5 takeaways from the State of Software Delivery Q2 Pulse report
Senior Technical Content Marketing Manager
AI is pushing code volume up almost everywhere. Shipping it is still the hard part, and the gap between leaders and everyone else is getting wider.
Today we’re releasing the 2026 State of Software Delivery Q2 Pulse report, a shorter check-in between our annual reports. We analyzed more than 20 million CircleCI workflows from March 2026 to see what’s changed since the comprehensive 2026 State of Software Delivery report we published in Q1. With the pace of change only seeming to accelerate each quarter, our goal is to equip engineering leaders with up-to-date information they can use to make the decisions that will set their team’s direction long into the future.
The headline hasn’t changed: teams are producing more code than they can efficiently ship. What has changed is how fast the leaders are pulling away, and what their operating habits reveal about speed and cost in the age of agentic development.
Here are the five findings that matter most.
1. The velocity gap widened again
In Q1, the top 5% of teams ran workflows on the main branch 8x more frequently than the median team. In Q2, that velocity gap hit 9x.
- Median throughput on main held steady at roughly 1.7 workflows per day.
- Top performers climbed to 15.6 main-branch workflows per day.
Everyone is writing more code. What separates elite teams from the rest is how much of that code actually reaches production.
2. Feature branches are busy. Main branches aren’t.
The split that defined our annual report remains the defining story in Q2:
- Feature-branch throughput grew 7.7% year over year.
- Main-branch throughput stayed entirely flat.
Main-branch success rates improved modestly, from 70.8% in Q1 to 76.7% in Q2. While that is a step in the right direction, it remains well below the mid-80s results of 2023–2024, and far underneath CircleCI’s recommended 90% benchmark.
Validation remains the industry’s biggest bottleneck. Teams continue to generate more code than they can safely integrate and ship.
3. A breakout cohort is leaving the pack behind
Percentiles show the size of the gap, but they don’t show what the fastest teams are actually doing.
We looked closely at 20 real organizations with the highest main-branch throughput in our dataset. Where the typical team runs fewer than two workflows a day, this elite cohort averages roughly 2,165 workflows per day on main, a number that surged 72% over the past year.
These teams span regions, industries, and sizes (from ~20 to ~1,400 active contributors). They are organizations that successfully figured out how to turn higher code volume into shipped software, bypassing the obstacles stalling the rest of the industry.
4. Merge Efficiency Ratio is the new metric to watch
This Pulse report introduces the Merge Efficiency Ratio (MER): the number of feature-branch validation cycles it takes to move a change onto main. Think of it as a direct measure of how much rework happens before code becomes production-ready.
- Median Teams: MER sits at 3.9 (roughly 5 total CI cycles per shipped change when you include the final run on main).
- Top 5% Performers: Operating at an MER of 2.6.
- Top 20 Cohort: Running at an elite 1.3.
The gap isn’t static, either. Over the past year, the Top 20 improved their MER by 21%, dropping from 1.62 to 1.28, while the median team’s MER barely budged.
When teams spend fewer cycles getting work to main, developers spend less time waiting, context-switching, and re-validating changes they thought were done. And where coding agents are doing that work, a lower MER shows up directly on the clock and the bill: fewer validation loops mean faster merges and fewer wasted runs and tokens. Lower MER is a developer experience win and a cost win—not just a pipeline efficiency metric.
5. Inner-loop validation is how teams break the bottleneck and control cost
The same validation bottleneck that limits throughput is changing the economics of software delivery. As AI increases code volume, every slow feedback loop gets compounding hidden costs: more CI runs, more developer friction, and more token spend.
We modeled a 50-developer team shipping ~3,000 changes per month at a median MER. With unoptimized agentic workflows and slow CI feedback, delivery costs can reach roughly $900,000 per year. A significant portion of this stems from “token reload penalties” when autonomous agents sit idle waiting for green lights long enough for context caches to expire.
The fix is straightforward: catch routine failures before code reaches CI. Moving fast checks (linting, unit tests, build steps) into the inner development loop gives agents feedback in seconds, not minutes.
For that same 50-person team, shifting validation earlier can cut total delivery costs by $700,000 or more per year. CircleCI features like Chunk sidecars and our new CLI built for agentic development are designed to give coding agents fast, in-context feedback in the inner loop, catching routine failures in seconds, before they ever reach CI. That way, validation keeps pace with agent-driven volume instead of becoming the bottleneck that slows delivery and runs up the bill.
What can your team do next?
Software delivery has shifted from a code generation problem to a code validation problem. AI makes writing code faster, but legacy push-to-CI workflows weren’t built for machine-driven volume. The result is pipeline gridlock, wasted retries, and rising costs.
The State of Software Delivery Q2 Pulse report breaks down the three operational habits behind elite throughput—high individual output, continuous validation beyond commit-driven CI, and low MER—along with practical starting points for each.
Want to see where your team stands? The Software Delivery Data Explorer has been updated with the same dataset used in the report. Compare your metrics against benchmarks by team size, industry, and region.
