lockstep
4 min readNaman Jain

DORA metrics: the five measures, and what AI coding changes

The five DORA metrics explained: change lead time, deployment frequency, recovery time, change fail rate, and rework rate, plus AI coding effects.

A team adopts AI coding agents and pull requests per engineer double. The VP calls it a success. Two months later, on-call is miserable: more rollbacks, more hotfixes, more unplanned deploys. The team got faster. It didn't get better.

DORA's metrics are built to catch exactly that. This guide covers the five metrics from DORA's metrics guide, checked on 16 November 2026, and how AI coding changes what to watch.

The five metrics

DORA groups its software delivery metrics into throughput and instability.

Throughput: how many changes move through the system.

MetricDefinition
Change lead timeTime from commit to version control to deployed in production
Deployment frequencyNumber of deployments in a period, or time between them
Failed deployment recovery timeTime to recover from a deployment that fails and needs immediate intervention

Instability: how well deployments go.

MetricDefinition
Change fail rateShare of deployments that need immediate intervention, such as a rollback or hotfix
Deployment rework rateShare of deployments that are unplanned and happen because of a production incident

The metrics evolved from DORA's original four keys; failed deployment recovery time replaced MTTR, and rework rate was added.

What DORA's research says

  • Speed and stability aren't trade-offs. The metrics correlate for most teams. Top performers do well on all five; low performers do poorly on all five.
  • They predict more than delivery. DORA says the metrics predict better organizational performance and well-being for team members.
  • Context matters. They're best applied to one application or service at a time. Blending them across teams with different contexts can mislead.

How to measure them

You can get most of this from tools you have:

  • Change lead time: commit timestamp to production deploy timestamp, from git and your deploy pipeline.
  • Deployment frequency: count deploys per service from the pipeline.
  • Change fail rate and rework rate: tag deploys that are rollbacks, hotfixes, or incident-driven; divide by total deploys.
  • Recovery time: from incident or failed-deploy start to recovery, from your incident tool.

Start with a monthly view per service. Trends matter more than exact numbers.

Pitfalls DORA warns about

  • Setting metrics as goals. "Every app deploys multiple times a day by year end" invites gaming, per Goodhart's law.
  • One metric to rule them all. Use several, with some tension between them.
  • Using industry as a shield. Regulation isn't a reason not to improve.
  • Disparate comparisons. A mainframe system and a mobile app shouldn't share a leaderboard.

What AI coding changes

AI coding agents mostly push on throughput: more changes, written faster. DORA's framing tells you where to look for the cost.

  • Watch instability alongside throughput. If deployment frequency rises and change fail rate rises with it, the agents are producing changes the system can't absorb.
  • Watch rework rate. Unplanned, incident-driven deploys are where reckless technical debt from AI coding shows up.
  • Lead time can hide review load. Agents shorten writing time; if pull requests then wait days for review, lead time doesn't move. See AI code review best practices.
  • Don't measure output. Lines of code or PR counts from agents aren't delivery performance. DORA's metrics measure outcomes.

Where decisions show up

Many change failures aren't bad code; they're good code built against the wrong assumption: an API format the team had agreed to keep, a migration rule from the last incident. That's a context problem, and it shows up in change fail rate.

Where Lockstep fits

Lockstep briefs Claude Code sessions on the team's approved decisions that apply to the work, captured from PRDs, Slack, and pull requests, so agent-written changes start from what the team agreed. Its checks are advisory: they flag likely conflicts for a person to review.

See how it works.

Sources checked 16 November 2026

For incident follow-up, see the blameless postmortem template. For goals that use these metrics, see the OKR template. For safer releases, see feature flag best practices. For a branching model that improves these metrics, see trunk-based development.

Frequently asked questions

What are the DORA metrics?
DORA's software delivery performance metrics. There are now five: change lead time, deployment frequency, and failed deployment recovery time measure throughput; change fail rate and deployment rework rate measure instability. They evolved from the original four keys.
What happened to MTTR in DORA?
DORA replaced mean time to restore with failed deployment recovery time: the time it takes to recover from a deployment that fails and needs immediate intervention. DORA's site has a history of the metrics explaining the change.
Are speed and stability a trade-off?
DORA's research says no: the metrics are correlated for most teams. Top performers do well on all five, and low performers do poorly across them.
How should DORA metrics be used?
Per application or service, over time, to find what to improve. DORA warns against setting them as goals, relying on one metric, comparing very different teams, and using industry rules as an excuse not to improve.
How does AI coding affect DORA metrics?
AI coding agents can raise throughput by producing changes faster. Whether that improves delivery shows up in the instability metrics: if change fail rate and rework rate rise with throughput, the extra speed is creating problems downstream. Watch all five together.

Keep reading