Breakeven Brothers

Alrighty, it's been a big day, Brad. How's it going?

Pretty good. Just got back from New York, so I did a week out there. Met some cool people at work, saw the *Book of Mormon* play. That was pretty funny. And yeah, it's been a good time in New York. A little busy, was out there with a few friends just hanging out, getting good food and good drinks. And now I'm back, and then tomorrow I'm off for more travel. So we'll be in Korea for about, I don't know, two weeks. So travel, travel, end of September, early October. But hanging in there. How about you?

Good, good. Yeah, um, busy. And obviously, we've got some big news on this episode. So this is episode 30 of our podcast, of course. The dirty 30. And the big news is that there was a big announcement today from OpenAI. It was their Dev Day. So we wanted to get on, not just because Brad's getting out of town here in a little bit, but also because of the big news that came out from OpenAI. We want to unpack everything that got released. It is all pretty fresh, so there might be details that we miss, but we did want to cover the big hitters as of today. Today is October 6th, as of recording this. Same day as the Dev Day. Everything's coming at us pretty fresh and, uh, yeah, looking forward to unpacking it all.

Yeah, they have a lot of stuff. I was just looking through it earlier today. I think they have a Dev Week, if I remember correctly, and I think that is like a multiple-day workshop, but I'm not 100% sure. Regardless, their opening keynote had a bunch of updates. They have apps in ChatGPT, I'm just looking at their announcement. Agent Kit for building, you know, production-grade agents. Sora 2, which is taking off with their new app. They have Codex, which we've seen and loved. GPT-5 Pro in the API. GPT Real-time Mini, so a smaller voice model. And then last, but not least, GPT Image 1 Mini. So, I guess a smaller version of their image generation model. So yeah, quite a lot to unpack. And then they have these high-level, Apple-like, you know, presentations where they say 4 million developers have built with OpenAI, which I'm not surprised. They're kind of the leading company with the first AI model and then just continuously doing well. And then 800 million-plus weekly ChatGPT users, which is pretty awesome. And then 6 billion tokens per minute on the API platform. So pretty impressive from them. I know they've hired a ton of smart people, and their uptime and reliability has been pretty solid recently. So yeah, a lot to unpack.

Yeah, well, it's funny. I'll put the tweet in there, but there's a person—not just a person, he's a very influential person in AI—Sunny Madra. He's the president of Groq, G-R-O-Q. And he tweeted about the announcement and some of the stats that OpenAI was putting out there. And basically, he estimates, he does some math—and again, I'll put it in the show notes—but he basically says that with those figures, that's about a billion dollars a month in revenue run rate, or about $12 billion ARR. I don't know, like, I don't have the math to kind of check him on any of that kind of stuff, but obviously a very knowledgeable person in the AI space with Groq. And crazy numbers to think about, a billion dollars a month in ARR. That's just wild.

Yeah, and they're still not profitable, I'm pretty damn sure, due to all their investments. But you can see a horizon in which they kind of explode at the very end once they have a ton of hooked ChatGPT users, a ton of hooked developers. Like, they're building on both fronts. I think from their Dev Day, I'm really excited that they're taking more chances. I think ChatGPT as a product, they can do really well with. But as they build out deeper integrations—like they had launched shopping, I think, within ChatGPT maybe like a week ago. With Dev Day, they have apps built into ChatGPT, so we can talk about that. But there's a lot of stuff moving in OpenAI. I think from the outside as a developer, I kind of grade them by their model performance, not necessarily their user-facing products. But I'm excited that they have a lot of smart people working on that part of the house, building out really good, sticky AI experiences that actually feel useful versus just a generic chat interface. You know, I built my own chat AI app, and it's really hard to make a good chat. And then when they add all these bits and pieces on top that make the experience that much better, it does kind of eliminate some of the startups that are out there today building on top of OpenAI. So OpenAI is a behemoth, but they do come out swinging, and they are building stuff, and they are continuously partnering with important people. And it kind of makes you wonder, like, what's next for them, given they released all this stuff today and probably have 5 to 10 things, you know, cooking in the background.

Yeah. Well, let's kind of compare, I guess, before we dive into the specific items that we mentioned, like the Apps SDK and stuff like that. I wanted to just compare the, I guess, reception and my own personal feeling. This is, you know, just my own anecdote, I guess. In the release of GPT-5 versus this kind of demo Dev Day. Like you said, it's different because, you know, a GPT-5 release is the release of a new model, whereas this is kind of more product-focused and more feature-rich. It's just different. But to me, when I watched the live stream today, I thought they knocked it out of the park. I thought they did a wonderful job. You know, we were a little bit critical, I think fairly, of the GPT-5 rollout and just kind of what was promised versus what was delivered. But for me, I was pleasantly surprised. And I think the night before, so last night, again, on October 5th, Greg Isenberg, who's a very, you know, influential person in this AI space, I saw a tweet from him saying that it's rumored that the Agent Builder will be dropping as part of this Dev Day from OpenAI. And so you and I exchanged DMs about it, and I was super excited, and I think it's a very logical next step for them. So yeah, I mean, top to bottom, you know, on all the different things that we'll cover, I was impressed.

So let's get into it. Let's start looking at the individual items. And again, this is pretty fresh. Like Brad and I saw the live stream and then went about our days and then tried to get caught up. But, you know, whenever something is this new, there are going to be details that we might miss. But just from a first glance, let's get into the Apps SDK because I think that was the first thing that they led with. And you know, for me, I think the way that they pitched this was like a built-in way to build apps within the ChatGPT ecosystem, and importantly, be able to monetize those apps as well, all within OpenAI's ChatGPT ecosystem. So I thought this was really cool. I think there are so many companies that are out there trying to build with OpenAI and build those apps. But deployment is always, I think, a challenge right now, but also, it's just so fragmented and there's such a wild frontier, I guess, in terms of app development. That having a place, if I had an app that I wanted to build and deploy, what better place to do it than in a ChatGPT-type ecosystem. And one thing I'll add before I finish here is, I'm going to kind of give this credit on the bingo card, because I think we had a bingo card item, it might have been yours, where it was like an App marketplace. I'll have to go back and look at the bingo card to see exactly how you phrased it, but this is what it kind of feels like to me. So what were your thoughts on the App SDK?

Yeah, I was just opening the docs and looking at all the details. I think it's really unique because you can build these first-class experiences within ChatGPT without leaving. And I think this actually becomes more of a platform. Where previously we had GPTs, where there's like a GPT store where you'd write some custom prompt, people could kind of pull those custom prompts into ChatGPT and use them. And I believe that was a store in the sense that OpenAI had paid out people who had a lot of popularity with certain GPTs. This feels like the next evolution of that, where we're getting built-in, quote-unquote, apps into ChatGPT's interface. So they're kind of demoing Spotify, for example, on their blog post, where you can ask Spotify to give you the latest songs from your favorite artists. And it just pulls up a kind of custom UI, I would say, within the ChatGPT app that's Spotify-specific. I thought, oh, that's pretty nice because, you know, I build mobile apps for a living. I know Apple has very specific guidelines on how things should work, how they should look, whether that's data or UI. And so I was pulling up their documentation, and they actually have a design guidelines page, which is very interesting. They outline the different experiences that OpenAI makes available with this new SDK and what they might be used for. So for example, you could show inline text, you could show an inline card, there are different layouts that describe how you should prompt a user, there are different interactions. So what it seems like to me is that they want to bring apps into the experience to keep you inside ChatGPT's chat interface. So instead of, you know, connecting with external things kind of like on the MCP side, this is getting you a first-class custom experience. I think this is a good idea for them because Meta has always wanted to own the hardware side to get this great integration. Apple and Google have built out iOS and Android and clearly have a large market share here. This is the ChatGPT app within the iOS App Store and Android App Store, so it's a little different, but it's kind of like a platform within a platform. I think they get a lot of control and a lot of say within creating this ecosystem. It's going to be interesting to see which apps make it first because with all things, you know, creating a platform, OpenAI is going to have to review your submission, make sure you're trustworthy, and not be filled with junk apps. I'm sure there are going to be apps that don't meet the bar and get rejected, or they make it and they have a bad experience and maybe some bad things there. So I think off the rip, it's very interesting and smart that they're creating a more holistic platform that's kind of tucked away within the ChatGPT app with custom design. And I think as a developer, I see this could be a pretty big opportunity to get something out there fast. Like any new, quote-unquote, platform, being first to market there could be pretty advantageous, whether you're an existing company or you're just creating a new app that's chat-specific. So honestly, I'm definitely going to take a deeper look at the developer guidelines because that seems like a great opportunity.

Yeah, yeah, it was super cool. They also demoed one other company, but we'll leave that off the podcast. But definitely encourage people to check it out. It's pretty cool. And um, yeah, it just makes sense and you can think about it from the perspective that this is early days and they already showcased some really cool feature sets. And I think people can maybe see the writing on the wall that so much more is going to keep being integrated within ChatGPT. And I feel like that's where OpenAI really has a leg up on the other models. I think Google is very competitive in that space though, because they're the next ones that have Google Drive. There are lots of companies that use Google Drive and all the other different products that Google has. So I'm excited—I'm even more excited now to see how Google swings back because it seems like that stickiness is like, you know, having a great model is great and that is important. But also having products that really get into the ecosystem and get adopted by your users, you know, meeting your users where they are is such an important concept that companies strive for. It's kind of a marketing 101 thing, right? So, yeah, I thought the Apps SDK was really cool. And the one thing I would say too on it is, I was looking at some of the examples that they put out there in their repo. And you know, it's really a nice evolution of I think how agents and agent development kits have come to be because I think before, you had these big—or not big, but these frameworks from the model providers. You had things like LangChain, which was kind of one of the early ones for setting up an agent or setting up a prompt and chaining different prompts together. And over time you've seen OpenAI release their SDK, Google release their SDK. Pydantic AI is really good as well. It's maybe not as well known, but it is, I've used it before and it's really nice. And when you look under the hood at what the, uh, excuse me, what the Apps SDK is doing, you can kind of see it build upon that. Like you can create these agents using the OpenAI SDK in this example, and then it kind of pairs it with this, um, you know, chat interface, like you said, like inline and stuff like that or cards. So it's just cool to see that linear—you can almost see that linear progression to get to this announcement today. So I thought that was really cool. Yeah, I'm excited. I'm curious to see what people build and I'm sure in a couple of weeks we'll hear about someone who's made a million dollars building an app on ChatGPT because they were the first ones there, right?

Yeah, I mean, it is kind of cool being featured within the ChatGPT app. I wonder—I'm looking at their blog post right now, and it does mention the SDK is out in beta with a developer preview. "We will open for app submission later this year." So it could be a while. I think as I look at the developer documentation, there are about 20 pages to go through, which is pretty hefty. I mean, if you think about it, this is like a whole new platform. It is kind of similar to others, but different in a large sense because you're talking about the design and OpenAI owning this end-to-end workflow. Of you working with, you know, building your app, and then you're deploying it to make it compatible, and then testing it, optimizing it, the App Store, all that. So it is a little bit interesting. I think very, very exciting. I'm surprised to see that there aren't more apps on it. I imagine, as a company, you'd want to be integrated. But it does drive less traffic to your website if ChatGPT exists as an app. So it's that kind of double-edged sword of, yes, I would like to enable my experience natively within ChatGPT, but no, I don't want to take away from the core experience. And then I have to maintain a whole new kind of platform, so to speak. So that can create a little bit of a technical burden there. So very interesting to see who's out in their first round. Interesting to see who will come out in their following rounds and if it will truly be open. I have a feeling that OpenAI kind of treats itself like an Airbnb or an Apple with pretty high design taste. And I wonder if they will hold a very strict bar for developers approaching this Apps SDK and see how that approval process goes. But yeah, I think there'll be a tweet that, you know, "I made millions off a ChatGPT app" by the end of the year once they open up submissions. I guarantee it.

Yeah. Well, and also too, I wonder if this will—and this might already be the case where the other competitors are working on a similar thing. And if you think about it, okay, you're a company and you have an application, but you want to build it into ChatGPT as well, just like let's take that Spotify example. Spotify, right? The music one. That's what they demoed, I think, today. So if you're Spotify, you've got to build it for ChatGPT, but then if Google has their own in Gemini, right, then you might need to build it for that app platform as well. So you could see, if you want to get the most coverage and be where your users are, you might need to build that once or twice with the different app SDKs if they all start doing that. So, yeah, it'll be interesting to see where that all leads. And like you said, the approval process and how that might differ from Google to OpenAI and so on.

Yeah, and I think I would just like to tie it back into their "Buy It in ChatGPT" product release. So this was actually announced September 29th, so not too long ago. And I was just looking at the blog post and they have an Agentic Commerce Protocol (ACP), such that ChatGPT can work with a merchant. And that merchant can expose the products and fulfillment options, you know, sales tax, et cetera, all that. So the whole checkout flow is encompassed by this Agentic Commerce Protocol, so they have integrations with Stripe, with Etsy, with Shopify—so not Spotify, but Shopify on the payment side. Such that you can ask, "Oh, I want to buy a new chair for my office." And I'm going to go ask ChatGPT, which does a great web search. It scrapes Reddit, scrapes Yelp, you know, whatever. Once you get all that data back, since this is such a trusted and reliable thing for people these days, having a "buy it now" button really pushes things forward—you know, you get the research done, you're ready to buy, now you can buy immediately. And so I think it's a genius move by them to get that working. It's cool they added an AI layer on top, the ACP, Agentic Commerce Protocol. And under the hood, it works with a few of these existing payment providers. So very, very cool. I think that will be even better with this Apps SDK where you can expose an experience, and that experience can then expose the "buy it now" feature. I think at the end of the day, a lot of these companies are either trying to get you to pay money or get you very, very close to that. And so building on top, one layer after another, for ChatGPT to pretty much take over everything. So really exciting for them. I'm glad OpenAI is focusing on a product experience and not only the core underlying technology because they have great advancements in both. And I think this one's a standout. I'm excited that they have it. And I think it unlocks a bright future for them with the ChatGPT interface and the new apps.

Yeah, totally. Cool. Okay, awesome. Let's move on to the next chunk that they announced today. And that to me was, I think they call it Agent Kit. And within the Agent Kit are things like Agent Builder and then also, um, I think it's called Chat Kit. And then, um, within that, you can do evals on, you know, your agent. So the, I guess, elevator pitch in terms of Agent Builder and Agent Kit is first, speaking about Agent Builder, it's more of a drag-and-drop type interface to build out agents all within, again, the OpenAI platform. So I was super excited about this, especially because myself, not having a technical background, a big problem with implementation of AI solutions is the difficulty and the learning curve in terms of understanding how AI works under the hood, behind the scenes. And being able to create workflows that are user-friendly and don't have a super big learning curve to create, so that way it can get implemented. So, the reason—and there are other things that are already out there in terms of agent builders. So, you know, when I was thinking about why I was excited for this, you know, when there's already Zapier out there, there's already Make, there's N8N. There are lots of different, quote-unquote, agent builders. But what I was really excited about with this was just the fact again that it's brought into that OpenAI ecosystem. I mean, I think almost every company, every enterprise out there, is going to have at least a ChatGPT enterprise license or a Google business enterprise license, right? So being able to have that just be part of what you have when you are signing up and using ChatGPT, I think, unlocks a lot of potential productivity gains. It's not some separate vendor that you have to go get onboarded with or it's not some separate charge that you're going to feel a bit more when you pull your card out and type it into Zapier to start a new subscription. And so I think it's really exciting just for unlocking it for more people, not because the Agent Builder itself is incredible. I've been working with it this evening, and I think it's really cool. There are some pros and cons that we can get into, but just the introduction of being able to build agents to such a bigger audience than before, I think is what I'm most excited about.

Yeah, I think when I saw the announcement, I immediately thought, "Oh, Ben's going to have fun with this," because he's told me a lot about N8N and I've never made the hurdle to go try it. There are a lot of tools I do try. That one I just never made the jump. I think as you described, bringing this agentic workflow builder that resembles other platforms into the first-class OpenAI experience, I think that will bridge the gap of, you know, I'm already a paying developer here. I already pay for their, uh, you know, ChatGPT $20 a month subscription. I use Codex all the time. So I use Claude Code and Codex at the same time. But at this point, I'm always interested in trying their first-party product offerings. And I think this one falls under creating an agent. And I like how they outlined it with build, deploy, and optimize, because again, they do have the evals. They've had that for almost a year now, probably. They do have their OpenAI Agents SDK in Python, and I think they probably have other languages now. But they have all these foundational building blocks, and I'm glad they are tying this all together in a new guide. So they say, you know, use the Agent Builder for a visual canvas to drag and drop and bring everything together, connect to our LLMs—you know, OpenAI's great models. Equip your agent with tools, you know, MCP servers, et cetera. Provide knowledge and memory, which is kind of persistent data that you want your agent to use. And then having this control flow. So this is that visual drag-and-drop, looping, branching, those sorts of things to make sure that agents can work together. I haven't used it. I know Ben has, at least a little bit today on the first day it came out, but I think the overwhelming feeling that I have is if people weren't going to use N8N, I think this will tee them up to actually use and build these workflows. And they have pretty good—from what I can see from their documentation—pretty good ways to actually deploy these things, where you can bring them front and center, you know, within a matter of hours, probably, to an end-to-end experience. And they already have the Google Drive connector, all these other bits and pieces that have powered ChatGPT are now being exposed with this new agent setup. So really exciting. I'm definitely going to have to play with it pretty soon. I think they again, they called it Agent Kit. It's a whole bunch of different things, whether they're slightly existing with a wrapper or brand new with this Agent Builder interface. I think all of it is pretty cool to see tied together and they definitely have a vision for how to do this. And we'll have to see over the next few days what people build because I think this is out now and you can try it now, which is really cool. And again, for the Apps SDK, you can't really get in until later. So it shows you what you can do. You can't really get there yet, but this one is like, "Hey, come try it and build some cool stuff."

Yeah, it's super cool. Um, and, you know, I've been toying with it. I have a little demo I'm probably going to record and put on the Augmentic Accounting channel that I have. But um, because I was working on a PDF extractor workflow, something pretty simple in terms of the actual process. But just to get my feet wet in terms of how it's being used and all that kind of stuff. But yeah, it's really cool. One of the cool things that they demoed with this was, to go to your point about deployment, the presenter, I think it's Christina Huang, if I remember that. But so she said, "I'm going to set a timer for like eight minutes," I think it was, "and I'm going to build an agent and deploy it all within that time." And so they went through the steps and everything like that, and they nailed it. Again, they did a great presentation. So I'll probably say that a few more times on this podcast. But yeah, deployment is such a huge hurdle, especially again for non-technical folks like myself. That seems to always be that last mile where it's like, I have this thing, it works great locally. Now I've got to, you know, if I'm speaking about web applications, I've got to go into a VPS and I've got to do the reverse proxy, like all that stuff. Just miss me with that. But if you can just deploy that with the click of a button, which is what the presenter did, that's huge. Like that is so nice. So again, still early days. Like when I was toying with it, I was getting some errors. Things weren't consistent. Like GPT-5 as a model for the agent just wouldn't work. So I've been using 4.1. But um, you know, they have the built-in tools like file search is what I've been using. But also they have the best, in my opinion, implementation of guardrails where you can prevent malicious actors with like hallucinations or jailbreaking or PII. So they have guardrail tools that are built in. And then of course, MCP is a huge one that I think is still in early days, but being able to, like you said, connect to your Google Drive and read documents as part of the agent workflow. Huge. I think it's massive. So yeah, more to come and um, yeah, once I have something solid to show, I'll put it on this channel too, or maybe just do a quick screen share because it's cool to see it in action, again, especially for non-developers. I know for developers it's like, "Oh, that's already there. We've already seen that kind of stuff." But for non-developers to have the power in your hands, it's a good feeling.

Yeah, and for folks who want to dive deeper into the Dev Day, they have an opening keynote which is about 50 minutes. They walk through the high-level bits and pieces of a lot of these announcements. Some of them didn't make it to the full keynote, but I think the meat of it is there. So it's a very good presentation, kind of Apple-like so to speak, showcasing that live demo in 8 minutes is really cool. They have other companies and first-party integrations that are featured there. So we'll link it in the show notes, but a pretty compelling presentation. And I think all the other AI labs are going to be a little bit nervous. Last time we recorded, it was last week and actually Anthropic had not released Sonnet 4.5. They have since then, and I think that gave them the spotlight for about a week. Now OpenAI came out swinging. This was a scheduled event, so I think people knew this was going to be a big one. But they did pretty well. I think the feedback that I've seen online from Twitter was pretty positive. And I think the opportunities are there—how things end up shaking out in terms of rough edges or how to get approved in the app space for their SDK, who's to say. But I think the initial reaction is definitely positive. And kind of moving on from their announcement list, we had talked about Sora. So Sora 2 is in the API. I think it's not extremely expensive, but it's not cheap. So you obviously have to have a use case that you can make money off of creating videos outside of just something to joke with. But I think for the creative professionals out there, it's going to be a huge deal. I think for the rest of us, you know, I don't use it too much. I didn't use it too much when the first versions came out. I even think with some of the Google products that do video generation as well, I do them for fun, not for anything to make money. So for the folks that are in that space, it's going to be excellent. For people outside of it, it's more of a toy, I think, just to mess around and have fun with. So glad they released it, but yeah, no crazy reaction from my end.

Yeah, I think it's super cool. I mean, super cool as in Sora 2. Again, for it being available on the API, I haven't used it myself. But it's a super powerful model. I mean, some of the videos that I've seen are pretty incredible. And it feels like we're kind of blurring that line of like, is this AI or is this real? Like some of the videos that they've shown. And I think in the demo, when they were speaking about it, they presented a cool use case where it was, um, like a marketing video, like a commercial, basically, of I think a person doing some shopping for furniture. And it looked really good. Like, you know, so if you think about the current state where if you want to shoot a commercial like that, you've got to hire an actor or an actress, you've got to get expensive cameras, you've got to get a setting, you know, maybe it's a big nice building or you're down at the beach. So much cost, and the fact that you can create an almost-real video for a fraction of the cost. I'm not sure exactly what Sora's pricing is, but it's certainly a hundredth, if not less, of the cost of that alternative of creating a professional ad. So I'm sure it's not there yet, but you can kind of see it's only going to get better. And so, you know, I'm not going to be surprised if we start seeing car ads, for example, where it shows a Porsche driving on the side of a road that looks all nice and it's like, that could have been from Sora. You know, are we able to tell or not?

It is, it is really good. I am slightly scared too. I think the hands are still a big tell. If you see anything that might be AI-generated, check the hands or if it has unnatural movements. There was an OpenAI release video in which they talked about something, I can't remember what it was, but there was big speculation that the video they had shown to the public was Sora-generated because of these very nuanced, odd human-like movements, I think in the actress. And so people were speculating, hey, this is probably like a Sora 2 video. And I don't know if that ever came out to be true, but I thought it was very interesting as all the videos that OpenAI showcases on their marketing page, people kind of have this in the back of their head of, you know, is this Sora? If it is, it's really impressive, but it almost feels kind of weird in a way. I think when Sam and Jony Ive had their video, which was many minutes, people had wondered, "Oh, is that Sora as well?" because of just some weird elements about the video that people are very particular about identifying and saying this looks a little weird. So yeah, we'll have to see if that makes it a little bit easier. I think the pricing is good, like you mentioned, but I think for people just trying to mess around, it's not cheap. You can't mess around forever. So cool release. I'm glad they added it. And again, I think for the creative professional, it'll actually be a huge, huge deal. For the rest of us, definitely cool.

Yeah. And I think the only other API release before we get into the last, I think, big chunk, um, is that they released GPT-5 Pro for the API, I think.

Yeah, that one's good. I think the pricing is really expensive. And in my mind, it becomes a little bit difficult to choose when to reach for GPT-5 Pro. I think if you pay for ChatGPT Pro or Plus, you only get so many queries to GPT-5 Pro. I think it was maybe like 15 or 20 per month, which is extremely low. Maybe they've updated in the past few weeks, but originally it was a very low limit. So you could tell that it costs OpenAI a lot of money. When they released the API pricing today, it also costs a lot of damn money. So you can tell that it's not cheap to run, but it's really good at what it does. But we kind of end up at this point where—and I know you have told me about this—like have we hit the limit where the models are just good enough? And to me, I don't think we're there yet in terms of code, but when you have to face the question of, should I use GPT-5 Pro or not? I'm personally not extremely familiar with how far it could go and how much better I could get code generation if I went to GPT-5 Pro versus not. And sadly, I don't have a good answer to that question because I think I would use it more if I knew. But given the steep price, I don't want to run through like 20 examples of here's GPT-5 versus GPT-5 Pro, because, you know, if it's not significantly different, I just wasted money. And again, I'm paying for Anthropic's Claude Code Max plan, which is $200 a month, and then I pay for ChatGPT $20 a month. So both of those give me their agentic CLI coding experience, but GPT-5 Pro is not included in OpenAI's $20 a month. So I can't even test this unless I go through the API. And again, API pricing is really expensive. So long story short, it's really, really cool that this is out. I think as a developer who's used a lot of AI tools, I'm completely unaware of when this would actually make a difference in terms of my day-to-day coding. I hope they come out with some guide or something to help me. Like if Codex, the CLI, could intelligently choose GPT-5 Pro for me on the hardest of hardest of hardest of problems, because, you know, like we had talked about, they are moving to this dynamic thinking mode where you choose a model, then that model internally says, "Should I think a lot or should I not think a lot for this query?" And GPT-5 Pro, I think, extends leaps and bounds above GPT-5 high, which is their high thinking mode. So I would personally trust OpenAI if they thought my query was really complicated and needed GPT-5 Pro. But I think their pricing doesn't really align with it where I pay $20 a month, I'm writing a lot of code with their tools, but GPT-5 Pro is so expensive they can't even offer a sliver of that within their $20 a month. So it's in a weird spot where you really need a hard use case. And to me, I have not found that use case where it matches because I write a lot of code with AI. So, yeah, I don't know. Maybe that's just my reaction, but I think people are excited about it from what I've seen online, but personally, again, I don't have the right understanding to know when it would be worth it to spend that extra money to get that higher grade of intelligence.

Yeah, yeah, same here. I mean, I, like I said before, and like you said, I don't think my personal use case would ever really warrant GPT-5 Pro. Maybe one day I'm building rockets, but for now I'm not. So, um, yeah, it's good enough for me. Um, but yeah, I mean, if you're someone who needs that, it's available through the API. So there you go.

Brad, the last thing I wanted to touch on in terms of the announcements—and to me, this was interesting because it was the coolest demo in my opinion, but it was the one I'm probably least close to—and that was some of the stuff that they were showing with Codex. So I'll explain a little bit about the demo and I thought it was super cool. Um, but I will be the first to say this is the area I've spent the least amount of time on. Um, so Brad, I'm not sure if you've had a similar, you know, not been able to spend enough time on it yet, but, um, in short, what they did was they had someone from OpenAI come up on stage and they have cameras, right? You're on a stage and they have lighting and they have cameras all around the room, right? And so the presenter, I wish I got his name, um, but he was saying, "I'm going to have Codex write me a program where I can control the lights and the camera that is in this presentation room." And he prompted Codex, and a couple of the things that stood out to me were, one, I think he was saying, "Hey, um, they're like Sony a7R cameras or something like that." So he prompted Codex and said, "I need you to write me code that can let me control the Sony a7R cameras and build it into an app." And so he wrote it into Codex. Codex went and did its thing. Um, it was really—they emphasized MCP a lot all day, actually, because they mentioned it in the Apps SDK, they mentioned it in the Agent Builder, and then they mentioned it a lot during the Codex thing. So I think it somehow used MCP to go get the docs from Sony on how to control the cameras. And it built an app. And so he live demoed it, and he showed the camera looking out at the crowd. And then he had his mouse and he was able to move the camera around. And you could see the camera pan to the left, pan to the right based on where he was pointing with his mouse. And, you know, of course, people were clapping, right? Because it's a live demo. So you never know how those go. We just talked about it with Meta a couple of weeks ago where it didn't go that well. Um, but then he leveled it up even further and was like, "I'm going to see if I can do it with an Xbox 360 controller." So he grabbed an Xbox 360 controller and he went to Codex and said, "Hey, that's great, but I want to control it with an Xbox 360 controller. Go ahead and do that." And then, you know, it went off, did its thing. And then he turned on the Xbox 360 controller, and then he was moving it around with the—I have it right here—moving around with the joysticks. And he didn't tell it to make the joysticks be the ones that move it. You know, he didn't say any of that. Um, but it just again, figured it out through its own intelligence that that was probably a good call to make. So there's more, but I'll pause there because any initial reactions from that? Again, he didn't write a single line of code. It was just like two or three prompts from him into Codex that got us to that point so far.

Yeah, I was just looking at it while you were describing it, skimming through. It looks like the presenter is Romain Huet, Head of Developer Experience at OpenAI. So yeah, pretty cool demo. I think when I look at Codex being used, especially for a new project, which it sounds like this was, I think it's a little bit easier just to be completely honest. I think building something new using agentic coding tools brings the code with no existing human interaction. Like the AI can create whatever it wants because there's nothing existing and create it in whatever way it wants. But I think modifying existing code is a much harder task, which Codex does do well at. But I think the live demo, as I look at it, is really impressive. Again, hats off to them for building on stage. I think Codex has come a long way where there's been a trend, like we talked about last episode, of shifting from Claude Code to Codex based on various Anthropic missteps and OpenAI spending more time there. So I'm glad they highlighted it. I think they announced maybe two or three new features in Codex, like a Slack integration. Supposedly enterprise controls. They had the Slack integration for at least a couple of months now, I think. I think they made it GA, so maybe it wasn't super stable. So now it's generally available. So yeah, a lot of things within Codex were more like, "Please come use it. We've spent a lot of time on it," and that becomes a recurring theme. Because previously they didn't spend a lot of time on Codex, the CLI. Now that they saw Claude Code's success, I think they're spending a lot more time there. So there are new admin tools, the SDK is now available, whereas Claude Code has the Claude SDK for agents. So it's like all these companies are creating this agentic harness where they make it easy to make a reliable agent based on all their testing. So OpenAI has an agent SDK, but they call it the Codex SDK, which is kind of interesting because it's different from their agent SDK. And then, you know, again, Anthropic released the Claude Code SDK essentially so you can build an agent. But overall, I think I'm pretty impressed with Codex, and I actually do wield Codex quite a bit now towards my hardest problems because GPT-5 does a good job. It just takes so damn long for the models to think about it and figure it out. So overall, I think the Codex demo was impressive. In terms of feature set, it seems a bit more focused on enterprise, which I think is a good idea. Because like you mentioned, everyone has a ChatGPT enterprise subscription for any serious company. If they can just tack on Codex as, "Hey, now here's a CLI tool to write more code, and it has the security bells and whistles and controls," companies will easily adopt that. I think people are looking for that right now today. So, smart for them to focus on that and make sure they cover the ground so that this gets adopted more and more.

Yeah. And so, agree on all those points. The part where I wanted to continue, which I thought was really cool—and again, just a clever way to show something. I think the technology is amazing. Um, they're definitely not the only ones in this space, like you said, but the way they did it, I thought it was really cool. Because once he had the Xbox controller, he then switched it into like voice-to-text mode where he was speaking with an agent, I guess. And he said, "Hey, agent, also give me some code to control the lighting." So before we were controlling the camera, and then he's like, "Okay, I can go ahead and control the lighting," and I think it was through MCP, don't quote me on that. Um, but then, so once he had the camera, once he had the lighting, he was like, "Hey, what do you see in this camera right now?" And it was like, "Oh, I see an audience of people that are ready to learn," whatever, blah, blah, blah. And then, he's like, "Go ahead and turn the lighting onto the audience." And all the lights in the studio—and they had a great shot where you could see the lights above the speaker—they turned and pivoted towards the audience, and then, of course, on the camera, the audience lit up with the lighting. Um, and then the last cool thing that they did was, he said, "Hey, get a list of all the attendees and display them in rolling credits, like a movie at the end of a movie." And so everyone who was there as a developer during the Dev Day saw their name pop up on the screen, and it was like rolling credits at the end of a movie, which again, I thought was just clever and interactive and engaging. So all that, again, from some prompts and some speech—like he was just using his voice. Just incredible. Like I could not imagine—and again, I'll keep prefacing this, but I think it is worth saying over and over again—is like, from a non-developer perspective like myself, to be able to see someone do that. And he said before he did the demo, he's like, "I don't know..." like if the Sony—and he might have been selling this, just to be clear, I'm not that gullible. But he's like, "I'm not sure how to even do this with the Sony cameras, but let's give it a shot." So like, that kind of curiosity that lets you go and try something and build it, even if you don't necessarily understand every single nuance in that codebase, and to see it come to life like that. Like that's what building's about, right? So, um, I thought it was really cool. Really, really well done.

Yeah, I've been preaching that so much recently of just pick up a coding tool, an AI coding tool to be specific, and just build cool stuff. Like I have thoroughly enjoyed scratching my own itch more so than ever because I've always been in that camp to build stuff for myself. But with these tools, yeah, it really becomes whatever your imagination is, you can get pretty darn close to that. I think that was what this demo revealed is, yeah, you don't have to know a ton about it. If you're even slightly technical or non-technical, you can get pretty far because these models have gotten so darn good. So really impressed with OpenAI's overall Dev Day. I think last year they also released a bunch of stuff during Dev Day. I think the competition is fierce right now, absolutely fierce, where any of these releases have high scrutiny. And it's very much a neck-and-neck game. And I think this puts OpenAI in a really good spot, kind of closing out the year. Again, we had talked about Google having a release week in December, intelligently waiting a gap from OpenAI. Maybe Google's changing their plans right now. You know, what they see is maybe they have successors to Agent Kit, you know, Apps in Gemini, who knows. But all this stuff is exciting nonetheless. I think over the next coming days and weeks, we'll really see how this evolves and how people pick it up. Because there is a lot of material to go through here in Apps in ChatGPT alone, and then Agent Kit as well. Like all this stuff is kind of bleeding-edge new. And again, it hasn't been enough time for people to really work with it and release their thoughts. And what we're giving you today is our opinions and what we see from their documentation and announcements. But I think it's going to be a really exciting period for people to pick up these new technologies. And OpenAI is going to spend a lot of time hand-holding initially, I bet, across all their forums and Twitter, to get people equipped. And then once people are well-equipped with that, they'll be sharing more knowledge so that the rest of us can learn and grow. So yeah, really exciting stuff. Hats off to OpenAI. I think they did a fantastic job.

Yeah, yeah. Agreed. Well said. Cool. All right, should we wrap it up there with some bookmarks?

Yeah, yeah. I'm in for that. I think, uh, I won't use this as my bookmark, but Ben sent me a really funny tweet from this guy named Keitsey and he takes a selfie of himself in an empty office space with a bunch of chairs. And he goes, "I'm having a meeting with all the people who have done something useful with MCP." So it's himself and a bunch of empty chairs. Obviously, the joke is no one has done anything useful with MCP. I think from this Dev Day, we've clearly seen that MCP is a focus for OpenAI. And they were kind of late adopters to adding MCP server support to ChatGPT as a client. Now that they've done this and their OpenAI Agents SDK has MCP and their ChatGPT apps have MCP. Like all this stuff is revolving around MCP, and I think that protocol is being developed and refined. We're getting to a point where you can't really ignore it. There's been some Twitter discourse of, should I use a CLI tool or an MCP to get things done? I think the official system-to-system way when we're talking about a big company to a big company is definitely MCP, with MCP servers powering that technology. So while I think this tweet is funny, it definitely highlights the point that MCP servers are only as useful as they are broadly adopted. And at the current moment, it feels very developer-focused. And I think we're still getting to the point where average users can connect things and do things. And again, part of the Agents SDK or Agent Kit is bringing that reality of MCPs closer to the doers. Once we get to that point, I think a lot more people will be creating MCP servers because they want to enable this agent-building workflow. So long story short, I think MCPs have had a very, very long road. From Anthropic releasing it almost a year ago to where we are today. And it's just hitting that inflection point where I think we'll get a lot more focus on companies building out their MCP servers to power all these experiences we just saw today.

Yeah, well, it's almost like that's a connector, essentially. Like when I think about MCP, it's obviously technically more nuanced than that. But I feel like it's just a connector to bring apps together. It's almost like the connector was there before the agents were there, you know? And it kind of hung out there for a bit. Now that the agents are there, the connector's there. Like it's already all set up. And a lot of times that happens in reverse. But, um, yeah, I agree.

My, uh, my bookmark is, um, it's just a funny one. Another thing I sent Brad. Um, but, uh, so this is from PJ Ace on Twitter. And it's a Sora 2 video of Sam Altman stealing art from Miyazaki, who's the, I don't know if he's the director, founder, but of Studio Ghibli. Um, which I actually just watched a Studio Ghibli movie for the first time with my kids two days ago. Or, yeah, no, yesterday. Yesterday, not two days ago. Um, and so, but it's funny because, you know, there's all this controversy, and I think Studio Ghibli... um, is it Ghibli or Ghibli? Do you know? I guess I don't want to butcher that live on the podcast. Is it Ghibli or Ghibli?

I think it's Ghibli.

Ghibli? Whatever, you know what I mean. Um, they're really cute anime animations. Um, but in any case, when people first started using the ChatGPT image generator, they were making, you know, Studio Ghibli-style pictures and stuff like that, which was a murky area in terms of legality. I think the Studio Ghibli people or representatives were kind of like, "Hey, that's basically theft." They were anti-it. Um, so someone made this video on Sora of Sam Altman going into the Studio Ghibli office and stealing the art. Just kind of funny. And it looks so realistic. Like it's crazy.

Yeah, I just—that does look really real. So, pretty impressive. I wonder how much that costs for 10 seconds.

Yeah, what do you think? Like if you had to ballpark, I have no idea on pricing. What do you think it would cost if you had to just take a swing at it?

My guess is probably like a dollar or two. Maybe slightly more. My guess is like two to four dollars.

Okay. Yeah. Who knows if that's right though. You know better than me. So that's for sure. Cool. All right, well, good stuff, Brad. That was a lot. We didn't even get into a couple things like Chat Kit, but we can save that for another episode because there was a lot to unpack there. And I think we covered the big hitters.

Yeah, yeah. Good stuff. Good job by, uh, good job by OpenAI. Way to go and we'll see who's next on deck to deliver something big.

Yeah, awesome. Cool. We'll see you next time.

See you next time.

Thank you for listening to the Breakeven Brothers podcast. If you enjoyed the episode, please leave us a five-star review on Spotify, Apple Podcasts, or wherever else you may be listening from. Also, be sure to subscribe to our show and YouTube channel so you never miss an episode. Thanks, and take care. All views and opinions by Bradley and Bennett are solely their own and unaffiliated with any external parties.