00:00:00 · Brett Caughran, Khe Hy
[00:00:00] S0: The challenge in AI adoption has been the heterogeneity of what investors want. [00:00:05] S1: More competition is good for us. More options, more choice, lower prices. A lot of them power Mac mini users that were using it for OpenClaw. Most have moved over to Codecs. [00:00:18] S0: I think to me what has been so, so exciting, and I think we're at a real inflection point on these tools on Wall Street is the fable and soul class of models have gone from answer engines, prior LMs were like Q and A, give me an answer, to actual work mechanism. Hello, and welcome to the next episode of Invest with AI, where we explore the intersection of artificial intelligence and fundamental investing. This is the no guest episode where Kay and I will chop it up and sort of do a little distillation of what we've been thinking about and what we've been working on recently. So Kay, what have you been thinking about and working on recently when it comes to AI and the craft of investing?
00:01:12 · Khe Hy
[00:01:12] S1: Yeah, great to be here. Great to see you, Brett. So the most recent, this will come up in another episode, but Grok Bot. So as you know, I'm a big kind of personal agent, want to use it from my phone, want to really kind of always be interacting with, you know, a personalized harness. I have not downloaded Grok bot yet, but it's kind of on my list. And it's interesting, man, because, you know, Grok was kind of written off two months ago. SpaceX goes public. They buy Cursor with that rich stock. Cursor has been crushing it on the kind of very narrow, you know, pre trained models, right, for coding. SpaceX got the data center and the compute. It's like a... You know, it's a three horse race now again. And I have not spent a lot of time on Grok. This has all happened the past couple of weeks. But it's nice that it's a three horse race, and it feels like Google is out of the picture temporarily. And then you have that, like, massive exodus of Demis or, you know, the organizational change of Demis and then Jeff Dean and his partner Sanjay, I think, like the legendary ML programmers leaving. So it'd be interesting to see. But, you know, from my perspective and our perspective, I'm like, more competition is good for us, right? More options, more choice, lower prices. So I did have something on Codex, but I don't know if you had any quick thoughts
00:02:54 · Brett Caughran
[00:02:54] S0: Yeah, it's remarkable how fast things change in this landscape. And it wasn't so many months ago that people were setting up their open claws and buying Mac minis. What's your sense of that? Like, are those mothballed now? Are we gonna see, like, thousands of Mac Minis show up on Facebook Marketplace?
00:03:14 · Khe Hy
[00:03:14] S1: I looked, and it's still hard to find them. Like, they're still sold out from Apple as of a few weeks ago. I think that so when OpenClaw to Hermes, which I never tried, but I've heard very good things. But then and then as of two days ago, Grok bot is just blowing up. And there is a little bit of the kind of SpaceX, Elon, Mafia kind of pumping it up right now. So in my mind, I'm going to wait till, like, a few of the normies come out and have a few weeks with it because it definitely seems like there's an influencer campaign happening as we speak on GrokBot. But the and this is actually a really good segue, the Mac mini question. It's the thing that I wanted to talk about today is that a lot of the power Mac Mini users that were using it for OpenCLO. Most of it have moved over to codecs, which I guess there's so many naming problems with ChatGPT for work or basically like Claude code for OpenAI. They've got a massive product confusion right now. But let's I'm gonna call it Codex so people know that it's like the Claude code for OpenAI. So Codex comes out, and it's first, it was pretty bad. This was, you know, six months ago or four months ago. Was lagging Claude code. And then they did a massive redesign of the app, 5.6. Soul came out maybe, I don't know, four weeks ago, six weeks ago. And a few things happened. One, kind of Soul reached parity ish with Opus. So that was a, like, a big step up. But more importantly, the Codex app leapfrogged the Claude code, Claude app from a lot of people's perspective. It was easier to use. It was faster. It's cheaper. It just seems like an overall better experience functionally. Some of the screens change. So it's just an enjoyable app to use. So you had user experience, cost, and performance all kind of like leapfrog. This was probably six weeks ago. The other thing that Codex seems to have done well on that I haven't really had the time to explore, I actually don't know where I would explore it, but their memory seems to be much better. Like, it just seems to remember stuff. And so now I do things in skills. I do things in scheduled tasks. Like, I don't even know which one's a skill, which one. It's just sometimes it's just a chat. It's just like, just do this. And the memory on Codex seems to have gotten a lot better. And so here's where the real kind of separation. So all that was, you know, step one.
00:06:15 · Khe Hy
Step two, and this is where the Mac minis come in, is that Codex obviously works on your local machine, kind of like Cloud Code. And, you know, it has access to your browser and, you know, it can use your computer and it has access to your cookies and all that stuff. That was really good. But it required your computer to be always on, right? So that's where the Mac Minis came in, is that people basically repurposed their Mac Mini to just be codex hosts. And codex, you know, it can run really well on your computer, right? Like everything we got used to with Cloud Code. But the thing that's a big difference maker is that you can directly access all of your Claude I'm sorry all of your codec sessions on your physical machine through your phone. So you basically have a mirror of your physical computer on your phone. So let's say you're logged into a site like DealCloud. DealCloud doesn't have an MCP yet. I think they just added one. And it's like pipeline management for private equity, and some public investors use it. And you want to update your deal tagging, your pipeline. And you're doing that using the browser, right? You're logged in browser because you don't have an MCP, right? So you say like, Hey, Codex, go in and, you know, there's deal A. We passed on deal B. We sent an LOI. And deal C, we're still diligencing. Go update deal cloud. So what happens is that because it doesn't have an MCP, it would pull up your Chrome browser. You'd be logged into deal cloud, and it would start clicking around, doing what you need it to do. Obviously, you can only do that on your local machine because it has your logged in credential. Here's where Codex comes in. Now you can initiate that from your phone. You could just be like, hey, update my deal cloud. Or maybe, like, a more simple example would be like, into all my bookmarked tweets and summarize them for me, right? Something that requires your logged in machine. And that's why the Mac Minis come in, is because then you want that machine to be always on. You don't want if your laptop's closed or sleeping or so on. So that ecosystem, and for folks that are very into like personal agents, customizable agents that are always on and that are mobile first, that kind of, that's where all the OpenCloud people went. Because it hit almost every problem, scheduled tasks being
00:09:13 · Brett Caughran, Khe Hy
[00:09:13] S0: a [00:09:13] S1: big one, right? Like scheduled tasks on Claude code require your computer to be on. So it's kind of running locally. So basically, you're still running all these things locally. But because you can access them from your mobile phone, your iPhone, it doesn't feel like you're constrained anymore. Like, you could just be on the go just doing your work. So that's I said a lot there, but let me, I'll pause and let you react and ask questions. [00:09:44] S0: No, it's interesting. I mean, the one fundamental question I have is what does this mean for Wall Street people who've been working to build fluency in the tools? I mean, a core part AI training on Wall Street the last eighteen months has been, really last nine months has been Claude fluency. And even a core part of your practice has been helping firms get up to speed on Claude code and Claude co work. Yeah. What does that mean now? Like what's the sort of the next step as Claude gets decommissioned in some of these instances?