Commit count is far from a perfect barometer of dev progress. It earned its place onto my 2020 list of "five worst software metrics agitating developers." Commit count is trivial to game. Unless you work in an organization that's dogmatic about its commit definition, every developer will have their own interpretation of "how much work goes into a single commit."

You could do worse. Compared to "lines of code," the signal/noise of "commit count" is at a clear advantage:

open_in_new

From a 2021 GitClear research study of 2,729 issue tracker tickets

Even more important than its correlation advantage, "commit count" is much less prone to short-term spiking. Whereas "lines of code" can sometimes vaguely resemble "repo evolution" if you stand back and squint your eyes, "commit count" is a comparatively stable day-to-day artifact of developer activity underway.

For these reasons, we wanted to revisit a popular piece we posted a few years ago, git stats on dev activity, based on 295,000 of dev-years. This year, we have accumulated 878,592 developer-years of commit activity to analyze (<-- we'll use that coloring elsewhere to highlight key findings if you hope to skim this 7-minute research in one minute).

The data presented below is, as of its publishing in 2024, the largest aggregation of git commit activity known to exist online (if you know a larger-sourced research paper on this subject, leave a comment? 🙏). In the tables & graphs below, we'll present a picture of what an "average," "good," and "exceptional" year of commit progress looks like.

link📚 878,592 randomly sampled years of GitHub developer activity

Every quarter, GitClear publishes a new original research paper in the space of developer experience, developer happiness, and developer productivity. Since we don't have a professional writer on staff, our research contribution is driven largely by leaning on the data that we collect in the natural course of processing commits for thousands of customers, and hundreds of open source projects.

linkData characteristics

Some metadata about the research that went into collecting the data we're presenting:

Gathered: Between 2020-2024

Source: Publicly visible data on GitHub profiles (e.g., your humble author had ~5k during 2023)

Inclusion criteria: A committer-year was included if and only if the year included between 100 (why?) and 20,000 commits (why?). Note that an average American software engineer works around 241 days per year.

Data size (n): 878,592 developer-years

Distribution: Global, all GitHub users who have not disabled profile stats

link🧮 Percentiles of "Commits Per Year" Relative to Developer Cohort

Recall that the bounds of this data is from 100 to 20,000 commits per year. We surmise, without any proof but with informed conjecture, that 80% or more of those who contribute 100+ commits per year work in a job related to software development. So it seems fair to label these stats as "Software Developer Commit Count Percentiles." So, let's do it:

linkAll Programmers (incl Open Source) Commit Count Percentiles

As produced by this console query:

open_in_new

Number of commits made annually among those who made at least 100 commits per year

To the extent that you believe that full-time developers average sometimes go a year with one commit every 2.4 days, the median developer between 2020-2024 contributed 417 commits per year. Thus, the median software developer contributes 1.73 commits per working day

linkCareer Developer Commit Count Percentiles

The lower bound of "100" figures to include some percent of developers that work on code sparingly. Maybe they maintain a small open source project in their spare time, or they are an Upwork contractor who alternates between phases of "being engaged in tasks" and awaiting the next gig. Another way to look at the data is to update "Inclusion Criteria" to require at least 241 commits (one per working day) up to 10,000 commits (to better exclude accounts with automated commit activity) per year.

This more narrow band of eligible developer-years reduces the data set to 599,872 developer-years. To get a sense for the size of this data set, gpt-4o's best guess is that there are about 180,000 developers at Apple, Amazon, Meta, Microsoft and Google, combined, in 2024. That makes the size of this data set roughly equivalent to 3.3 years of developer stats for the five tech companies combined.

Source query.

open_in_new

The table data above, graphed to visualize how annual commit count varies by percentile

To the extent that you believe Career Developers average at least one commit per working day, the median Career Developer between 2020-2024 contributed 673 commits. Thus, the median Career Developer averages 2.8 commits per day.

link📆 Percentiles of Programmer Days Active per Year

To produce a strata of "programmer days active per year" for the "Career Developer" cohort, we used a basic query to find 578,917 developer-years worth of data to analyze. Here was what our data set reported:

This one is pretty interesting! Let's graph it:

open_in_new

The median developer in the data set has around 70 work days per year without a commit

Among those who average a commit per work day, only 20% of developers actually make a commit every work day. The median Career Developer has about 70 days per year where no commit is made, around 30% of the total work days. This assumes that the average developer will have around 230 work days per year (blending Americans with ~240 work days with Europeans and their ~220 work days annually).

link🌎 Concentration of Senior-Level Developers by Country

Which countries are mentioned in the profile's "Location" field among developer-years with 1,000 commits or more?

A lot of these locations don't have a huge sample of those who divulged their location, but still pretty interesting to see how there is almost a 2x difference between the density of 1k annual developers in Tokyo vs. Shenzen. Also, if, like the author, you were curious where Vilnius would be found on a map:

open_in_new

An extremely dense enclave of high-activity software engineers, but also, beware the tiny sample size (n=63)

link🧑‍💻 Stats for "1,000 Commit" Developer-years (Senior-Equivalent Engineers)

In the full data set we analyzed, 1,000 commit years were fairly rare: only the top 30% of developers who average one commit per work day manage to reach 1,000 commits per year. But, as an employer of US-based software engineers, more than half of the hundred developers that our startup-sized company has employed were able to reach this threshold. This is to say, among professional Seattle-based programmers earning $150-250k annually, 1k commit years are much more common than in the overall developer population. Let's look at where the stats land for this particular demographic.

Our database finds 193,651 developer-years in the data set that accumulated 1,000 up to 15,000 annual commits:

The median developer in the Senior Developer cohort contributes 1,563 commits for 51,130 Diff Delta annually (context for that much Diff Delta: it's enough to build two sites like Noteapps.info). How many days do developers have to be active to reach these heights?

open_in_new

Most developers with 1,000+ commits per year contribute on 200-250 days

For those aspiring to be an L6 or L7 developer at the tier 1 tech organizations, it's reasonable to ballpark this cohort as the people you compete against in the top positions. For those in support of work-life balance, it is encouraging to see that, even in this competitive group of upper-end talent, the median developer is active only on work days. It's not until one reaches the 70th percentile that we encounter the developers who don't let PTO or weekends stop them from engaging in their programmer pursuits. 🤓

link🤖 AI Impact on Commit Count

Finally, let's see if we can find evidence that the proliferation of AI that began in 2022 has changed the volume of commits made by developers. To compare apples-to-apples, we'll consider only developers for which we have data before-and-after the 2022 epoch of Copilot and AI ubiquity. Specifically, this data set includes 60,658 developers who authored at least 100 commits every year between 2020 and 2023. Do they support the narrative that AI is boosting developer productivity?

Here were the database queries used to produce this data. Here are how the numbers look graphed:

open_in_new

So much for the unambiguous takeoff of developer productivity

It's hard to imagine how AI might actually be reducing commit activity. Still, from a correlative standpoint, there is no ambiguity that the 60k developers who have posted at least 100 commits per year have seen a gradual decline in their annual commits over the last four years. Compared to 2020, the median developer authored 9.5% less commits in 2023 than they did in 2020.

Since this finding is contrary to what common sense would predict, we consulted Diff Delta (a higher-correlating metric with "software effort" in our research) to check whether it agrees that the median developer is contributing less in 2023 than 2020. The count of developers with Diff Delta stats for each of the years totaled 339. Their median and average Diff Delta per year were:

GitClear customers have gradually grown their Diff Delta each of the past four years -- more in line with what we would expect to see as powerful AI tools proliferate. It's interesting that, like the larger sample GitHub data, we do not see evidence that developers are more active with their commits than years past. Rather, it seems that, in our smaller sample of developers, each is squeezing greater amounts of code change into each commit than they had in years' past.

link🔮 Takeaways & Predictions

When attempting to understand developer activity based on commit count, much depends on where the lower bound is set. As we see in the analysis above, the commit count of the median developer depends largely on what cohort of developers is being analyzed:

open_in_new

Median commit activity by constituency, analyzing 666,867 dev years

Among the cohort we have labeled "Senior Developers," for locations with at least 100 developers reporting, the highest concentration of Senior Developers were observed in Tokyo (55%), Belgium (43%) and Seoul (43%).

In terms of predictions, this data evaluation suggests that AI assistants are not increasingly the volume of commits made by the average developer. Rather, the data we've analyzed suggests that the "Age of AI" is thus far correlating with an increase in commit change density, but not in commit count.

If further research supports these findings, we would predict that:

What do you make of the data? Feel free to drop us a line hello@gitclear.com if you have other ideas for how we can answer your management questions using our terabytes of developer performance data.