Cursor's blog on scaling Git makes me think that Anthropic just likes reinventing the wheel with their infinite token capacity: https://cursor.com/blog/git-at-any-scale
Cursor scaled git to way more writes/s than Anthropic did with their rewrite.
> I don't swear that this is how GitHub actually works, but I've had knowledgeable friends swear up and down that it is.
All the details on how GitHub's infrastructure has evolved over the years are very publicly detailed in the GitHub engineering blog and in technical talks. There are no "secrets" or "rumors" here, all the information is one google search away. Perhaps you need to re-evaluate your priors on how knowledgeable your friends are.
Hey few things that are deal breakers for me (I tried Intellij IDEA community edition, AI assistant)
1) Import of historical data from agents. I often look up past agent sessions and require search into them as a primary feature of the harness. I know for a fact that historical conversations are not encrypted for both Codex and Cursor on a local filesystem, so Intellij tool could import them into ai assistance plugin even before user is authenticated with either Cursor or Codex. There are few annoying things in search for both Cursor and Codex extensions, so having great search over past conversations could be a great introductory feature
2) Auth .. I am not sure about this one technically because I haven't looked myself into how agent auth is done for Cursor and Codex, but at least for Codex, their IDE extension for vscodium/cursor/zed/etc. is able to use existing authentication from the CLI codex tool, so I do not need to login into every new IDE I try Codex extension in. With Intellij Codex auth did not work at all for me.
3) Cursor authentication worked, though I needed second one for Intellij while being logged in with CLI, but the model reasoning level is not selectable from within ai assistant, one more deal breaker for me unfortunately (I found it stays like this for a long time: https://forum.cursor.com/t/intellij-cursor-acp-reasoning-lev...)
What you are saying sounds like an assertion that the problem described here[1] - by Vicent Marti - that packfiles aren't contiguous blocks of data and so every fetch fans out to more fetches from the client - is not actually a problem? Or is there an innovation in zerofs that somehow solves this for git? From what I can see it offers NFS and NBD interfaces and GitHub tried and rejected both of those very early on.
The problem he described seems consistent with known patterns from datalake storage architectures, its a big part of why columnar storage is the only game in town for most use cases.
This is why TFA has a lot of interest here - he has changed the storage layout of git packfiles to a columnar model to address this challenge.
You sound like someone who has not been informed about the problem GitHub is facing. No one else is facing anything close to it, so there is no meaning to "the bad one".
IMO one of the biggest losses of the OpenAI/Cursor breakup will be the loss of OAI models on CursorBench [1]. Their bench has always been one that most-fit my mental model of how good each of these models are. I find AA’s Intelligence index to often be out of alignment with my own subjective evals.
Cursor seems to have intentionally dropped semantic/embedding-based codebase indexing. Their explanation is that newer coding agents are now good enough at searching repos directly. They can grep in parallel, inspect directories, read likely files, and refine the search themselves. Cursor says this now works as well or better for most cases. [1]
So the architecture shifted from something like “chunk → embed → vector DB → retrieve” to “local search index/ripgrep → agent explores.” There is still an index, but it’s a conventional local search index. No embeddings are involved in that retrieval path. [1]
The interesting part is that Cursor was still reporting a ~12.5% accuracy gain from semantic indexing in January 2026. [2] So this looks less like “semantic search never worked” and more like agentic search improving fast enough that the extra infrastructure stopped being worth it.
I use Cursor at work and love it. It's a great IDE, the best IMHO. But it becomes really expensive.
Since the introduction of the Cursor Token Rate, my company tells us to use Claude and Codex directly with Anthropic (or Vertex AI) and OpenAI, respectively.
I'm now using Grok 4.6 most of the time. Sonnet some times. But that's it. Cursor-native models 99% of the time. I'm pretty sure Anthropic will follow OpenAI soon in pulling their models out of Cursor because it's now a true competitor.
Looks like a React component tree binding to SVG nodes. Right click and inspect element. Right click on the svg element in the (chrome) inspector. Choose `break on` > `subtree modifications`. That led me to look through the network tab for this: https://cursor.com/marketing-static/_next/static/chunks/0sqf... (look for the text "GitObject" in this file)
No they were not.
Especially if you were an end user, you wouldn't have been able to tell it wasn't their own model. That exactly isn't being clear nor public.
Cursor scaled git to way more writes/s than Anthropic did with their rewrite.