You can pick an agent framework this month and not find out for another year that the people who wrote it have stopped adding to it. Of the five frameworks StillShipping puts side by side, one states its own maintenance status on its front page. The other four publish nothing about it.
The one that does is AutoGen, in its own words, read off its README on 2026-09-04: "Maintenance mode. AutoGen is now in maintenance mode. It will not receive new features or enhancements and is community managed going forward. New users should start with Microsoft Agent Framework."
What StillShipping is
StillShipping is a freshness board for agent tooling: orchestrators, spec frameworks, command line agents, IDE extensions. Every tracked tool carries one of three verdicts, maintained, stale or dead, and every verdict is recomputed each night at 02:00 by ops/refresh.mjs from the GitHub API. It's for a developer choosing between two frameworks, or already depending on one and wanting to know whether anyone is still there. What you get is a row per tool with a score out of 100 and every signal that produced it, a page of the ones that have stopped, a page stating the scoring rules, and the same data as JSON. Reading it is free. There's no key, no account, and no cap on how often you can call it.
The score is built from what the API can actually see. A push inside a week starts a row at 100, inside a month 92, a quarter 72, six months 55, a year 32, two years 12. Past 365 days of silence the dead verdict is forced whatever the rest says, and an archived project drops to 0 immediately. 65 and above is maintained, 25 and below is dead, and the gap between them is stale.
The release signal costs 12 points, or 2
A project that tagged a release every 40 days and hasn't tagged one in 300 loses 12. A project that has never tagged a release at all loses 2. The reason sits in the comment above the branch in ops/score.mjs: the absence of a habit is not the breaking of one. Something with no releases is unreadable on that axis, not evidently dead, so the heavy penalty stays reserved for a project that had a cadence and broke it.
| Project | What its README publishes about its own maintenance | What it publishes instead |
|---|---|---|
| AutoGen | Maintenance mode, with a named successor | A migration guide to that successor |
| LangChain | Not published | Routing advice inside its own ecosystem |
| CrewAI | Not published | Adoption, "over 100,000 developers certified through our community courses" |
| GitHub Spec Kit | Not published | An "Experimental goals" section on what its authors are researching |
| BMAD-METHOD | Not published | Prerelease builds and setup, in a separate installation guide |
Every cell was read off the project's own front page on 2026-09-04.
The cells re-check themselves
Each quoted cell stores the source URL, the date it was read, and a literal string that has to still be on the page for the figure to still be true. ops/qa/compare-gate.mjs --verify re-fetches every source and fails on a missing string. On 2026-09-02 it caught one: CrewAI's README had moved from uv pip install crewai to uv tool install crewai, and the crewai[tools] extra was gone from the file entirely.
Here is the board: https://stillshipping.kynth.studio/kd
One shipped product, taken apart, once a month. What it does, what it cost to build, what the pipeline behind it looks like, and what the numbers did, read off the repository and the live site, not written from memory. Join the list.
Top comments (0)