Engineering Performance · part of Developer Intelligence

Engineering performance,
measured from code to production.

AI-native engineering metrics for velocity, code quality, and production reliability. Measure Engineering Output from merged work, adjusted for complexity, change size, missing tests, and AI slop.

Try Now Get Pricing

GitHubGitLabCIDeploysPagerDutyClaude CodeCodexCursor

What shipped

Example · Payments · Last 12 weeks

Engineering Output per Engineer

A snapshot of how the team is shipping. Switch metrics and cuts to explore the detail.

67.55 / week +110.26%

Total OutputAI vs HumanBy CategoryBy People

105

140

P90

P50

Jun 15Jun 29Jul 13Jul 27Aug 10Aug 24

Dashed lines: illustrative reference valuesTeam

Output / week / eng

By Category

Features39.89%

KTLO35.31%

Bug Fixes24.8%

Unclassified0%

Engineering Output per Engineer is Engineering Output per engineer per week: merged work weighted by complexity and adjusted for change size, AI slop, and missing tests. Benchmarks come from the illustrative data shown here.

Trusted by AI-forward enterprises

Velocity, adjusted

A merged PR is not a unit of work.

Engineering Output weights every merged change by its complexity and takes output back off for quality: AI slop and missing tests earn less. Ten trivial PRs and one hard migration stop looking the same.

Complexity from the diff · logic, scope, architecture, risk, and novelty, assessed in the repository’s context

Quality discounts · AI slop and missing required tests; later rework is tracked separately

Four ways to explore Output · Total Output, AI vs Human, By Category and By People, with clearly labeled illustrative reference lines

Read how Engineering Output is scored

How a PR becomes output

Example · four merged changes

11.90

Engineering Output points

eligible merged changes

4 factors

one scoring model

Merged changeComplexity weightSlop / test factorsOutput pts

#4821 Retry queue, idempotent writes100 changed lines · size factor 0.66712slop ×0.85 · tests ×16.80

#4790 Checkout p95, pricing cache150 changed lines · size factor 0.756slop ×1 · tests ×14.50

#4833 Bulk config regeneration (agent)450 changed lines · size factor 0.92slop ×0.3 · tests ×0.80.43

#4802 Typo fixes across docs10 changed lines · size factor 0.1671slop ×1 · tests not required0.17

CutsTotal OutputAI vs HumanBy CategoryBy People

Engineering Output combines complexity weights, a bounded change-size factor, and penalties for AI slop and missing tests. Rework is a separate outcome signal. The same scoring feeds Engineering Output per Engineer above.

Quality, reliability, CI/CD health

Speed only counts if it holds in production.

The same telemetry reads quality and reliability: code rewritten within 30 days, test coverage on what changed, the AI Slop Index, change failure rate, incident recovery, and the health of the pipeline that ships it all.

Rework, not review comments · Code Durability is the share of merged code still standing after 30 days, by team and by author type

Coverage on what changed · tests added against lines touched, per PR, so a green suite cannot hide untested new code

Pipeline health · success rate, median CI time and deploy frequency, so a slow pipeline shows up as the bottleneck it is

See AI Impact

Quality, reliability and pipeline

Example · Payments · Last 12 weeks · AI-assisted vs human-only

Quality

Code Durability

93.2%

−0.9 pts vs human-only

30-day rework 6.8% vs 5.9%

Test coverage on changed code

71%

+4 pts

tests added against lines touched

AI Slop Index

−6, lower is better

on AI-generated code, 12-week trend

Reliability

Change failure rate

3.2%

−0.7 pts

deploys causing a failure or incident, baseline 3.9%

Defect rate

2.1%

−0.3 pts

bugs per merged PR within 30 days, baseline 2.4%

Time to recover

42 min

−11 min

median, incidents from this quarter's deploys

CI/CD health

Pipeline success

94%

+3 pts

runs passing on first attempt

Median CI run

9m 40s

−2m 10s

commit to green, main branch

Deployment frequency

38 / wk

to production, prior 12 weeks 29

Computed from merged PRs, CI runs, deploys and incidents. AI-assisted means a merged PR with positive AI attribution or an agent author; the baseline is the same team's human-only work on the same repositories.

Team pulse

A weekly view of delivery and quality.

Review what merged and deployed, what stalled or was reverted, and where your team can improve. Team pulse brings delivery evidence into one weekly view, with links back to the work behind it.

Shipped · up to three highlights written from the week's merged PR clusters, each linking to its PRs, tagged Feature, KTLO or Bug fix

Did not ship · open PRs with no commit or review in five days, PRs waiting more than 48 hours for a first review, and PRs reverted or rolled back, each with its age

Areas of improvement · each line names the evidence in this team's own PRs and sessions, the change that would move it, and the hours at stake

Ask AI about any week

Team pulse

Example · Payments · Week of Aug 24 to 30

PRs merged, 14 AI-assisted

deployed to production

did not ship: 4 stalled, 2 blocked on review, 1 reverted

incidents from this week's deploys

Shipped3 highlights

1Retry queue moved to idempotent writes; duplicate charges on retry fell to zero in staging.Feature#4821#4833AI-assisted

2Checkout p95 down 210 ms after the pricing cache landed.Bug fix#4790Agent-authored

3PCI evidence export automated ahead of the Q3 audit.KTLO#4802#4815Human

Did not ship7 PRs, 4 shown

Ledger reconciliation v2Stalled 9 days#4766 · no commit or review since Aug 21

Refund webhook retriesBlocked on review 52h#4830 · one reviewer requested, none started

Fee schedule migrationReverted Aug 27#4811 · rolled back after a change-failure alert

Card vault key rotationBlocked on review 61h#4819 · same two reviewers requested

Areas of improvementranked by hours returned

1Review latency19 PRs in 12 weeks waited 48h+ for a first review, 17 of them on the same two reviewers.Add a third reviewer to the payments rotation.14hper week

2Verification before close34% of AI sessions closed without a test run; those PRs needed 2.7× more follow-up fixes.Make a passing test run the exit condition for agent sessions.9hper week

3Diff size4 PRs over 500 lines took 2.8× the median cycle. Naming the files in scope up front yields 2.4× smaller diffs.Put the file list in the first prompt.6hper week

Stalled: an open PR with no commit or review for five days. Blocked on review: review requested, none started within 48 hours. Reverted: a revert commit or rollback deploy that names the PR. Improvements are ranked by estimated hours returned to this team's cycle time each week, from its own last 12 weeks.

A closer look

Engineering metrics with clear definitions

Engineering Output is the primary delivery metric. Read it alongside cycle time, code quality, and production reliability to understand what shipped and whether it held up.

Engineering Performance metric reference
Metric	What it measures	Unit	Source and window
Engineering Output	Complexity-weighted merged work, adjusted for size, missing tests, and AI slop	Output points	Analyzed merged PRs; complete Monday–Sunday weeks
Engineering Output per Engineer	Weekly output divided by engineers with positive output that week; averaged across complete weeks	Points / active engineer / week	Same eligible work and complete weeks
AI Output Share	AI-assisted plus agent-authored output divided by total output	Percentage	Attributed Engineering Output; same complete weeks
PR cycle time	Time from first commit to merge	Hours or days	Commit and merge timestamps; selected period
30-Day Code Rework Rate	Recently added code rewritten or deleted within 30 days	Percentage	Code history; a 30-day observation window
AI Slop Index	Assessment of low-quality or unnecessary AI-generated code	Scored quality signal	Analyzed changes; selected scope
Deployment frequency	How often software is deployed to an environment	Deployments / period	Deployment records and environment mapping
Change failure rate	Deployments associated with a failure or incident	Percentage	Mapped deployments and incidents; selected period
Incident recovery	Time taken to restore service after a failure	Minutes or hours	Incident and delivery records; source-dependent

Questions about Engineering Performance

How does Engineering Output differ from counting pull requests?

Each eligible merged change earns points based on complexity, a bounded size adjustment, and penalties for missing tests and AI slop. A routine update and a difficult architectural change contribute differently. PR counts remain useful supporting evidence.

Does rework reduce Engineering Output?

The current score applies missing-test and AI-slop penalties at merge time. Later rework, code turnover, and incidents are separate outcome signals; they do not directly discount the current Engineering Output formula.

How does Engineering Performance relate to DORA metrics?

Engineering Output describes scored work delivered. DORA views describe delivery speed and stability using deployments, lead time, failures, and recovery. Use them together. CI/CD and DORA views require the relevant data mappings and admin access.

How should I interpret the product examples?

The product views on this page use illustrative data to explain the metrics and workflows. For metric definitions, sample calculations, and assumptions, see our measurement methodology.

Connect your repositories. Review your team's delivery.

Build a baseline from repository history and connect agent sessions to understand AI contribution. Start with one team's repositories, or get pricing for the org.

Get Pricing Try Now

Engineering performance, measured from code to production.