AI coding tools have increased how much work each developer can take on. Code is quicker to write and pull requests are easier to open. But understanding AI’s impact on engineering requires looking across the development process, including how quickly changes move through delivery and how reliably teams release them.
DORA metrics help teams examine the speed and stability of software delivery as part of that broader picture. Antenna now measures all four automatically from the tools your teams already use. There are no surveys or spreadsheets to maintain, every team is measured the same way, and your delivery history is available as soon as you connect your development tools.
DORA metrics in the AI era
DORA, short for DevOps Research and Assessment, is a research program at Google Cloud that has studied software delivery performance for more than a decade. Its four key metrics describe both the speed and the stability of delivery. Deployment Frequency and Lead Time for Changes measure throughput, tracking how often a team ships and how quickly each change moves through its process. Change Failure Rate and Time to Restore Service measure stability, tracking how often deployments fail and how long recovery takes.
The metrics are most informative when read together. More frequent deployments represent progress only if failures remain stable. DORA's research has repeatedly found that the highest-performing teams achieve both speed and stability.
That balance has become harder to judge as teams adopt AI. The 2025 DORA report found that higher AI adoption was associated with greater individual effectiveness and a small increase in delivery throughput. It was also associated with greater delivery instability.
Delivery throughput and stability diverge with AI adoption
Estimated association between AI adoption and key outcomes
- Individual effectiveness
- Approximate standardized estimate 0.172, with an 89% credible interval from 0.146 to 0.198.
- Delivery throughput
- Approximate standardized estimate 0.029, with an 89% credible interval from -0.001 to 0.059.
- Delivery instability
- Approximate standardized estimate 0.097, with an 89% credible interval from 0.067 to 0.129.
DORA Research, State of AI-Assisted Software Development (2025)
The report describes AI as an amplifier: it magnifies the strengths of teams with sound engineering practices and the weaknesses of those without them. Faster code generation creates little value if changes then stall in review or cause more failures in production.
DORA metrics complement measures of AI adoption. Usage data shows whether developers are using AI tools. DORA metrics show whether the software reaching customers is becoming faster and more reliable as a result. Tracked over time, they give each team a baseline for assessing how AI is affecting delivery and an early warning when speed comes at the expense of stability.
Consistent measurement across teams
Many organizations still assemble DORA metrics by hand from surveys, spreadsheets, or scripts maintained by individual teams. These numbers are difficult to keep current. They are also difficult to compare because each team defines metrics in its own way.
Antenna measures the four keys from your connected delivery data and applies the same definitions to every team:
- Deployment Frequency measures successful production deployments per day.
- Lead Time for Changes measures the average time from a pull request's first commit to its merge.
- Change Failure Rate measures the share of successful production deployments that later caused a failure.
- Time to Restore Service measures how long recovery takes after those failures.
Antenna rates each metric as Elite, High, Medium, or Low using fixed benchmark bands. Weekly, monthly, and quarterly trends show how each team's performance changes over time.
Few organizations perform strongly on any single DORA metric, based on data in the 2025 survey. Most teams have room to improve, whether in how often they ship or how quickly they recover from incidents.
How common is strong software delivery performance?
Share of DORA survey respondents meeting each selected reference point.
- Deployment frequency
- Multiple deployments per day: 16.2% of respondents.
- Change lead time
- Under one day, from commit to production: 24.4% of respondents.
- Change failure rate
- 4% or less: 16.7% of respondents.
- Failed deployment recovery time
- Under one hour: 21.3% of respondents.
DORA Research, State of AI-Assisted Software Development (2025)
Benchmarks show where a team stands, but a more meaningful comparison is often with its own history. Teams that run different services work under different release requirements and constraints. A shared set of measures lets managers discuss those differences on common terms and see whether changes to the delivery process are having an impact.
Trace each trend to the work behind it
Each metric links to the deployments and pull requests that produced it. You can review them for the full date range or for a single period. Team filters and custom date ranges keep the analysis focused on a particular group's work.
For example, when deployment frequency and change failure rate rise together, you can review the deployments that failed and the pull requests they shipped. Each fix can be traced back to the deployment it repaired. Patterns across those failures show where better testing or changes to the release process would have the most effect.
When lead time grows, you can review the lead time of each merged pull request and investigate the slowest changes. A team might respond by shipping smaller changes or prioritizing reviews. Following all four metrics over the next few periods can then show whether speed improved without a loss of stability.
Understand failures and recovery
By default, Antenna estimates change failures from Git history. It recognizes reverts, hotfixes, and rollbacks and traces each one to the production deployment it repaired. Recovery time runs from that deployment until the fix reaches production.
Teams using PagerDuty can measure change failure rate and recovery from resolved high-urgency production incidents. Antenna associates each incident with a recent successful deployment and measures recovery from when the incident opens until it resolves. You can review the deployment alongside the incident and open the incident in PagerDuty for more context.
Together, delivery records and incident data give teams evidence for reviewing what failed, how they responded, and where changes to their process could help.
Getting started
Setup takes a few minutes. Connect GitHub, GitLab, Bitbucket, or Azure DevOps, and Antenna backfills your history so trends for all four metrics are available right away.
Antenna detects how each repository deploys automatically; review the deployment configuration to confirm the production signals match how your teams release software. If you use PagerDuty, connect it to Antenna to measure failures and recovery from production incidents.
Learn more about DORA metrics or request a demo to learn how to measure delivery performance with your team.