DuckDB isn't losing its contributors. It's renewing them.

I suspected the people who build DuckDB and its extensions were drifting to other projects, and I suspected AI was pulling them away. So I tracked 562 of them across five years of GitHub activity. The suspicion was half right, and the half that was wrong is the more interesting part.

Disclosure: I maintain more than thirty DuckDB community extensions through Query.Farm, so I’m one of the prolific extension authors counted below. Everything here comes from public GitHub data and is reported in aggregate. No individual is named.

The headcount says nothing is wrong

Start with the obvious check: are fewer people contributing? The answer is a flat no. In the last twelve months, 343 people had a pull request merged into duckdb/duckdb, up from 184 the year before and the most ever. 264 of them were first-timers. Merged PRs doubled, from 2,410 to 4,823. Community extensions grew even faster: 32 were added in 2024, 104 in 2025, and 279 so far in 2026.

But a growing headcount can hide a lot of turnover. Look at who comes back. Of the people who merged a core PR in a given year, about two thirds don’t merge one the next year: 61% of 2022’s authors, 67% of 2024’s, 62% of 2025’s. Most of the people in any given year’s count contribute for a while and then stop.

So I followed people, not totals

Therefore the right question isn’t how many people contribute, but what happens to each one. I took everyone who had become a core contributor (two or more merged PRs) or an extension maintainer before November 2025, and checked their status today. “Active” means any DuckDB activity in the last six months. That window comes from the data: the median gap between a contributor’s PRs is nine days, only 6.5% of gaps exceed six months, and 59% of people who go quiet for six months never merge another core PR.

For community contributors (everyone outside the DuckDB organization) the result looks, at first, like an exodus. Of 205 established community core contributors, only 32% were active in the last six months. Another 11% are fading (last seen six to twelve months ago), 34% are busy elsewhere on GitHub with no DuckDB activity for over a year, and 22% have gone quiet in public altogether. The groups from 2021 and 2022 are mostly gone.

But they were never really DuckDB people

But “moved on” assumes they were all-in to begin with, and most weren’t. I measured how much of each person’s time went to DuckDB as the share of their active GitHub weeks that included any DuckDB work. In the last year, the median community contributor did DuckDB work in 20% of their active weeks, and that figure has stayed between 9–20% every year since 2022. Members of the DuckDB organization sit at 100%.

Therefore most departures probably aren’t people leaving DuckDB for something else. The likelier reading is that they contributed the one or two things they needed and returned to their main work: the fix landed, the feature shipped, and they went back to their day job. That would also explain why the drop-off rate has stayed flat since 2022 instead of rising: this is how contribution to DuckDB has always worked.

It also means the core team does most of the work, which is typical for a company-backed open source project. Members of the DuckDB organization wrote 70–80% of merged core PRs in every year from 2020 to 2025. In the last twelve months that fell to 69%, the lowest since 2019, as the newcomer wave grew. The community’s share is rising even though individual community members come and go.

Where their attention goes

But “back to their main work” isn’t the whole story either, because that main work is changing. For everyone who moved on, I classified the repositories they now work on (with TypeSafe’s Jev model, using each repo’s name, description, topics and language). Most of them went to the things you’d expect from database people: developer tooling, personal projects, and other engines like Apache Arrow, Polars, ClickHouse and Velox.

One area is growing much faster than the rest. In 2022, 9% of active established DuckDB contributors touched an AI/ML repository. In 2026 it’s 27%, three times as many. Over the same year, the number of these people doing any DuckDB work fell for the first time (173 → 147). AI isn’t where most of them went, but it’s the area taking a growing share of their time.

Extension authors should be the flightiest. They aren’t.

Therefore I expected extension authors, who often build extensions alongside other work, to look worse still. They look much better. Of the 80 maintainers who published their first extension before November 2025, 72% were active in the last six months, 10% are fading and only 12% have moved on. Their extensions are maintained too: 370 of 415 had a commit from one of their own maintainers in the last year. Of the 45 that didn’t, 30 are being kept going by other contributors and only 15 have had no commits at all.

A methods note, because it was the obvious objection: many extensions keep a branch per DuckDB version, and 16,762 of the 36,630 extension commits from the last year exist only on non-default branches. Counting every branch hardly changed the picture (368 extensions with a maintainer commit on the default branch vs. 370 on any branch). The same maintainers do both.

Extensions are now the main way in

But extension authors aren’t the same people as core contributors. Only 15% of maintainers also have two or more merged core PRs. In 2024, most first-time extension authors (15 of 26) had already contributed to DuckDB core. In 2026 it’s 17 of 166. The number of first-time authors nearly tripled in a year (58 in 2025, 166 so far in 2026), and roughly half of all new extensions now have a first-time author.

Therefore the extension registry does something core contribution can’t easily do: bring in a steady stream of new people who stay. It’s also concentrated. 7 authors with ten or more extensions each maintain 123 of the 415, while 184 authors have exactly one.

But AI is pulling at them too

But sticky doesn’t mean dedicated. The median extension author did DuckDB work in 48% of their active weeks, and DuckDB is the main area for only 38% of them. Their biggest single destination outside DuckDB is the Apache data projects (Arrow, DataFusion, Iceberg, Doris, Sedona), where 29 of them contributed this year. The fastest-growing area is AI agents and MCP: 6% of active maintainers touched that space in 2024, and 28% do now.

So what’s worth watching

The hypothesis was that DuckDB’s contributors and extension authors are moving on. Individuals do move on, quickly, and AI is taking a growing share of the same people’s time. But the community isn’t shrinking. More people arrive each year than leave, and the newest arrivals come through extensions and stick around.

Therefore the things worth watching are narrower than an exodus:

• A small number of very active people. As in most company-backed projects, the core team writes most core PRs, and a handful of prolific authors maintain a large share of the extensions. Their continued involvement matters more than the overall headcount.
• Part-time attention. Even the stickiest group spends most of its weeks elsewhere. DuckDB competes for their time, and AI is taking a growing share of it.
• The people fading now. 11% of established contributors were active in the past year but not in the last six months. Historically about half of people silent that long don’t come back, so this is the group most worth reaching out to.

DuckDB’s contributor base renews itself rather than staying fixed, and so far it has renewed faster than it has shrunk.

How this was measured

• People. 346 GitHub users with two or more merged PRs in duckdb/duckdb (13,544 merged PRs) and 251 maintainers listed in the community-extensions registry, 562 unique people in total. “DuckDB organization” and “Labs” mean current members of the duckdb GitHub organization, mostly DuckDB Labs staff, based on GitHub’s author association. Bots excluded.
• Activity. GitHub’s per-user contribution data for five November-to-November windows, plus dated contributions for active weeks, plus every commit on every branch of all 411 extension repos for the last year. One person’s contributions to one repo in one year are capped at 1,000.
• Status. Active: DuckDB activity in the last six months. Fading: six to twelve months ago. Moved on: none in twelve months but at least ten public contributions elsewhere. Quiet: neither.
• Areas. Repos under duckdb, duckdblabs or MotherDuck and registry extension repos are DuckDB by definition. Jev checked the 623 repos that merely mention DuckDB, and moved 105 general projects (e.g. Lance, Rill) to their real area. Everything else was classified into areas and sub-areas by Jev. 1,975 of 7,205 area labels had confidence below 0.6.
• Limits. Only public GitHub activity is visible, so “moved on” means moved on in public, and private or GitLab work is invisible. GitHub only counts default-branch commits in per-user data (PRs count on any branch). Contribution volume roughly doubled in 2026 across every group, in DuckDB and elsewhere. I didn’t investigate why, and the findings rely on head counts and active weeks rather than commit totals so the jump doesn’t distort them.