Short answer: Jordan Metzner says Cursor is his go to IDE but model makers may build the best tooling. Anand Chandrasekaran expects coding models to saturate, making tooling and price decisive. Arun Kalaiselvan calls Google the silent killer.
Watch this answer from Episode 24 (starts at 31:08) · Episode 24 notes and transcript
From Episode 24: This AI Fills Healthcare Shifts in Minutes, starting at 31:08.
Sam Nadler [31:08]: Alright. Alright. Now, I'll kick you off with you. Cursor CEO believes OpenAI, Anthropic competition won't crush his startup. I know you're a heavy Cursor user, and but you guys also have some thoughts on, how it's it's position as the application layer versus the model layer.
Jordan Metzner [31:30]: I mean, have every IDE installed. Like, I have Visual Studio. I have Antigravity. I have Windsurf. I have Cursor. My go to is cursor. It's like I think it's definitely the most developed. And when I can use, like, multiple agents simultaneously, it's definitely the best, like, product user customer experience. I've been using Antigravity a little bit, but, it's free, so I use try to use up my tokens until they kick me off. But it's not as polished as a product. It does have some cool, like, one off features that are pretty cool. So I thought, like, obviously, there's just so much copying going on or, like, I don't know, like, feature, like, feature parity might be, like, the proper word. So, like, if anybody launches anything, everyone else will just launch it. If it's cool enough and, like, the time to market is so fast. Like, and my Cursor, I can't even use it for one day without requiring, like, a feature up you know, a product update on it.
Jordan Metzner [32:17]: So I think at the end of the day and honestly, it is surprising to me how good Claude Code is and, like, the path that Cursor has gone, the path that Anthropic has gone because like I think that Anthropic could like, choose a selfish model where they would literally sell their, code editor only in their tools and nowhere else. And I don't know. To me, it would seem like an even bigger opportunity. Maybe I don't like, I'm sure they've done that evaluation. Dario's a pretty smart guy, but, sell to everyone has been their model. But, at the end of the day, it's whatever the best coding model is. In my personal opinion, I try to think of costs not being a factor because my coding time is so expensive to me. Right? We have you know, these Anthropic models are so expensive. We have seen that they cost more money. Like, I see my cursor credit card getting hit all the time. But, realistically, like, I don't know. Does this, like, essentially see the inflationary values here, where this just becomes cheaper. I think so because already I'm running four agents simultaneously when, like, before I wasn't. So, I think we'll see that.
Jordan Metzner [33:23]: But yeah. I don't know. What do you guys think? How are you guys coding? And, like, how do think?
Arun Kalaiselvan [33:26]: We are controversial on Cursor.
Anand Chandrasekaran [33:28]: I mean Okay. Cool. We like Cursor. It's it's a good product. But this when I read this article, I kind of understand where she's coming from. Right? Like, so think about it like this. Two years ago, Claude was the only coding model. Today, we can actually do some work with Gemini three. Still not yet there. But in two years from now, I'm go there is going to be a saturated simple state for coding from most of the models. So at that point, you are no longer going to be dependent on Claude. And at that point, it was going to become tooling is going to be the most important aspect. And then it will be like, is going to be giving you all those things at the less the least cost. And that time, Cursor will be probably the top in the market line because they've already been trying to do the best tooling and the way the interactions and all for a long time, and they would be in a very matured state, and they will all be able to offer things for its much cheaper price.
Anand Chandrasekaran [34:28]: I don't know where Claude Code will be at that point, but Claude Code is already very expensive when compared to everything else. And then I'm I'm assuming there's going to be some enterprise system play going on by Cursor where it's it's like it's going to be adapted by many people. So I do think that it's going to what he's saying is right and if that's the way it would go because there is a huge competition to build the best coding model still exist. And once it's solved, like, we're still fighting with Claude and everybody's compete. I don't know long
Arun Kalaiselvan [35:01]: You can solve that, though. There is no clear definition of best coding model. Like, there is no clear definition of best engineer. Right? Like, it's a if you can get there to the ninety fourth percentile correct, then that's all that really matters.
Anand Chandrasekaran [35:14]: That's what I'm saying. Right? That's that's Okay. But
Jordan Metzner [35:16]: But let me push back on you guys for a sec. Yeah. If you say that, like, talking, as we spoke earlier, like, using this Pydantic model, like, talking to the models, if you know how the model is built, the more you the closer you talk to how the model is built, the better are you at like getting the model's response rate. Then you would have to argue that those who build the model would be best at creating the tooling around said model. So while you well, let me just like finish this thought. Right? While that might be so might be true, so if inherently, like if all these models to your point kind of reaches logarithmic curve, then he who has one of these top models and is able to best tooling around their set model and cost might be the number one definitive thing in the in that sense. Right? Because like, hey, do you want an AI code developer that's doing it at, a dollar an hour or at $200 an hour or at 50¢ an hour or at 5¢ an hour and like, does, like, going all the way back to, like, the chips, like, is Google Silicon able to, like, then allow you to use Gemini at this logarithmic curve and they built the best tooling around it and then their IDE because, like, you have to understand there the Google bought, this windsurf. Right?
Jordan Metzner [36:23]: Got You've Microsoft that has VS Code that's already open source. Right? We haven't seen anything from Amazon but, like, maybe they would buy it. Maybe they would buy cursor. Maybe it looks like cursor is almost too big and it's gonna run alone. So, is the cross stack tooling across all these things important? If we reach this log driven curve and it doesn't matter and all the models are the same, then you just want the cheapest best one. And so then he who builds the model may be the owner of that and cross tooling might not be important at all. I don't know. I'm just trying to flip the other side of it.
Arun Kalaiselvan [36:49]: But here's the deal. At that point, is tooling even important? If you have a AI coding system that's doing it, why do you even use need to use the tools? So for us with Cursor, we did a whole six month pilot with Cursor. We were on their enterprise ecosystem. We actually found that they introduced more issues into our code base than they solved for. So we ended up completely shutting our cluster from our eco Cursor from our ecosystem, and we switched to a Copilot.
Anand Chandrasekaran [37:15]: Yeah. I think that was back in the day when they were trying to figure things out.
Arun Kalaiselvan [37:19]: This is about, what, two years ago at this point?
Anand Chandrasekaran [37:21]: One and a half years ago. Yeah. But Cursor is now very mature. I mean, yeah, I see. Like I said, I've actually had a chance through a friend to directly talk with one of their co founders or CEOs. I'm not sure. But, like, we had discussions and I gave them reports and stuff. But my point is to talk about what you are saying. Yeah. But you're right. With if Claude becomes the winner of that, then yeah. But more often than not, like in the technological world, there is always an alternate in anything that you do. And I don't believe that there is going to be one single there is always going to be two or three. Yeah. And there's always going to be this open source thing. And like, if we but look at Andrej Karpathy doing things. Right? He's like built he built his own new LLM model that you can just run, and then now he has built this judge concept that that that you can do use for anything. Right? So there are gonna be people who are gonna be pushing open source stuff all the time.
Anand Chandrasekaran [38:14]: In the last ten years, the most successful code editor has been VS Code from Microsoft, which was which has been enterprise for the longest time, embraced open source, and they became that. So that's that's how I'm I'm I'm seeing it.
Arun Kalaiselvan [38:26]: So as far as evolutions go, yes. What I don't know is what the next evolution looks like. And we went from I was a religious VIM user. I went to Sublime and then from Sublime, we went to Visual Studio. Right? So it's a and of course, cursor is is basically VS Code. Yeah. It's just a
Anand Chandrasekaran [38:47]: Yeah. Is VS Code. Yeah. Antigravity is VS Code.
Jordan Metzner [38:50]: Yeah. Of course. That's why I'm so long Google because I think, they can have a top model provider for cheap and, they got the ID
Arun Kalaiselvan [38:56]: I do think Google is going to be the silent killer in this current ecosystem. The way we've seen the Vertex AI team operate in the last three years has been fantastic. I think Amazon has a lot of catching up to do if they want to be competitive. They I've I've seen this in my blockchain days as well. Right? Like, all the clouds got it wrong for the blockchain world. And they've at least done a better job on the AI side of adoption, but Google is, like, leaps and bounds beyond where Amazon is. I can't really comment about where Microsoft is right now, but we are definitely all in on the Gemini team.
Jordan Metzner [39:37]: Yeah. That's awesome. Alright. Great episode, guys.
Sam Nadler [39:40]: Yeah. No. I just wanna say thank you so much. Thanks for joining. And, before we cut, just like and subscribe. We're here every Friday. Thank you, Anand. Thank you, Arun. And please check out Arya Health. Thank you so much for having us.
