Why we're publishing our DORA metrics

For years we've said our engineering team is one of the best in the business. Now the numbers are public, updated daily, with a live log of what we've learned from coding with AI.

Chris Cooney3 min read

Why we publish our DORA metrics: Coralogix feature releases per month from 2022, flat until late 2025, then climbing to 130 releases

For years we've told customers, analysts and anyone who would listen that the Coralogix engineering team is something special. Saying it is easy, so we stopped saying it and started publishing it. Our Engineering Benchmarks put our live DORA metrics, our release history and what we've learned from coding with AI in public, and they update every day.

Why DORA metrics

Every software company says it ships fast. DORA metrics are how you check: the industry's shared measure of how often a team deploys, how long a change takes to reach production, how often a change fails, how quickly service comes back when one does, and how much work gets redone. We chose them because they are hard to flatter and easy to compare.

Over the 30 days to 16 November 2026, Coralogix deployed to production 125 times a day, with a 32-minute lead time for changes, a 0.1% change failure rate and an 18-minute time to restore.

The DORA panel on the Engineering Benchmarks page for 18 October to 16 November 2026: deployment frequency 125.2 a day, lead time for changes 32 minutes, change failure rate 0.1%, time to restore 18 minutes, rework rate 0.48%
The DORA panel on the benchmark, for 18 October to 16 November 2026. It updates daily.

From 4 feature releases a month to 130

From May 2022 to August 2025, Coralogix shipped about four feature releases a month. AI-assisted delivery began in September 2025, and in October 2026 we shipped 130, 33 times the old rate.

Bar chart of Coralogix feature releases per month from May 2022: about four a month until AI-assisted delivery began in September 2025, then rising to 130 in October 2026
Feature releases per month since May 2022.

Shipping faster didn't cost us stability. Between January and October 2026, production releases per day rose from 48 to 119, developer pull requests from 4,116 to 10,149 and features shipped from 19 to 58. Over the same months, new bugs fell from 337 to 288, Statuspage incidents from 7 to 3, and median lead time from 44 minutes to 36.

Month-by-month grid for January to October 2026: agent share of pull requests rising from 6% to 49%, with pull requests, releases per day, features shipped, median lead time, new bugs and Statuspage incidents, each month shaded from its weakest in grey to its strongest in green and its best month in gold
January to October 2026. Each row runs from its weakest month in grey to its strongest in green, with its best month in gold.

What AI coding taught us, in a live log

Hundreds of people at Coralogix use Claude Code, Cursor, Codex and Copilot every day, and their telemetry lands in Coralogix, so we can measure what every change to our process does. Coding agents now make 49% of our pull requests, up from 6% in January.

Along the way we found new ways of working that we think other engineering teams can use, so we publish them. The benchmark includes a learnings log that we add to every time we find an approach with a meaningful effect on AI coding. Each entry records the problem, the signal that exposed it, what we changed and what it saved:

  • Model routing. Haiku 5.5 became the default for maintenance subagents, with plan mode on Opus and execution on Sonnet. A maintenance session went from $0.92 to $0.41, with the merge rate unchanged.
  • Skills. Our pr-description skill read only the branch diff and the linked ticket instead of the full repository history. Each call went from $4.10 to $0.55, across 2,300 calls a month.
  • Governance. Every service account got an owning team, a budget alert and a cap on retries. The share of spend from unattended automation fell from 18% to 9%.
The first three entries of the learnings log, each with its problem, signal, solution and savings: governance cut unattended automation spend by 50%, model routing cut the cost per maintenance session by 55%, and a leaner skill cut the cost per call by 87%
The latest entries in the learnings log.

AI cost per merged pull request is down 44% since the first month, to $11.80.

Explore the Engineering Benchmarks

The Engineering Benchmarks update daily. Check the DORA panel, see how pull requests, lead time and incidents moved as agent adoption grew, read the full learnings log, and come back next month to see what we've added.

If you want the same view of your own engineering team and coding agents, start a free trial or talk to an expert.

Coralogix

One lake.
Every signal.

Logs, metrics, traces and security data in one queryable place, ready for your agents and your people.