OpenAI DevDay: Worth the Hype?

Shane: Hello everyone and welcome to a special edition of Agents Hour. You know, something happened today and we figured why not jump on a quick livestream and cover it. What's up, dude?

Abhi: What's up, dude? Long time no see. Yeah.

Shane: Tell everyone why are we here today?

Abhi: So today was OpenAI DevDay, and I think a lot of excitement around the day. Last year, the OpenAI DevDay supposedly killed a bunch of startups. That wasn't true. So we're gonna recap it for y'all and give y'all our take on you know how we feel about things and there's a lot of funny shit that happened too so excited to go through it.

Shane: So we'll start by first going through just everything they announced. You've probably been seeing a lot of online discussions around it anyways, if you're paying attention to this. But we'll cover it all and then we'll talk through some of the hot takes. I think my first comment as we go through this is that last year I specifically remember OpenAI telling us the future of agents was a drag and drop workflow builder. Yep. So let's take it all with a grain of salt. I've never used that product. I don't know anyone that has. I don't even know if it's a good idea. Exactly. So we will take everything they launch with a bit of a grain of salt, but there are some cool things we can talk about. So let's go ahead and jump right into it. This is just openai.com. If you find the DevDay 2026 recap, you can see this was from September 29th. So if you were watching this not live, you know, watching on either Spotify, Apple Podcasts, wherever you get your podcasts, or YouTube, that's the day it happened. So they have the full kind of like 50-minute kind of keynote presentation here. You can see Sam Altman. We're not going to watch the whole thing. You can go watch it if you want. We're going to go down and just talk through what they released. So the first thing they released is Dots.

OpenAI: Hey, I'm your dot. I'm new in ChatGPT. I'll be here to handle everything that comes your way.

Abhi: We got new Tamagotchis, dude. New Tamagotchis, man. They're going head on. Yeah.

Shane: Okay, I think that's enough. We get the idea, right? Everyone's seen it.

Abhi: Yeah, it's a fucking Tamagotchi.

Shane: If you've paid attention to Grok Bot, that was kind of the first launch in this direction, which honestly was an extension of OpenClaw and Hermes, right? OpenClaw and Hermes were really the pioneers, like OpenClaw got a ton of traction. It was hard to set up. Then you got Hermes. It was more interesting. The memory layer made it even more interesting for folks. Still like not the easiest to set up. Then Grok Bot came in and kind of built like what felt like the first consumer personal agent. And then Muse quickly followed suit. And we talked about this on the show yesterday actually from the Meta Connect announcements, and then now you have OpenAI Dots. It's like they are all just following each other.

Abhi: And they're all obsessed with these little characters and things like that, which I guess are cool, you know, but I don't know. I don't think I'll ever use this personally.

Shane: ...making people feel like AI is gonna destroy us all, which is the other, you know, aspect. Maybe that's a psyop, dude. I feel like if Anthropic released one, it would be like your agent's like a dragon or something. Like it's probably gonna kill you. But you know, here it is.

Abhi: Yeah.

Shane: So I do appreciate that, you know, they're kinda cute and cuddly, but I also think it's a little bit just cliche, feels like a Tamagotchi. But I do think the idea of personal agents is going to win long term. I just don't know which of these players is actually gonna win it. I mean I think the biggest thing is most people are going to just use it for simple things, probably. But if it got good enough to like book your travel or book hotels or whatever, I could see general users using it.

Abhi: Well then it comes down to brand loyalty and like, you know, capability.

Shane: Yeah. And my question is, and it's actually kind of why I think, surprisingly, maybe Meta's in the best position to win it. Yeah. Because they, if they could make it an extension of Facebook and Instagram. I don't know. OpenAI does have a lot of users now.

Abhi: But we'll see if we're in a dot bubble when Google, you know, Gemini releases theirs and you know maybe this is a consumer app trend.

Shane: Yeah, I mean Gemini has to be thinking of this, right? But we have Hashim in the chat saying, I just set up my dot. I don't even have access.

Abhi: Not available in South Dakota.

Shane: Yeah, I tried to go access it and it said, you know, it's not available in your region, so what's up with that? And apparently Microsoft's Autopilot is based on OpenClaw and came out this week as well. Even Microsoft's trying to get in the game. Microsoft should just be Clippy, right? It's like it's Clippy but good. Yeah, Clippy but good. What Clippy could have been. For those of you who don't know what Clippy is, I'm sorry for you. Go for it. Go Google it. We're old. All right. Also, they've released GPT-6.1 Sol. It's a major upgrade to GPT-6 Sol with exceptionally strong performance on agentic coding. Along with computer use and professional work, it delivers near-Astra intelligence to everyone at a fifth of its standard input and output token prices, giving developers more flexibility to execute complex workflows within the same budget.

Abhi: This is definitely in response to Opus 5.5. For sure.

Shane: Yeah. I don't know. I think we'll test it. I do know that there was a lot of pressure. Opus feels good from when I've been using it. So as much as I was on the Astra and Sol bandwagon for a bit, I think like many users out there, you quickly realize it's pretty easy to jump back and forth between the two.

Abhi: Yeah. Like, our team was big 5.6 Sol users and now we're 5.5 Opus users and we have not had any evidence throughout the day today that 6.1 Sol is any better than Opus. We'll do more research. So curious to see what happens next.

Shane: Yeah, we'll see how the vibes, if they live up to what we're getting from Opus, over, you know, spending a little bit more time on it, but so far not too promising. Alright, Ultrafast is a premium speed tier for workloads where speed matters most. Ultrafast offers up to 8x faster token generation, 300 tokens per second, in Codex and up to 6x in the API. I guess you can get your tokens faster.

Abhi: Yeah, if you have Cerebras stock, good for you. I suppose this is yeah, it's all Cerebras.

Shane: Private Intelligence. OpenAI Private Intelligence helps businesses use frontier AI with greater confidence that their data is protected. Zero data retention with private safety processing. Enables automated safety reviews without giving OpenAI personnel access to the underlying content. I mean I think this is just their goal of trying to appeal more to businesses and enterprises because people are always worried about if we use your models, are you going to train on it and then compete with us, especially as they're releasing more consumer-based applications, right? And competing with their users. But this is basically to say if you want this, you probably have to pay. I'm assuming you pay, Enterprise. It's like, yeah, you can have it, but it's gonna cost you. We won't train on your data, but you know. All right. So they've launched some new tools for developers. Codex in the cloud. Everyone's doing managed agents in the cloud. Now so is OpenAI. They're trying to push it a bit more. It's available on a bunch of their plans. Essentially, you can have reusable development environments so you can start the tasks more quickly, give your team a shared setup. The thing is that they've had this for a while, right? Or maybe even a year ago. But now they're putting more into it. I guess they're renaming it. They're improving the app. I think a lot of people will be, you know, running coding agents in the cloud, so I'm not surprised. They have a refreshed Codex CLI. Cool. The new code review experience in the ChatGPT desktop app makes it easier to review changes across projects. You can read summaries, explore diffs, and ask Codex about potential issues before sharing feedback on GitHub pull requests or GitLab merge requests. With automatic reviews, Codex can also take a first pass in the cloud while you're away.

Abhi: Beginning the factory business.

Shane: Yeah, they're trying to keep you in the app, which I guess makes sense. They're gonna want to do everything in the app. I imagine eventually they're gonna have like managed deployments or monitor. Like after you merge, it'll have like a deployment monitor or something. Spin it out, right? Should make sense. Codex Security Cloud. It gives defenders a better set of tools to harden their infrastructure, scan entire GitHub repositories on demand or on a schedule with ongoing checks of new commits. Codex investigates findings, removes duplicates, and prepares fixes in the cloud, even with your laptop closed. It includes access to models offered through Daybreak Blue without a separate Daybreak application. So there's a lot of security companies they're now competing with.

Abhi: Yeah, Socket, Aikido, etc. So they're getting into that business.

Shane: Yeah, even, I mean Superagent as well has, you know, some security stuff. CodeRabbit...

Abhi: ...has a security product.

Shane: Yeah. Decisions API. This is a Jev competitor, right? Yep. Trying to compete. Had to have a response to Jev. I wonder if they had this in the works or if they just like spun it up really quickly in response.

Abhi: When we did the classifier work, you could use Luna as an evaluation model in AI SDK. So it has existed, but you know, this is more of an official API for it.

Shane: Yeah, so the Decisions API enables real-time decision-making by focusing Luna's intelligence on a specific set of user-defined questions with finite predefined answers. Developers supply context using text or images and get back answers they can use to classify content, route requests, or choose an agent's next action. So again, just very much trying to compete with Jev. But does Jev support images? I don't think it does. Does it?

Abhi: No, I don't think it does, but I'm sure it will.

Shane: Maybe it does. I'm not sure. But I have never used Jev for images. I guess I hadn't really thought about it. This one obviously supports it. I guess we could maybe check, see if Jev does.

Abhi: Right now it's text only, but it'll be multimodal.

Shane: All right, so Agents API with computer use. So the Agents API now supports computer use, so developers can build agents that interact with software to complete tasks. It also brings Codex's multi-agent capabilities, tool search, tool calling, and context compaction into your application. So this is definitely a push behind, you know, Claude has Managed Agents, right? OpenAI has Agents API. Which is you don't actually even have to build your agent. We'll just host it and run it for you. You just have an API that your application hits. But now it has computer use, I guess, as well. And now they have Bedrock Managed Agents powered by OpenAI. So OpenAI worked with Amazon on Bedrock Managed Agents to take the core capabilities of the Agents API and add customization to work natively in AWS and integrate with AWS resources. So it's basically like OpenAI agents in your cloud.

Abhi: You know, your cloud, your Bedrock capabilities with your AWS spend. Makes sense.

Shane: Yeah. All right, let's talk about more customization in ChatGPT with plugins. So improved plugin creation, submission, and discovery. They're making it easier to build plugins and help people discover them. I don't know what they're actually launching besides like just improving the process, but I do think that this is where ChatGPT has a chance to win. I would say Claude is competitive here too. But I do think that there is a chance that these desktop apps become the new browsers, right? You search within them, you buy things from within them, you are interacting within them, your agent is doing things on your behalf, booking flights, whatever. If that were to happen, the plugins are the new app stores. I don't know if it's gonna happen, but I think it's a good bet by them. And I think if you are an app developer, it's risky to build on someone else's ecosystem. I think you've seen that with like Facebook and other things in the past. But there's usually a point in time, especially if you're early, where there's like benefits to being early. And then at some point you will probably get like competed away. But there's usually a way that's a good idea. Yeah, yeah, there's a window. Sites can now host plugins. So you can now add supported ChatGPT plugins to the sites you build. So teammates in your workspace can use the same app with their own connected data and permissions. All right, so they're really pushing plugins. MCP events for plugin automations. Support for the proposed MCP events specification. So plugins can start automations when something happens in a connected app. For example, you can ask ChatGPT to watch for new tasks on a project board. When one comes in, ChatGPT can read the linked documents and draft a plan. Even while you're away.

Abhi: Sounds like signal providers. Yeah, more autonomy. Sorry, our homie Max Stoiber worked on a lot of this part of the product. So shout out to him.

Shane: Yeah, shout out to Max. We gotta get you on the show, dude. We'll hit these last ones pretty quick. Wow, there's a lot of littler ones that came out. So ChatGPT Space, not necessarily a little one. It's a new home for your team to collaborate with AI to get work done. Create a dedicated space where teammates, ChatGPT and your dot can build on shared knowledge. It's almost like a Google Drive of... They're trying to own the whole canvas, right? The whole space. They have Pages. Pages are a new type of document built for human and agent collaboration. Again, that could be... Yeah, that's like a Notion. Collaborative slides. You got Google Slides. Google Slides. Create team and share tasks. That's like a project management task. It's like all the different apps that you could build that you might want to build AI into. They're just saying no, we're gonna pull it into our app. You can @ChatGPT in Slack and Microsoft Teams. So they're basically wanting to have your own channels. Meetings plugin, it's like Fireflies or Granola or whatever. ChatGPT can just join. They're basically just trying to like again launch all their different like all the people that would build on top of their APIs, they just launch competitors in all that space. Anything that's like business or yeah, any anything you'd use as like a small or medium size or even enterprise business, right? They just want to own all of it. Shareable profiles let people showcase sites and plugins, so they're even trying to be like the social part. Yeah, like the internal social network. What was the internal social network that... Yammer? It's like a Yammer almost. Yeah, it's like you build your own like internal social network slash wiki or whatever. Do more with your ChatGPT subscription. You can sign in with ChatGPT so you can use your ChatGPT plan with other tools. This is they want you to be able to auth with OpenAI just like you auth with Google, I guess. There's a new Pro tier. Now this is maybe a spicy one. For those who want even more usage and speed, the new Pro 500 plan offers our highest usage allowances. But what happened to the $200 plan? Well they brought it back, but...

Abhi: They nerfed the fuck out of it.

Shane: So yeah, you can buy the $200 plan now, or if you have one, guess what? Your usage was just significantly cut. They also have OpenAI Marketplace, so they can apply part of their existing OpenAI commitment toward approved partner software. So they're trying to be AWS Marketplace as well. So you can actually, again, buy software right through OpenAI. Okay.

Abhi: That's a lot. It's a lot, dude. A lot of different things, but like none of them were... I mean a lot of them are like infringing on a lot of startups existing and stuff like that, but you know, the hype of last year is way more than the hype of this year. You know, they're not making these bold claims that they're gonna destroy the whole startup ecosystem as they did last year. And the application layer seems to be untouched still, though they are trying to work in that space.

Shane: Yeah, I mean I think the biggest thing is that they're trying to compete with, you know, Google Docs and Google Drive. Like everything in Google Drive, they're trying to compete with security companies a bit. And then all the meeting plugin organizations and companies that are just building bots to join meetings.

Abhi: ...how can you really make that commitment, you know?

Shane: I guess the question you need to decide is where do you want your lock-in? You get to choose your lock-in. Do you want to be locked in to models and providers? And if so, then you could choose like OpenAI for everything or Claude, but then again people do that, right? People are a Microsoft shop or a Google shop, so maybe you'll be an OpenAI shop or, you know, Anthropic shop. So I could see that, but now you're bound a bit to their intelligence. So every time there's a new model release, you can't actually try it. It's just a kind of lock-in that I don't know at this early of a stage if I would want.

Abhi: Yeah. Maybe the dust will settle and there will be like a model lab like you know, Google or a Microsoft or whatever where you're doing your work in. But still it's very fragmented.

Shane: Yeah, I mean I could see for some tools it would make sense to just consolidate. But I could also see an option for some more like open tools that you could bring your own models to, right? Or use you know, open source models, like tie into models that are hosted on your own infra eventually because open source models are getting cheaper, right? That's the only hesitation I have is like do you want to own the intelligence or do you wanna like rent it from someone else? Which it kind of feels like in this case you're kind of like relying on OpenAI to stay in the lead or at least very close to the lead. Maybe for some teams that's fine.

Abhi: Curious how other companies are going to respond to all these announcements, and I'm sure some of them already are. Maybe we'll see some more releases this week. I wanna switch gears, like we went through that. Obviously we have some skepticism. I wanna show y'all what the internet is saying. Some of it's pretty funny. I've kind of selected some tweets. Let's start with the response to Pro plan changes. So $200 is back, and this joke here is: now with 50% less. Rug pull special. Yeah, they did nerf the $200 plan. They are urging you to go to $500. I'm curious if the other companies will offer $500 plans now too, because the price has moved, right? The price has moved for the Max plan. Not everything is accessible to Pro. Some are gated by business, some are gated by only Enterprise, whatever. Pro is involved with a lot, but you're gonna get less usage for the money you pay. That's just the way it is. Then, you know, we talked about Devin. You can use your Codex subscription in Devin. Really curious how this is gonna work for the Devin product. Today, you know, people are paying Devin for API key usage essentially. I'm wondering does that cut into Devin's revenue? Are they making money off tokens? You know, we're all paying Devin per usage. What does usage mean these days now?

Shane: Yeah. Yeah. Are you still going to pay the same for usage and you can use your Codex subscription? I feel like that's a tough... There's a lot of like interesting UX problems to to deal with that, right? Of like how does this actually work? How do I as a user understand if I'm paying Devin for this or am I paying Codex? If I can use Codex, why do I pay Devin? Is it because of their agent harness, their skills, like whatever? Like maybe you do, but what do I pay? Because I would expect I pay less.

Abhi: Right? Then why wouldn't you just use Codex at that point?

Shane: Because you can tag it and talk to it. Feels like a bit challenging. Like, as Devin, I don't know why you do this if you're having such good success. There must be a reason.

Abhi: Yeah, some Game of Thrones shit for sure. In the same vein, the homies over at Pi are officially, you know, OpenAI, you can use Codex in the Pi harness. You already could do this and obviously Mastra Code supports this too, but we previously were using login methods that maybe were not public, let's say. And now it's official. Good, we'll look into what the API changes are here and then just do it ourselves the same way. If something has changed. But I was looking at the code and maybe it's just more so that it's official. Then just focusing on the homie Max, Max Stoiber, you know, the plugin extensions we were talking about. There's also MCP extensions. So I just wanted to, I know I texted Max, congratulations on DevDay. You know, hopefully we can get him on the show to kind of just give us a deeper look at plugins and where they all fit into things. Okay, let's do some other, maybe not so great takes. First, on the Decisions API. This is from Diogo from Jev or from TypeSafe, I mean. And so he says here, begun, the clone war has. Just kidding. I love OpenAI. And I think more competition and validation is great for developers. Assuming the model is good, please make it good. Hopefully this is a sign for the future that building in a System 1 compatible way is the future. So kind of a you know interesting dig. Maybe we'll see if Luna as a System 1 model is good or not. Also, it's funny that Diogo hates the word "decision" in terms of phrasing of these models. Yeah, I mean I think...

Shane: I saw an interview where he definitely said it was kind of like either short-sighted or just didn't really like it as the you know the concept of calling it a Decisions API or a decision model. Right. And I think OpenRouter calls it decision models as well.

Abhi: Exactly. Some other mishaps. Apparently the Dots demo didn't go well. I mean live demos are tough, so I really, my heart goes out to this person here. But they had to, you know, cut the audio. And, you know, movie magic around it.

Shane: It's not the first time that OpenAI's had some demo issues, which live demos are terrible, right? Especially if you just launch it. But if this is your biggest conference of the year, you really, because they've had issues with like weird slides before too, you know? It's just, you think they'd really check this. I mean I do feel bad. Because the demo gods are vicious sometimes. So I get it.

Abhi: That's not the only bad thing that happened with Dots. So this is from Sui. No way. OpenAI just announced Dots. SpaceXAI bought dot.com. It redirects straight to the Grok Bot download page. And let me prove it to y'all. It does. I just click that and now... So the Game of Thrones be throning.

Shane: That's a funny troll.

Abhi: That is an amazing troll. But also, you know, the Grok Bot has been in OpenAI DevDay threads talking about how the Grok usage has not changed and is not 50% less and stuff. So there is some you know, there's just some banter from SpaceXAI. Yeah. But yeah, there's more stuff. There's a lot more negativity, but we didn't want to show any more of that. These are all the poking fun.

Shane: This is kind of like an emergency broadcast, I guess, or like a fun broadcast that we just we didn't want to wait all the way till next Monday to cover it. So let us know if you're in the chat. Let us know what you think about this kind of like shorter hot take episode just on the fly, just behind like the DevDay, and we probably could do it. For other types of like really big AI events, like NVIDIA's big conference when they do announcements. But let us know if you like it in the comments. Whether you're on YouTube, drop us a comment. We read most of them, like we try to respond. Whether you're on X, drop us comments. If you're on Spotify or Apple Podcasts, give us that review. Only if it's five stars though. Otherwise, find something else to do. And that's it. That's the show today. Kind of short, kind of sweet.

Abhi: Yeah, we'll see y'all next Monday. Alright everyone. See you next time.

Shane: Every week in AI, something insane happens.

Abhi: And there's so much drama. Every Monday we break it down live.

Shane: We do the news, we bring on guests building in the space.

Abhi: And we go deep into the stuff that actually matters.

Shane: Agents Hour, every Monday, noon Pacific. Follow, don't miss it. Peace.