I've used Helix for all my recreational programming projects (in Rust) for about 6 months and I've written about 10kLOC of code with it. Still using vim at $WORK for a variety of reasons (giant c++ codebase that does not easily plug in to clangd/LSP), but I might be doing the switch soon.
Installation was easy and the default configuration is good. Plugging in rust-analyzer still needed a line or two of config file editing, but that should not be necessary for very long. Hint: use `hx --health` to check that all the tools are properly installed and configured (it shows a matrix of programming languages and their associated tooling, with red/green check marks).
My Helix config file is about 5 lines long, but my Vim config is in the hundreds of lines.
I had very little friction coming from Vim and a little bit of Kakoune.
I particularly like the fact that all the important features are built-in and not half baked scripts with unintended side effects (e.g. opening a Makefile in Vim triggers a ftplugin script that messes editor-global config instead of buffer-local, with the default out of the box scripts).
Lots of people are requesting a plugin extension system, which is kinda understandable since every editor seems to have one. But I haven't had a need to add any kind of plugin, and it's easy enough to fork a shell for an external process if needed.
It's still rough around the edges, but good enough for a daily driver.
Thanks to everyone working on the project.
For me, coming from writing Clojure with Neovim + vim-sexp + vim-fireplace, I would need Paredit and REPL integration.
Paredit is doable with Treesitter, and I'm actually really excited about Helix's Treesitter integration--- every time I write a non-Lisp language, I miss structural editing, so it'll be nice for other languages catch up to Lisp on that front. It doesn't really look like it's quite there yet, though. The built-operations that use Treesitter are spare [0], and while you could implement operations like promote, slurp, and barf with them, you'd have to clobber some register. I don't see at all how you could implement splice without an actual language, since you would need a way to select all siblings of the current node.
REPL integration absolutely needs a plugin language. Unlike the LSP, there isn't a standard protocol for communicating with a REPL, so each language requires its own REPL client. Unless they want to ship a REPL client for every language under the sun, they'll have to provide some way for users to implement their own clients. That said, every time I look at Conjure, they've added REPL clients for more and more languages [1], so maybe it is feasible to have it built into the editor.