Hey, guys. Okay, let's try this.
Uh, to give you a little bit of context, B and me have been hacking on this editor idea for a couple of weeks now, and it's been fun, you know, working with Tory and Bun servers and whatever.
But I realized this week that building an editor is really hard. There are a lot of features that I would expect from an editor to be present before I am able to use them on the daily.
Like, there's no way, realistically, without there being, like, years of work in this that are ever going to be replacing VS Code or even Sought for that magic.
Sure, there is, like, some tiny editors that exist and people are getting viral with it, with like, for them, this artist project.
But realistically, this is not interesting, and that is this general problem with this project that I'm not really know what I'm going to build, because I'm not using this editor at all yet.
So everything I do is just theoretical, um, and it's completely, yeah, you either are... (sighs) I think there's two ways to build a product.
Either you, you know, you talk to customers and they give you daily feedback, and then you iterate to make things better, or you are the customer and you kind of build something you use daily and then like really figure out how to enable that workflow.
And with a custom editor, I just, I don't see that. (sighs) And yeah, I've been thinking a little bit more about my workflows as well.
I recently using Cursor, um, as both my editor and also in the past few weeks a little bit as, as my agent in trying Opus and I had like credits left on Cursor and it was there and so I tried it, whatever.
I think it's nice to have a very, this very integrated experience, but I don't think I care too much about it.
Like, (sighs) I still am pretty sure that agents, coding agents are just something that you're gonna replace whatever is the best right now and I'm using Cursor right now because it's available to me and it's free, but also...
Free as, y- yeah, I had already like paid for a license so I didn't have credits to use. (sniffs) But I'm just gonna use Cloud Code eventually, if they would have at least 4.5 Opus support to my $20 plan.
Uh, or I, also might just use Codex again, like next week there's some inside scoop that OpenEd is gonna bring out the new model again. Uh, so, you know, you don't know basically.
This whole model layer of innovation just feels like there's just so much happening that as long as you're not like doing it full time with a team of multiple people, like the AMP team, there's hardly, like, any innovation that you can do.
(sniffs) And then there's, like, there's one argument, you want like one interface to... Like, the, yeah, want a unified interface regardless of what agent you pick. I don't think necessarily this is true, like not for me personally.
I don't really mind switching between Cloud Code or Opus and actually there had been a time...
I think now the features are more stable but in the beginning there has been a time where like it hasn't been that and some, like, uh, some of those agents delivered a feature with UI that they still shipped and you were only able to use it basically if you used the stock UI.
There's no way any third party UI could adapt quickly enough.
(sniffs) I think that's kind of maybe overall the feature pretty much the same although like editing last fi- uh, last command in history making edits and stuff still not something every agent supports, I think. I don't know.
Anyway, that's not the point. The point is, I was pretty frustrated because I'm not really sure what I'm about to build or what I want to build and yeah that's been kind of nagging on me the past few days.
So I've been chatting with Valerie a little bit this today and he also said that, you know, he starts having similar thoughts.
It's hard for him to keep motivated on something that he's not using blah blah blah and then we were kind of talking about you know what is the thing that we could do to make it both like where our skills are already there and you know where we know we can we can build something but also something that we can integrate into all workflows very easily and you know Valerie knows from working on AMP just like you know I guess everybody in this team that there's like demand for stuff like uh tab completions for like non-VS code extensions and really like getting something out there that's gonna be something people like.
I know that too, like people are really pissed that, what's it called?
Super Maven is no longer available apparently that was like a thing that people used to use the NeoVim and yeah this guy this doc sky whatever who builds open code also made if he wasn't building open code he would build like a super completion service and I'm talking to people here that build world-class tab completion service.
There's got to be something that you can do and you all kind of got, not everyone but Valerie and Tom kind of got a little bit, you know, another word, I don't want to swear here, I have the child in me, but yeah you're no longer working at these companies basically so that's a huge opportunity to build something like that and I still have this feeling that...I wanna, like, make innovations in the UI layer more, just because that's what I'm really comfortable to do.
But that's, I don't think that's smart. Like, I don't think it's smart to fork the S code or something. And this whole direction of, like, building something that we can build in, integrate into our workflows very quickly just feels much more reasonable.
So yeah, the kind of proposal, and why I even built up anything of that, is, is that I'd like to have a catch, like get, get your idea on what would it take for you to, like, start another AI coding extension with me.
You all have been doing that before. I've been doing that before. I worked on, like, the code extension from Sourcegraph, the search one, before the AI was a thing, and then obviously Cody, where we all worked together. And you have been doing it with M and with Client.
That's, like, where we have a lot of experience for, and that's something that we can all integrate into our workflows really easily. And I've been thinking, like, what would that extension b- make unique?
So it's been... This is like, uh, I don't really have, I haven't really thought about VS code extensions a ton in the past few years, a little bit because this code, uh, K8 also has a VS code extension, but, uh, not in terms of, like, AI engineering.
But I want to build something where it's really like you h- you have your agents, and those switch every time, and this is fine by us. We don't... I don't want us to build an agent. This is like...
I don't see us, like, being able to compete.
But we do all the rest that agents don't do. Like, we give you completions. We allow you to, like, have, like a sidebar where you can see all your latest transcripts.
I've been building something like this for myself when I was between Claude Code and Codex, just like a C like command, and it shows you all your previous chat histories, just parsed from your local files.
And we could put that into a sidebar, and you could cut back, see your existing agents, can make it work for Claude Code and Cody, sorry, Claude Code and Codex immediately.
I have the code, but then we can also add more stuff if we want to, like cursor agents, just like a sequel lite file that we can parse.
So that, plus obviously tab completions, which is a whole nother bit, f- like fall. We don't have data for this, so we need to get into this business.
But then Valerie shared with me that there's some maybe, uh, a v- commercially available model that we could use in the meantime. And when I was looking for that... Oh sorry, baby woke up. Gonna be a little bit...
Okay, I need to be a little bit louder, uh, quieter. (laughs) Yeah, and I was looking for that a little bit and apparently it's, some other companies are doing the same. I think there is, like other...
Oh God, I shouldn't have done this in the walk. I cannot even look at the references. The point is, I think it'd be fun to just hack on something and just see where it leads, not necessarily have, like, any commercial interest.
We, we're just trying to see what we can do in our workflows to improve things.
I want your brains on, like, thinking through, like the kind of everything besides agent flows. And, uh, I think there is like...
Sure, there is like a giant market, there are like many competing products and anything, it's gonna be hard to stand out, for sure.
But also like the fact if we do, it is an extension, (sniffs) it's very low commitment for people to try it out. And I was just looking at this earlier.
Cody has like 800,000 downloads. That's pretty insane. Of course, times have changed. Wouldn't build a ............................
But yeah, kind of feeling more excited about this, which is funny because that was the initial idea all along, like Cody 2.0. And took a while for me to see that, yeah, that's, that's kinda the obvious thing we should do.
And then we use it, we make it better for us. At some point we share it. Maybe we can even ask, (sniffs) uh, you know, Sourcegraph, probably doesn't feel too great about that.
So maybe we can even get like a... I don't know. I don't know what I'm trying to say here.
Uh, yeah, in terms of like technically doing things, we love to discuss this as well, but I'm just trying to see like, would you guys be excited to, to work with something like this?
Kind of get all of our learnings we had from Cody, from M, from Client, from Taming extension, (laughs) I'm obviously using in this project, I admit.
But can we get all of these learnings and, uh, yeah, build something fun to use. Start as a VS code extension, get something in o- in other people's mind.
Be kind of the open source, ............................ Be kind of the alternative here for people that, yeah, I'm gonna stop rambling now because it's normal substance.
Curious what do you guys think.