonymized, no customer or repository identifiable. The wider 200,000-merge figure is the full platform over the period; the benchmarks use the filtered set so toy and bot-only repos don't skew them. Full methodology sits at the end.

The gap between green and merged grows with your team<br>When several approved PRs merge around the same time, each one is green on its own, but combined they can still break main. A merge queue catches this by testing every PR against the real state main will have at merge time. When a batch fails and the queue has to pull it apart to find the culprit, that is a break it prevented before it reached anyone.<br>Across the cohort that happened on 5.1% of private merges and 1.1% of open-source ones. The average hides the real story, which is team size:

Broken-main rate by team size2–5 devs

0.77% · 1 in 130

6–15 devs

0.98%

16–40 devs

2.49%

40+ devs

12.5% · 1 in 8

Queue time scales too: sub-minute under 15 devs, ~11 min at 16–40, ~18 min at 40+. All four buckets ≥ 21 organizations.<br>At two to five engineers, a green PR breaks main about once in 130 merges, rare enough to handle by hand. At forty-plus engineers sharing a repo, it is one in eight. That is most of a team's week. The curve bends between the 6–15 and 16–40 buckets, which is exactly where teams usually start feeling merge pain.<br>The mechanism is just math. More engineers means more PRs in flight at once, and more chances that two independent, individually-correct changes touch the same assumption. The failed batches show it directly: on private repos they bundle about six PRs on average, so one bad change can stall five good ones queued behind it.<br>This is the case for a queue in a single chart. Below roughly fifteen engineers, most teams get by without one. Above it, "green PR, broken main" goes from a curiosity to a standing tax, and the cost of catching it by hand grows with every hire.

Case study<br>Cerebras runs the hardware CI behind its chips through a Mergify queue, batching compatible PRs into a single multi-hour test cycle so a broken main never idles the rigs. The batching lever from the section above, in production.<br>Read the case study →

The queue itself is fast<br>None of this means the queue is slow. The median PR spends about seven minutes in it, and ninety percent are through within the hour, even on large teams where you'd expect contention to pile up. Throughput sits around three merges per active day for...