How much faster everyone got with agents.
How much engineering each engineer ships, against their own history from before the agents.
Engineering shipped per engineerTrailing 12 mo
Your baseline
2.3× more shipped per engineer.
The AI engineering benchmark
AI is how we code now.
AI is how we code now.
See where you stand.
Every merged change, scored. See how far AI has taken your team.
3×the engineering your team ships, with GitVelocity
Nothing touches your code · bring your own model key · free
Engineering teams on GitVelocity
Agents are the new fuel.Engineers are still the drivers.
Engineering, not token spend
Engineering finally has a speedometer.
And it isn’t your token bill.
Every merged change is scored for how much engineering it took. Months of those scores make an engineer’s multiplier.
Config key renamePull request·34 files23/100
Scope16
Architecture1
Implementation2
Risk3
Quality1
Perf & Security0
Concurrent batch processorPull request·11 files68/100
Scope15
Architecture12
Implementation17
Risk10
Quality11
Perf & Security3
Tap the card behind to compare
What it scores
We measure scope by how far a change reaches, such as the files, systems, and boundaries it touches.
Why it countsBig changes count, but size is only one of six parts of the score.
What it scores
We measure architecture by what a change does to the system’s design, such as new abstractions, services, and dependencies.
Why it countsMost changes don’t touch the design, so they score low here. Real redesign work gets credit.
What it scores
We measure implementation by how hard the logic is, such as rules, algorithms, and concurrency.
Why it countsA short change can still be hard work. This gives it credit even when the diff is small.
What it scores
We measure risk by what it takes to ship a change safely, such as migrations, breaking changes, and rollbacks.
Why it countsShipping a risky change without breaking anything is hard work that a commit count never shows.
What it scores
We measure quality by how well a change is built, such as tests, docs, and clear code.
Why it countsAgents make code fast to write. Tests and docs make it safe to keep.
What it scores
We measure performance and security by extra work beyond the defaults, such as speedups and hardening.
Why it countsIt’s a small bonus. Most changes score 0 or 1 here, and that’s normal.
Founders and CTOs
How they’re making the whole team faster.
“100% of the code we write is AI-first, and almost every feature we ship has an agentic element. GitVelocity provided insight and visibility in two ways that helped us 10x our velocity: detailed and actionable feedback on every PR, and rubric scoring to evaluate each engineer’s productivity. We are a competitive group, so seeing the benchmark has been motivating the team and the individuals.”
Nathaniel Jewell
Founder & CEO
10×per engineer
“We changed how the whole company builds this year, including people who are not engineers. What I did not expect was the week we turned GitVelocity on for everyone. Nobody was told to work differently and the work changed anyway.”
Chopin Yen
CTO
6.5×per engineer
“I could always tell you whether engineering felt busy. I could never tell you whether it was getting better. That is the gap GitVelocity closed for me.”
Peter Thompson
Co-Founder & CEO
1.8×per engineer
“Measuring developer productivity is a huge challenge, and GitVelocity is a major step in the right direction. We use it to measure how much our team is accelerating over time and communicate it to the whole company.”
Bruno Dreux
CTO
2.9×per engineer
“Our developers asked for a score on every push, so they could track quality while they iterate rather than waiting for the final number. It’s part of how they work now.”
Eduardo Maia Machado
CTO
3.1×per engineer
“Before GitVelocity, I was going off gut feel. Now I can see how the team is trending over time with a real number on our engineering output, and how we stack up against other engineering teams. That’s a big upgrade for any CEO.”
Colton Calandrella
Co-Founder & CEO
2.6×per engineer
Velocity Map
Where would your engineers land?
A month of PRs at DuckDB, Tailscale, rust-lang and more, scored by GitVelocity.
See your multiplier →Legend
- Height
- points earned in 30 days
- Contour
- every 500 pts, steeper near the top
- Ring
- a PR's six scoring dimensions
- Logo
- a team: every PR merged into that org
How much faster everyone got with agents.
How much engineering each engineer ships, against their own history from before the agents.
The gain is not evenly spread.
Rank how well each engineer directs their agents, against their own past.
Engineering, not volume.
A rename and a race-condition fix are not the same work. The score reads what a change does, not how many lines it has.
Your engineers, in the field.
See your team in the context of every engineer out there.
The gain is not evenly spread.
Rank how well each engineer directs their agents, against their own past.
LeverageAgainst each baseline
Engineering, not volume.
A rename and a race-condition fix are not the same work. The score reads what a change does, not how many lines it has.
Scored change
New feature PR
Concurrent batch processor with retry and backoff
68
/ 100
Scope
15/20
Architecture
12/20
Implementation
17/20
Risk
10/20
Quality
11/15
Perf & Security
3/5
What counts
PR count and line count are terrible measures. The GitVelocity score reads the change itself.
Your engineers, in the field.
See your team in the context of every engineer out there.
Your engineers in contextThis quarter
Grey dots are the wider benchmark. Green dots are your engineers.
Also in your terminal
Score where you work.
The scores and trends in your dashboard are also in your shell. The GitVelocity CLI pipes them into Claude, or into whatever you already run. Point your agent at the MCP server, or build on the public API.
The benchmark
See where you stand among 4,000 engineers.
Your 2.3× multiplier ranks in the top 12%.
YouEveryone elseTop 0%
Check our working
See what the number rests on.
No hidden weights. The six dimensions, the size multiplier tiers and our model benchmark are all public.
Scale
0k+
Merged changes scored. Every one read by a frontier model, not counted by a heuristic.
From seed-stage teams to public companies
Calibration
4,000+
Engineers in the benchmark. We score what the best of them merge.
Including some of the most-watched open-source repositories
Coverage
0+
Repositories connected, in whatever your team already ships. Nothing in your codebase, no config to maintain.
Every language: Python · TypeScript · Java · Go · Rust
Free
$0
You supply the API key, so the model calls run on your account and cost you pennies per change. We charge nothing.
Anthropic by default · OpenAI and others supported
Why the multiplier
Benchmark your workflow, not your agents.
| Approach | What it measures | What you get |
|---|---|---|
| GitVelocity | How much engineering ships, over time | Your AI multiplier — org and per engineer, human or agent |
| Story points | Guesses before work starts | Always wrong, easily gamed |
| DORA / process metrics | How work moves through the pipeline | Measures process, not engineering. Blind to whether your team ships more of it. |
| PR count / lines of code | How much stuff moved | Picks the wrong teammate 1 time in 5, and an agent inflates it in seconds. |
| AI coding & eval tools | How much AI produces, or agent quality | None measure whether your org actually ships more engineering |
Your keys
Bring your own key
Add an Anthropic or OpenAI key. We call the provider with your key and they bill you directly, usually a few cents per change.
Your code
We never store source
We send the diff to the model to score it, then delete it. What stays in our database is the score, the six sub-scores and a short written rationale.
Not surveillance
Work, not workers
GitVelocity scores merged work the team can already see. No keystrokes, no hours, no screens — every engineer you invite sees their own scores and the rubric.
Your data
Your data stays yours
We do not sell or share your data. Built to European standards, GDPR and DORA compliant, with annual Tier 1 penetration testing.
See it on your own repos
See how much faster AI actually made you.
Connect your repos and see how you’ve done.
Bring your own model key