Programming language adoption patterns at Meta | XCancel

Programming language adoption patterns at Meta

David Tolnay

@dtolnay<br>50m

12

1,155

Meta’s code review team provides the company with a collection of database tables for analysis of source control activity. These tables contain a representation of all code changes submitted to Meta repositories by Meta employees, including information about filepaths touched, code review revisions, line count, categorizations such as generated code and third-party code, and AI attribution. From these tables derive progressively more enriched tables that feed many infrastructure teams’ dashboards, projections, planning, and staffing.

Today we present one particularly enlightening set of visualizations juxtaposing long-term year-over-year growth rates across languages.

The methodology involves comparing 90-day windows ending 12 months apart, i.e. separated by a 9-month gap:

Within each window, for each language, we count how many devs committed code in that language.

We divide every count by the total number of devs who committed code in any language in that window, to get a "market share": what fraction of devs committed code in the specific language. This normalizes for overall changes in company size.

Lastly we divide market share by the same language's market share in the 12-month-prior window, producing a year-over-year market share growth ratio. Using a multiple of 12 months for this is important because it cancels out seasonal fluctuation in code output, notably during Northern hemisphere winter holiday season and summer peak vacation season, which is a well-established pattern in this dataset.

Concretely, a data point of +98% for Rust on August 10, 2026 means that from May 12, 2026 through August 10, 2026 (90 day window) the fraction of devs committing code in Rust was 96% higher than the fraction of devs committing code in Rust from May 12, 2025 through August 10, 2025 (90 days, 12 months prior). Data points can theoretically range from -100% to +∞%.

Here is every language of current relevance in Meta's codebase on one graph. The horizontal dotted guides are drawn at +0% and +100%, and the vertical axis is cropped at -50% to +150%.

To help lay some foundational intuition about what's in this graph, let's look at a few languages individually, starting with Rust. Rust is the only language that has outpaced the company's rate of growth for this entire 8-year period. In other words it has never dipped below +0% year-over-year market share growth.

In the opposite situation, here are Haskell and Java, which have each been losing market share continuously for 8 years. With the sunset of the company’s last major Haskell project in 2024, Haskell ended the year at -89% year-over-year growth, because 48 people had committed code in Haskell around the end of 2024, whereas 447 people were committing Haskell at the same window the previous year.

Another interesting case is that of Hack, C++, and Python, which are immune to trends. These are the 3 most used languages at Meta and all three float along at +0%, matching the company's overall growth to within about ±7%. This means any rise or fall in number of devs is almost entirely explained by that same change in company size.

Here is Kotlin, a language that sustained rapid growth, powered in part by programmatic rewriting of Java to Kotlin in mobile apps, followed by rapidly saturating the total addressable market for Kotlin (much like Hack, C++, and Python seem to be saturating their markets in the previous graph).

Here is C#, which grew rapidly in AR/VR projects immediately after the company rebrand from Facebook to Meta in October 2021, but lost significant market share later on.

With the intuition built from examining the adoption curves individually, there are insights to glean as well from the all-in-one graph. Here is the same graph again:

One remarkable pattern we are watching unfold is the group of 5 languages exhibiting a simultaneous surge in usage around February–March 2026. This timeframe correlates with the uptake in agentic coding among Meta engineers in Q1 of 2026.

Stratospheric among the breakout languages is TypeScript, cropped from the chart and sitting currently at +1200% growth from the prior year. TypeScript has historically been a niche language at Meta with less than ⅒ the userbase of Flow/JavaScript. In just a few months, TypeScript has closed that gap. Both engineers and managers are producing all kinds of dashboards and personal widgets in TypeScript that they never would have bothered to do without AI.

The second most rapid recent growth is in Rust. Meta engineers have written twice as much first-party Rust code so far this year than in the previous 9 years combined.

Reading down the right edge of the graph, the final three languages with rapidly growing adoption are Swift, JavaScript, and Go. (From further investigation into the growth of JavaScript usage, that one consists...