Four in ten of the most popular AI tools did not exist eighteen months ago
41 of the 100 most-starred AI projects were created in 2025 or 2026, and eighteen of them did not exist in January. The old guard is not dying — among the top 50, not one project has stopped. It is being overtaken.
TL;DR — the short version
41 of the 100 most-starred AI software projects were created in 2025 or 2026. Eighteen of them did not exist in January.
The churn is not caused by death — it is caused by replacement. Among the top 50, exactly zero projects have stopped receiving commits. The old guard is still healthy. It is simply being overtaken.
133 projects created in 2026 already hold more than 10,000 stars, and the largest passed 238,000 in seven months.
Every “best open-source AI tools” list has a shelf life, and it is much shorter than the people writing them assume. We can put a number on it.
Rank every AI software repository in the catalogue by stars, take the top hundred, and ask a simple question: when were they created?
Four in ten are younger than eighteen months
41 of the top 100 were created in 2025 or 2026. Eighteen were created this year — which means that a list of the most popular AI tools written last Christmas is already missing a fifth of its entries.
Look at the very top and the effect is stronger, not weaker: 44 % of the top 50, and 40 % of the top 20. This is not long-tail noise. It is happening where everyone is looking.
The projects doing it
These are the largest repositories created since January, with the star counts they reached in a matter of months:
| Project | Created | Stars today |
|---|---|---|
| ECC | January 2026 | 238,557 |
| skills | February 2026 | 208,700 |
| claw-code | March 2026 | 195,004 |
| gstack | March 2026 | 126,797 |
| graphify | April 2026 | 103,999 |
| ponytail | June 2026 | 98,219 |
| caveman | April 2026 | 96,698 |
Two things stand out. The pace — ponytail reached 98,000 stars in about two months. And the theme: almost all of them are tooling for coding agents. Harnesses, skill libraries, context tools. The 2023 wave was about wrapping a chat model; this one is about feeding an agent.
The part we got wrong before we checked
The obvious explanation for a turning-over ranking is that the old entries are dying and dropping away. We assumed it, then measured it, and it is false.
| Rank band | Projects with no commit in six months |
|---|---|
| Top 50 | 0 (0 %) |
| Top 100 | 4 (4 %) |
| Top 500 | 43 (9 %) |
| All 8,246 tracked | 3,220 (39 %) |
Not one of the fifty most-starred AI projects has gone quiet. The abandonment problem in open-source AI is real — 39 % of everything we track is dormant — but it lives in the middle and the long tail, not at the summit.
So the top does not turn over because incumbents rot. It turns over because new projects arrive fast enough to push healthy incumbents down the page. That is a much better sign for the ecosystem, and a much worse one for your bookmarks.
The other side of the number
Stars measure attention, and attention arrives faster than judgement. A project with 98,000 stars and two months of history has not survived a dependency upgrade, a breaking API change, or its author getting bored. Some of the names in the table above will be in a dormancy article in 2028.
That is the trade. The new entries are where the ideas are; the older ones are where the proof is. A reasonable rule: take architecture ideas from what is rising, take production dependencies from what has been maintained for two years.
How to keep a list current without living on GitHub
- Re-check quarterly, not yearly. At this rate, a year of drift is 40 % of the field.
- Sort by recent momentum, not by total stars. Total stars is a cumulative measure and structurally favours whatever is oldest.
- Watch the category, not the tool. The winner in agent tooling has changed three times in eighteen months; the category has not moved.
- Treat anything under six months old as promising, not proven.
Our emerging page tracks exactly this: healthy projects gaining ground that are not yet in anyone’s list, rebuilt every morning. The projects list sorts by momentum as well as by stars.
Method
8,246 AI software repositories tracked as of 7 August 2026, ranked by star count. Courses, awesome-lists, roadmaps and books are excluded by the same name-and-language rule used in our dormancy analysis — several of the most-starred entries on GitHub in AI are reading material, and including them would measure something else entirely. Creation dates and last-push dates come from the GitHub API. “No commit in six months” means no push in the last 180 days.
One limit worth stating: a catalogue of 8,246 projects is not all of GitHub, and our inclusion criteria favour projects that are visibly about AI. A tool that became AI-adjacent later may enter the ranking later than it deserves, which would slightly overstate how new the top is.