Short answer: Jordan Metzner says a friend told him Kiro runs an older Opus underneath, while developers want Opus 4.6. August Kiles says Cursor or Copilot with Opus 4.6 beats Claude Code with Opus 4.5, because the model matters most.
Watch this answer from Episode 31 (starts at 15:03) · Episode 31 notes and transcript
From Episode 31: How AI Is Replacing 100-Hour Due Diligence (Claude 4.6, Private Equity, and Emblem), starting at 15:03.
Sam Nadler [15:02]: Yeah. August, thank you for giving us, the high level on Emblem. If anyone's interested in checking out Emblem, go to emblem.pe, and let's jump into the news. We Jordan's alma mater is Amazon, so we cover it quite a bit. And we also love Claude Code here, so we thought this would be a fun story. But seems like Amazon engineers are, clamoring to get their hands on Claude Code versus the internal tool, which I think is Kiro. So, Jordan, just as a future Amazon or as a past Amazonian, what do you Yeah.
Jordan Metzner [15:39]: So I have a few friends that work at Amazon, and I remember talking to them a few months ago, and they were really behind on using LLMs for coding. They were, like, three or four OpenAI models behind. I don't think they had any Anthropic models. And then I just talked to some a few friends recently, and now they said they're kind of full stack on this Kiro. But one friend told me that in the background, it's just an older version of Opus. So it's just a wrapper on a wrapper, which isn't surprising. Right? Like, I mean, obviously, these companies are worried about security compliance and, like, regulations and a bunch of other things. Same time, wanna kinda push their developers to having the best tools. But you see, developers are kind of, like, going crazy because there's such a step function improvement between Opus 4.5 and 4.6 that, developers wanna get their hands on it as soon as possible.
Jordan Metzner [16:25]: So it does seem that, like, at least at the FANG kind of, MAG seven companies, the gap from launching new AI models to the market to, like, getting them implemented at the enterprise level is is getting shorter and shorter. And that obviously is, dangerous for startups and good for, probably the mag seven. But, as they get closer and closer to be able to launch, like, these new models when they come out, then they become, more and more powerful rather than, having to use kind of old technology. I think, like, we're lucky in that sense because, like, as soon as the new models come out, we're able to adopt them pretty rapidly, even same day. I don't know how you guys feel about that, August. But, yeah, I think it just shows that this time is collapsing. Maybe they're not there yet, but they're gonna be there really soon where, like, new model launches, even maybe we'll see that they get it first. I mean, I think I saw yesterday, OpenAI announced that everyone at NVIDIA is getting Codex, so you can just imagine that, like a new version of Codex is ready, and maybe they launch it to their clients first, so that, the whole model kind of flips, and these enterprises are getting access to the models even before the public does.
August Kiles [17:30]: No, I think you're totally right. I think we will see the enterprises getting first access sooner rather than later. Especially like I would imagine with Gemini, it'll be these new model, the new versions will be used in house at Google, long before they're used anywhere else. We already see with Anthropic, right, they're using Claude Code heavily internally, presumably with new versions of Opus, pre release. I guess I have a few thoughts on this article. It was interesting to hear, kind of that insider info that they're behind on which model of Opus they've got running. Because, I think that's really significant because the latest release was truly an inflection point in my opinion. And I have found in my own work that, especially if I really do a good job, because I do write a lot of code as head of product at least at this early stage of the company, but if I do a great job with the architecture and the prompting, it's actually pretty rare that I need to make any adjustments to what, the models end up creating, and that's a lot different than it was a year ago or even four months ago. So if they're leaving that on the table, that seems like something that's going to have to be corrected fast. My one contrarian viewpoint on this is that I don't really think that, I think people tend to like bandwagon these different products like Claude Code.
August Kiles [18:56]: At the end of the day, everything out there is really good and using, say Cursor or Copilot with Opus 4.6 is going to be so much better than using Claude Code with Opus 4.5 that it's not even funny. I personally do still use cursor sometimes. I really like certain things it can do. You can launch the browser inside there and it can For making UI tweaks, that's really helpful. The debug mode is really helpful cause it'll tell you, okay, go and click on this, and then it automates the logging and everything. So I think that all of these tools are great, and that the common thread is simply how good are the models behind them. I think it's the same thing, know, in terms of say Emblem and our competitors, right, we're all using the same models. We all good rag systems. So a lot of it is truly the product itself and the user interface and how quickly and how well can we build exactly what our users want. Because, our competitor with the new version of Opus is always gonna be better, on the back end than us with the old version of Opus, and vice versa.
Jordan Metzner [20:13]: Yeah, totally. I completely agree with you. I think it's like a new regimen jumping up to 4.6, and hopefully we'll keep seeing these step function improvements that kind of make things get a lot better. It's obvious that like the next step is kind of these like swarms or teams and kind of managing these teams of teams, and you know, I've been playing with it a little bit. It seems still a little rudimentary in like how successful they work, and at least on the Claude side to me, they've they haven't been able to end kind of eliminate major issues that I have to kinda go back and clean up or whatnot. But, yeah, I found the Codex models to be really good. And I think that, they announced this new frontier tool that they did launch to enterprises and haven't launched it to the public. So, maybe it's already happening and we're kinda behind the curve there. But, yeah, some great insights. Sam, what about you?
Jordan Metzner [20:58]: And I know you're not even using Cursor anymore.
Sam Nadler [21:00]: No. But I agree with August. I think, like, whether it's it doesn't matter what IDE you're using or not. I think that the model there has been such a change in my workflow. I'm not a developer by trade, August, and, like, honestly, lately, I've been, like, pushing code.
