Posted by icy 19 hours ago
I think sovereignty over what information you consume is more important than ever. I had to use Twitter for work to get news about <topic> but the amount of virulent propaganda, totally unrelated to <topic>, that you end up absorbing is unforgivable. Even if you think you're smart and don't pay attention to propaganda, by design it hits you at the subconscious level so you can't block it. The only social media I have left is LinkedIn and I really hate it but it has made a direct positive material impact in my life ($$$) so I try to hold my nose while I use it. I really would rather use some kind of federated LinkedIn, but when I last checked nothing like that existed yet.
It's so so so early. But I love how it moves from a world of maintainers & pull requests to a more ambient "this is what is working for me". I think this really is a next kind of leap. I don't know if we can keep relying on maintainer folks to guide each project forward like we have, if our agentic selves can be bandwidth limited & still go where we need to, channeling all our energy through individuals.
We need a federation of maintainers. A distributed of maintainers. Maintain ought be social. Tangled is great and I hope we can go beyond federation to many tangled, to widely widely tangled. And I hope we can go past maintainers too, past pressuring single people to have to decide it all. I think v-it really preceeda such an interesting agentic leaping off point that we are at, so interestingly.
If you push a lot of new features but your baseline is constantly failing, then something is wrong.
For example, in our company, most commits on main currently have 3-5 authors (we squash): 1-2 humans, 1-3 agents (cursor cloud agent getting started, ppl pulling it into cursor locally to continue, then review using copilot review, modify using copilot agent) then use a vibe coded github app offloading UI test execution to a beefy baremetal machine to adjust baselines.
Copilot review in particular is just so good, better than any agent i know (incl opus 4.7). It just allows you to skip the first few review rounds by humans and fix simple but hard to spot logical bugs, keep docstring & style up to date across the codebase, before you give it to a human - which means everyone can focus on writing more code.
Setting all of this up, at a massive scale, is just not feasible for any of these projects.
Github as we know it is gone, forever, it will never come back, except for niche hobby clones with .001% capacity that nobody will use. Agents are re-defining what software engineering means, they already have, right now,and are continuikg to do so, it's just that hackernews is lagging 6 months behind for some reason.
I don’t know their internals, though clearly they choose to tightly couple every major GitHub system to the AI offering, in my eyes that seems like part of the problem (plus Azure cloud migration on top because Microsoft sounds like a disaster).
Anyways, you sound angry.
I like to talk trash about Microsoft as much as anyone, they made insanely bad product descisions in the past (copilot in ms word is one of many) but this is not one of them.
My POV: Github actions are inconsistent in billing, security and require alot of attention to do right. Github has worse uptime than alot of free online videogame services, when most enterprise and business world leans on it for developers. Leaving a lot of users with terrible experience the past year having to constantly examine github firefighting for issues around availability, security, and billing instead of doing work that makes the company/people money.
Example walk through of securing github actions for ci/cd and managing SBOM python dependancy/supply chains (giant complexity) [1], Github has remote code execution[2], Uptime by 3rd party tracker shows 86% past 90 days. (First quarter in 2 years where they didn't have atleast one month above 90% uptime) [3]
[1] https://astral.sh/blog/open-source-security-at-astral [2] https://www.wiz.io/blog/github-rce-vulnerability-cve-2026-38... [3] https://mrshu.github.io/github-statuses/
This is likely on the back of Mitchell Hashimoto (Hashicorp founder) announcing he’s moving off of Github as well.
And really just years of Github feeling inconsistent, bad UX, no good solutions for open source developers in terms of AI spam etc.
Wow, it was a really long time ago it started going down the lane of the chute, can't believe someone missed it, made big news at the time back in 2018! This was the turning point: https://news.ycombinator.com/item?id=17221527
Check a local repo and go to pr's, there's a big banner telling you there's an ongoing ncident
Undoubtedly these various hosts will come under pressure from spammers and the like and they will react by placing extraordinary barriers around accessing the code.
That’s fine but it reminds me of the later stages of online forums, where it was impossible to browse most threads because you had to create an account and then build up community points until the screenshot of the kernel panic on the ZTE phone would be visible so you could see if it’s the same problem as yours.
GitHub was big and powerful enough to not need all of this but now we’re going back to the era of decentralization and I suppose with that come the pros and cons.
https://gitgrasp.com/ fixes this.
AI.
They're working on the scaling issues apparently due to huge demand.
"createIssue(title=string, body=string, labels=[string])" would be the same in Git's source code as it would be on a REST API server. The point of this is to standardize the software development lifecycle everyone uses around Git. That way you can do all the work we all need, with any VCS, without tight coupling. That's been the missing piece that nobody has made yet.
Want just the CI/CD component? Use that part of the schema. Want just the Issues? Use that part of the schema. Now you can write any tool you want, and just implement the features you want, and say "this follows the SDLC v1 CICD standard", or "the follows the SDLC v1 Issues standard". Much simpler to add extensions or support different use cases, without implementing everything you don't need. Yet everything's compatible.
We need that implementation-agnostic standard, so we can make transport-agnostic protocols, so different providers, clients, and servers can all talk to each other, without a hundred different bespoke "things". Rather than write your plugin-downloading app only against GitHub or against Federated-Whatever, you write it to use "httpSLDCs://some-server/v1". Don't want to use https? Use "grpcSDLC://some-server/v1", or "atSLDC://some-server/v1". You layer the application-specific protocol on top of the transport protocol, and express that in a URL. That's how we did 'federation' in the 80's/90's/2000's.
(also: did nobody come up with a better name? Tangled? Knot? you want your solution to be a tangled knot?!)