Limitless Podcast - OpenAI DevDay 2025 Explained in 20 Minutes

Episode Date: October 7, 2025

Covering highlights from OpenAI’s DevDay, including ChatGPT's surge to 800 million weekly users. We discuss the new Apps SDK for seamless app integration and the AgentKit, allowing users to create AI agents without code. We also give takes on the implications for user engagement and preview the upcoming GPT-5 Plus API.------💫 LIMITLESS | SUBSCRIBE & FOLLOWhttps://pod.link/1813210890https://x.com/LimitlessFT------TIMESTAMPS0:00 OpenAI's DevDay Announcements4:28 ChatGPT's App Store Evolution8:41 AgentKit13:13 AI Agent Development15:10 Competition17:00 Reviewing DevDay Overall19:27 Closing Remarks------RESOURCESJosh: https://x.com/Josh_KaleEjaaz: https://x.com/cryptopunk7213------Not financial or tax advice. See our investment disclosures here:https://www.bankless.com/disclosures⁠ 

OpenAI just finished their Dev Day, live from San Francisco where they announced a bunch of cool new developer tools that will be accessible to everybody today, but also for people like us who are users of them very, very soon. So there was a series of announcements. We're going to get right into it. We're going to try to cover everything in 20 minutes. So if you stick with us, you will know everything about everything that Open AI just announced today, starting with their agent SDK, right? EJS, can you walk us through what I guess your first reactions maybe to what you just saw before we even get into everything? Well, my first reaction is probably your reaction to seeing this image right here, Josh. Now, there are a lot of big numbers on this, but I think the thing that kind of stood out to both

of us actually, we were discussing it before we started recording this was 800 million weekly chat GPT users. Do you remember when this was 500 million quoted like three months ago? They were the fastest to get to 1 million users, and we thought that was incredible. They did it in like five days. They're getting close to a billion. It's remarkable, the progress that they're making and how fast and how big they're getting. So you mentioned something, well, you basically, said apps inside chat GPT. And I think you were referring to one of three major announcements that they made on this live stream, which was the apps SDK, which basically allows you to connect your

app through chat GPT such that any one of their 800 million weekly active users can get access to your app. Congratulations. Open AI has officially become the welcome mat to the internet. Formally, it was Google. And now if you want your app to have access to the 100. of millions and billions of users on the internet, you probably kind of want to go through Open AI. What's cool about this new update, Josh, is it has full access to MCP, which is an open protocol. And you and I have gone back and forth about like closed versus open. MCP basically allows any app to connect to Open AI or to any AI model. And also importantly allows you to connect any of your data to it as well. And the reason why this is important is it's

one thing allowing chat GPT access to a tool, which is what we've seen OpenAI enable about seven months ago, right? And it didn't really get anywhere. Now, if you have access to MCP and a direct line to the chat GPT store, you can start to see this as a bit of an evolution. So they had an example here, Josh, where, you know, you could be talking to chat GPT and saying you want to book hotels and it selects Expedia or booking.com or something like that. They have something where, you know, you can access Canva to create like your own poster. And the major step that they've had from enabling GPT plugins to what they have today is chat GPD kind of decides what app it needs to pick for you instead of you picking it for

it, right? And it kind of has the context throughout the conversation without needing to remind it like, hey, you need to use this information to do A, B, and C. So for the people who are listening to this that are not developers like myself, this might not seem super interesting. and I want to kind of frame this in a way that is interesting because us as users are going to see a lot of these interesting downstream effects kind of as they go. This is the early seeding of it. That's kind of what this event is today.

It's for developers by developers. And there's just a lot of these complicated terms like an SDK and an API. But the reality of what they're doing now and the reality of this first batch of news that we're talking about is chat Chbpt is creating an app store for the chat Chepti application is an easy way of framing this. And basically what's happening is normally in order to use like we're seeing on screen, an app like Expedia, you go to Expedia.com, you open up your mobile app on your phone, you go on the desktop app on your computer. But it's not natively integrated into the chat GPT experience. So a few of the things that they showed, the few examples like we're seeing right here, is now when you want to book a flight through Expedia, you don't actually have to leave the OpenAI

ChatGBTGPT application. You can do it directly from within the application. And that's kind of the novel breakthrough with this SDK that they're releasing to everybody, is now you can engage with applications that you use on a daily basis, but they're kind of enhanced and augmented by inline AI. So if you want to book your flight now, chat ChbT has the context of your calendar. It has the context of your emails. It has the context of all of the chatting that you've been doing with it. It can go and it can seek out those flights, for example, and suggest which one is to book. And they're doing that with a whole bunch of different examples. They had one with Zillow where you can actually browse homes. And then, for example, if you find a home in a relevant area,

you could ask, well, how is the school system there? And how close is this? the nearest grocery store and you could get additional context that you never really could prior to this integration. So, EJA, for me, one of the more interesting parts of this entire presentation that they did was just how they were trying to approach the problem of solving it directly within their app instead of going to the browser. I think with perplexity, they created their own browser. With Anthropic, they created a browser extension where it kind of embeds itself into the browser. ChatGBT is saying, no, no, no, no, no. We have 800 million weekly active users, you're coming to us.

And you're going to give us to your tools. And we're going to augment the experience for our users. But please come to us and we're going to make that experience great. I don't think that this is a surprise, Josh, because I don't know if this is a hot take at all. But I don't know if it matters if it's a browser or directly in a chat interface like chat GPT here. I think it's become a business development game, a partnership strategy game, AKA, if Open AI can partner with all the biggest companies and shops in the world, e-commerce

sites in the world, then they will probably likely provide the best shopping experience, the best agent experience, the best app experience. And I think there's good reason for them to pursue something like this, right? I actually mentioned on the episode that went out today on Open AI's personalization and memory features, if you guys should check out that episode, if you haven't seen it already, that personalization and memory are open AI's biggest modes.

It makes the product incredibly sticky because chat GPT suddenly knows everything about you. It goes from feeling like a tool to feeling like a friend. And this is kind of the similar jump that they've made with apps. If I am using chat GPT and it has all the context on me

and it understands the conversation and I've just had about it, for example, I don't know what fridge I want to buy for my new house. So I'm obviously going to chat GBT. I'm checking out what the latest brands are, what the better price points are within my budget. And then I find an answer, right? But if chat GPD can then go out and buy that for me, I don't want to leave the site.

Like, why would I want to go to that? Why do I want to scroll and click open another tab, sign in via Gmail or whatever that might be? There's so many reasons which I will be lazy for it. And this makes chat chad chit be incredibly attractive. Yeah, I think it's a big breakthrough for why people are going to be excited to use chat chabit. I'm personally excited to try these features in line because it kind of augments. the whole experience. Like every time use the internet, it's now enhanced by all of the

context that you've given it. So we frequently ask, and you mentioned on the episode that dropped today, is memory going to be portable? And maybe the answer is it doesn't matter because all of the stuff we're going to be doing is actually in the chat GPT app itself. So that could be an interesting way this plays out. I also just want to say that

if I'm being honest, I'm kind of unimpressed by this product launch. And I have to get it off my chest because about eight months ago, OpenAI released what they called the GBT App Store, some version of an app store. And back then, it was a plug-in interface where they advertised, hey, you can now plug in a Rezi or Open Table or a Zapio or whatever that might be. And suddenly you get access to these apps, to these tools. But they didn't call them apps. They call them tools. And they said, you can now at this tool in chat GPT, and it'll use

it. And that kind of ended up flunking, Josh. No one actually spoke about it in the last eight months since it launched. And this doesn't seem to dissimilar. In fact, they're using a bunch of the same partners that they started. So I'm optimistic in the sense that this is an evolution from that, where, you know, it comes across as more natural. The products get integrated into my conversation. I don't need to tell it what to do. But I'm still not wowed by it. And maybe that's just me being a duma, but I don't really see this as the chat GPT app store that this tweet that we have on our screen resembles or depicts it as. I see it more as something before that where it's just for devs only, and I haven't really seen the magic.

Yeah, we'll see. It's funny. I downloaded a total of zero apps in that first version, EJez. I never used one. I think it was totally useless. Well, it seems like there's a possibility that the incentive structure has been flipped now because of the amount of users that chat ChupT has, where there's a very strong incentive to come and build for the platform, and there's a very strong incentive for Chad ChupT to offer these because the more value they add, the less you have to leave the app. So we'll see. I guess it's one of those things we'll see. That was one of the two major announcements that they shared. The second is there. agent kit. So, EJES, I know you love your AI agents. Can you please tell us what did they just

announce in the world of agent generation? So I love AI agents. I haven't really loved how they've materialized in the real world just yet. And my biggest gripe with AI agents is they just haven't been able to do stuff. So when Open AI today announced something they're called, calling the agent kit, I got really excited. They advertised it, Sam advertised it as taking agents from prototype to production. And there's three core capabilities that they announced. One called the agent builder. The other called the chat kit,

which is kind of like a chat interface that you can embed in any app or agent that you build, and evals for agents. So basically measuring how good your agent is. But the most exciting one of those three is the agent builder, Josh, which is a no-code interface that anyone, including you or I,

can use to spin up an agent in a matter of minutes. And that's actually what they did live on stream. And what's cool about this is have you ever used kind of like a graph builder software, Josh? I know you use it to kind of like create like design flows. It looks pretty much identical to that. So you can kind of like spin up a box,

literally a shape of a box, and that represents your agent. You can then click on that agent and give it a personality, explain what it needs to do. And that uses chat GPT on the back end and it automatically creates some code and creates that agent for you.

And you can kind of like move that agent across to maybe work with another agent that you've created, and it kind of becomes this kind of gamified system of making an agent. And this agent builder, they describe it as a canvas to build agents. And it builds on top of the existing APIs that they have, as well as a few new ones. And then Chat Kit is this embeddable chat interface for your agent. So if you imagine, Josh, you and I create an agent that goes out and there's a bunch of research for the next couple of episodes that we're going to do for this week, right?

So imagine it gets all this information. It pulls all these article links and it does all the research. It gives us some nice, concise summaries. But we don't necessarily want to read all those summaries. So maybe we just want to talk to the agent itself. We can just slap in an I message-like interface and talk to it. And most importantly, not only can you and I talk to that agent, but other agents can talk to it. So the main takeaway for me here, and I want to get into your thoughts in a second, is we've gone from needing to know and write code to

create these automated programs that do stuff for us to not just being able to write in plain English and have the thing do the thing. Yeah, that was the thing that stood out most to me during this section is normally when I watch these developer days, I feel mostly excluded because, you know, I don't really write that much code. I'm not super familiar with how all this stuff works. But as I was watching the demos of how this agent builder worked, it was very similar to tools I've used before, like you mentioned. Like I use Excaladra because I love to create these visual design graphics. Even DaVinci Resolve, the way that we edit a lot of these podcasts, it has this kind of visual node graph where you just drag things and you connect things and it's very much like a playground and they

turned agent building into that type of playground where it's all in plain English. It's all very visual. It was really exciting because I think a lot of people are soon going to try to experiment with being a creator themselves and building things where you genuinely don't need to use code. And now if I need to have something built, I can just try to make my own agent. And the lady in the example who is, I think her name was Christina, she was on stage. She built an entire agent in eight minutes using zero lines of code and it worked amazing. So for that, I'm really excited. We have this take on the screen from Greg Eisenberg. It says agent engineering is the new prompt engineering, which very much feels true. Again, the hottest new coding language is English. If you can prompt

an agent to do things, it's like a prompting a large language model on steroids. You can kind of compound, you could use all these tools together. It creates this really fun opportunity. So I think the downstream effects of this are going to be worth observing. I guess we'll follow along and see what people are actually building with this because it left me feeling optimistic that like, hey, maybe we actually can try to build some tools ourselves. Do you agree, disagree? I agree. And I think that we might see some of the best examples come from the people that combine this agent kit with the app SDK that we just spoke about, right? Like, it would be cool if I could spin up an agent that doesn't just kind of like do fancy tool calling, but can use the apps that I have

spoken with in my other conversations, right? It already has that context. That being said, I'm a little skeptical, Josh. Again, like, I don't know, I've got my, I must be feeling pretty dumerish today for my takes, but I'm just kind of like, okay, the example that they live demo didn't really impress me. Like, the one that you built in in eight minutes was a registration agent, which can basically guide you on your way through Open AI's Dev Day. So it tells you when the next coming talk is happening and where to walk to. I don't really care. Who cares? Like, go and do my shopping for me, manage my accounts and make me a million dollars. I'm down for that. Maybe I'm asking for too much, but I don't really feel the magic. Okay, so maybe I have one

example that could perhaps change your mind that I'm thinking of as you're describing this is, is all of our transcripts for all our episodes, they're public. We put them on our transistor page. If you want to go subscribe to the RSS feed, please go do that. Within those transcripts are all the words that we say labeled with who says it. And if we wanted to, in eight minutes, just like Christina did in that example, we could build an agent ourselves just by dragging and dropping and adding a few prompts that can analyze all of the transcripts that we've written, turn itself into a mini-GPT, and answer questions as if it was Josh or E-JAS. And we could kind of query it against ourselves. We could ask, well, what do you think Josh would say in this situation and get an answer relatively

decent based on creating this agent that you've trained in just a couple of minutes? And I think, again, a big constraint is creativity because there is a lot of ways that you can build these to do interesting things. And the hardest unlock is figuring out what to do, how to use the tools to make it interesting. So maybe that's one cool one. Like I wouldn't hate it. I would probably use that for all the YouTube comments that we get. Our auto responder. Yeah, because I mean, I mean, some of you listeners are so, so active and I really, really appreciate that. The feedback we've been getting the thoughts, the comments, the disagreements is what we live for. And Josh and I've been trying to stay up late at night responding to you guys.

But hey, maybe an agent could have helped with that. But Josh, what's funny about this announcement is they weren't the only ones to announce an agent builder. In fact, I think what was it, eight to 10 hours ago, someone else announced something. Can you walk us through this? Yeah, this is very funny. So as I was scrolling my timeline on X, looking for just like snippets from this presentation, I stumbled upon the 11 Labs announcement, which I think was like five hours ago.

And the video, if you could restart it just for people to watch, it's basically the same. the same exact visual interface that OpenAI just announced today using the same terminology. So what you're seeing on screen is it's this very visual kind of graphing chart where you have these parameters and you could draw and connect these nodes on a graph. And it's really funny to see the convergence of ideas on how things are going to be built, like literally within hours of each other. So I think 11 labs, say front ran Open AI. But the idea is that there is a directional trend here and that everything will be easier

to actually engage with. you will not need to write code. You will be able to just do, command the AI to your will. And as these tools get more powerful, you will have more and more creative leverage. And it's just going to be this really cool, self-fulfilling developer world where it's easy to become a developer because all you need to do is speak English to the model. I'm telling you, man, it's going to become a business development game. I mean, everyone's going to create some of these automated workflow products, 11 labs,

Anthropics are probably going to come up with done. Google's probably good cooking up one that they're going to announce this week. I know they've got a bunch of exciting announcement this week. So I think this type of interface is going to become pretty commoditized. What it's ultimately going to come to, and I've said this on previous episodes, is how many users do you have and how high quality is the data that you own, how useful is that data, basically. And I think Open AI in both regards still maintain the lead. And to see like a 300 million weekly active user jump in a matter of months is just insane. And I'm still, you know, if I had to pay up.

pick, if I had to pick a winner, open air is still well. Yeah. Okay. So, so overall general thoughts of the presentation. Do we like, dislike, bullish, bearish? I liked it. I am neutral on whether it's bullish or bearish because, and I explained to you before the show started, Josh, that Sam, not too long ago was promising us AGI. He was using all the acronyms under the sun to describe this amazing intelligence that can do anything in the world for you. And since, then he's kind of pivoted. He's focused more on applications. He's sold us to social media apps or to release our sweet.

Go check out that episode that we did if you haven't seen it already. And so I'm kind of left fitting like there's a hole in my stomach that needs to be filled. This wasn't it, but I'm hoping this is the step towards something that fills that hole. Nice. Okay. For me, it's funny. This might sound ridiculous and I might dislike this take in hindsight. but it was my favorite presentation that Open Eyes ever done.

And this was mostly vibes-based. I don't think in terms of actual delivery, it was anything exceptional. But the way that they held themselves in the presentation, the look, the styling, the delivery, the people who they brought on stage, the examples and how they delivered them, it was really this, like, flawless 60-minute presentation that was very highly produced, looked very refined. They clearly spent a lot of time on it. And I just really admire that relative to a lot of the other presentations that we're seeing,

where even like the GPT5 presentation, I got bored when they started showing health demos and stuff. And as an non-developer, I was engaged and excited and enthusiastic about this presentation. So in terms of vibes, at least, I really appreciated the thought and care that they put into this. And I think that probably just wrap this up. Two other small little noteworthy things that I just want to drop at the end is that they did release the GPT5 Plus API, which means developers can now query the smartest GPT model that's out there. And also, SORA too.

So the Slop episode that we recorded last week, well, there's an API, which means that any developer can reach out to this API and generate a near infinite number of AI slop or AI beauty, depending on how you want to look at it. And I'm sure we'll be seeing a lot more outputs of both of those models in the near future. But that wraps it up. SORA two, just before we wrap up, is number one on the App Store, Josh. And if any of you listeners want access to it, because Josh is a hater, but I'm kind of a lover. If you want to invite codes we both have invite codes so comment we do have four left give us your hottest take give us your hardest take and we'll drop invite codes in the comments actually yeah that's a great incentive i will i have

four left and i read every comment so leave a good one and i'll drop i'll drop a code in there um but yeah that wraps it up 20 minutes you guys we did it that was good that's an update that is everything that open i announced today some not so not so noteworthy great time to be a developer let us know what you think in the comments below thank you for watching as always and we will see you guys in the next one