chapters transcript notes
click any line to jump to that moment in the video
0:00 [music] thinking. [music] 0:14 Hello everyone and welcome back to Agentic Thinking. As you can see, it's just not Matias and I. today is an exciting day. Today is Open AI's Dev Day 0:26 2026. Some announcements are coming. , I just wanted to share with everyone online here. I'm bringing up all my best friends in AI. Ryan, hello Ryan. 0:36 Welcome. Hey, how's it going? I I get to join these let's go watch a movie in class thing today. [laughter] [clears throat] This is fun. Awesome. Matias, 0:47 who's the other host of us on Agentic Thinking, Matias, again, wonderful to see you. I'm very glad you're here. I'm sorry it's late in your evening, but we're welcome to have you here on this. 0:57 Absolutely. Great to be here. , and we're going to have to be very behaved today with four people. We've never had as many on the show, ? We're pushing the edges of what is 1:08 possible here. We'll see what happens. And then in our last Brady Bunch Square that Tommy is very excited about, we have Tommy from the Explicit Measures podcast. If I had an an audio, I'd be 1:18 "Yay! We've never met before." , we've never met. He's a [laughter] brand new friend. All . With that being said, we are going to jump in. We're a little bit behind the actual event itself. , 1:28 we're going to participate here, but we're going to do the whole mystery science theater 3000. We're going to react to the video as we see it. We're going to do a bunch of review here. , this is Open AI Dev Day initial 1:38 reactions from AI people in the space. , with that being said, let's jump in. And no one knows what's coming. No one knows what coming. We don't know what's any announcements here. , we're just going to play this by ear and see 1:49 what we get. , this expecting Sam Ortman keynote at this point, I think. Yes, very true. I'm going to cut the music on their show here. This is the intro screen here. , I'm just going to 1:59 cut the music because I don't know if it's copyrighted or not, but this is their intro screen. I did want to point this out. , as we look at this, Tommy and I and to Ryan have all used 2:10 Grockbot. Oh boy, does this intro screen look a ton a Grockbot copy. I'm just going to say it . What do you think, 2:20 Tommy? We've never heard of Muse, which is what [clears throat] OpenAI is saying or Grockbot. Yeah, this is there's no way they're not going to have some 2:30 Look at this. , come on. Come on. They're This is a straight ripoff of what Curser is doing with Grockbot. they just said, "That's amazing. Let's go do it." . 2:40 . Anyways, this is the intro screen. I just wanted to have your initial reaction. Ryan, what do you think? Any thoughts on is this going to look a lot Rockbot? I feel Yeah, it's it's what AI does you see it in one and 2:52 it comes in the other, ? It's this is gonna be their flavor of it. I feel [clears throat] Matias, have you played with Crockpot? Is that something that's on your 3:02 rad? You're doing your own thing kind of. You're playing with it, but not really into it yet. Yes. Not quite. Yeah. But I'm I'm I'm carefully listening when whenever you get excited about it, which happens 3:12 twice a week. [laughter] you'll turn me over one day. We'll see. , Open AI is going to get it by default. , you'll be able to get your own Grock bots with with 3:22 them as , hopefully here soon. , I'm going to speed ahead. This intro screen's interesting. I don't want to point that out, but let's just skip up here until they have the actual episode. , I'm going to just speed 3:32 ahead here. Yeah, until we get to the actual episode. He's going to try to sell me because I stopped my OpenAI subscription. this is going to be, can you sell me again? 3:42 They're going to take my money again. Yeah. , Ryan, you were just talking about that. Ryan, you were saying as soon as something comes out, someone else copies it and repeats it and it becomes this next great thing. 3:53 You're , "Oh, shoot. I need to go try that again." Yeah. I feel you're constantly switching, constantly changing your tools. But that's feels it's the era that we're in . It's always 4:03 changing. All , here we go. The audio is generated. This looks the 4:13 center that we were forgi this year. It's it's an AI video online. Thank you [laughter] for coming. The people I think this is going to be our best dev day yet. We have shipped a lot since last Dev Day 4:24 and you all have built amazing things with it. You find uses for these models that we hadn't thought of. You've also helped make all of our products better. Can you guys hear it? . 4:34 Thousands of requests, mostly on Twitter, with ideas for what we can improve. We try to read all of them and we work through them as fast as we can. You asked us to bring codeex to Linux. 4:44 we did that. You asked. Yeah, [applause] Linux. People are clapping for Linux. Come on. Is it in or that's what I need to know? Codeex project. you can do that 4:55 too. And you asked for codeex on your phone. we brought it there everyone else. Yeah. Again, copy. It gives us a lot of joy to build things 5:06 that you find really helpful. And the next thing is one of the things that you've all asked for the most. It's something we've been quite excited about, too. , let's get into it. We're 5:16 kicking off today with a product that we believe is going to feel a whole new way to work with AI. This is the real deal AI that we've always imagined for ourselves. It's an AI helper 5:26 that always has your back, inspired by the cool vers what we all watched in movies growing up. It can hugely magnify your ability and expand your potential. working 247 to 5:36 take on things that you don't want to do you can focus on higher level work. you may have seen early signs of what this could look in virtual assistants that help you book a 5:46 reservation or buy a plane ticket. And these things are useful to hand off, but they are trivial to relative relative to what the technology can do. And , we wanted to be way more 5:56 ambitious. Think about that AI from the future that you imagined. It knows what's going on. It knows what matters to you. You don't have to tell it every little thing to 6:06 you. It's Star Trek is what it is. It's computer. Computer do this. Computer do that. It's just the f to me that's the future of AI is build Starship Enterprise. 6:17 That's what I'm talking about. Oh, come on. Really? Come on. [laughter] Change the B to a D. Dots are remarkably capable always on 6:27 agents built to handle really anything you can [laughter] think of. I love it. and give you more time. 6:37 Let's meet. The marketing team on this is amazing. Hey, they're calling it bots. How about we call it dots? is it dots or dotbots? 6:48 What's what's an anthropics version going to be? Oh, look. They've got images and [laughter] tots tot. Hey, I'm your dot. I'm new in chat GBT. 7:00 I'll be here to handle everything that comes your way. Oh, dude. Ready to jump in when you need me. Even the shapes are the same. You got it. I'm Alfred. 7:11 I'll keep things moving and check in when something important needs your approval. Love it. Let's get to work. I've been getting your wish. The next update. You can put little bow ties on them and put glass on. 7:24 That's their update. They got a beret. [laughter] Sounds good. , I've cut the app migration in your Google Meet notes. I've already mapped out the changes. Get it started, but make sure you first 7:35 check every flow against the current product before I review. Also, update on wedding planning. Bad news, the cake vendor canled. I will say this already is fundamentally 7:47 changing how I think about bots, and I'm automating a lot more of my workflow with a Grockbot. if this is comparable to what I'm doing already in other systems, this is a big deal. 7:58 regard, we're teasing here a little bit, but this is a major deal. I think it's going to really shift how people in the business dayto-day run with and use bots. This is huge. 8:08 it may have impact on how you And then you can publish the sites. Yeah. Go ahead, Ryan. What were you saying? I was just going to say it might it looks they're trying to target even just normal everyday use for 8:19 the normal person, ? Not necessarily someone who's in except for change it that the consumer experience is first and then get into the growth numbers. How's it looking? The But Tommy, what were you saying Tommy? 8:30 When you talked the screens, too. Preview and results are in the PR. Oh, the screen. Oh, never mind. They're looking at their phones the stage. Oh, a product. [laughter] 8:41 I'm going to drag around a $4,000 screen. Hey, product idea, Tommy. Maybe that's what I want. I want a big screen with my bot on in in my house . 8:52 I will make my dobbot get to it. I can't miss that recital. Tommy, have this made. We need to sell this on PowerBI tips. Dude, the cursor is even the same as what's in 9:02 Yeah, Groc. The dot screen. The bot screen. [applause] I this. This is good. It's 9:13 what feels new about my dot is that it gets to know how I to work over time and where I want to focus my attention. It can do things review what came in overnight and can in the morning with anything before I start my 9:24 day. I've been really surprised at what a difference this makes to my life. I feel I've finally gotten some of my attention back. I no longer feel quite as addicted to my phone in the same way. It's the way I've always wanted to work 9:34 with wrong. I'm more addicted to my phone than ever with the bot the dumb thing. small things those plane tickets or the time for dinner or help with a home renovation that you've been putting off. But because DOTs are powered by Astra 9:44 and a lot of other things we built around them, you can delegate ambitious pieces of work the way you would to a high agency engineer or a chief of staff that you work with. You just give DOTs a 9:54 responsibility monitor incoming bug reports, test and fixes and draft PRs, kick off next year's budget planning cycle, and keep all the teams on track. find out what's slowing down our app. 10:05 Build a fix and show me how much faster it runs and your DOT will just get to work and keep working. , here's a good example for dots. Get our app off this legacy API before it shuts down. 10:16 Oh jeez, you're not going to give a bot that prompt. The API might be used all over a codebase, and it's not obvious what's going to break as you start making changes. But Dots can trace those dependencies and figure out everything 10:27 that needs to change. They can write the code, run the tests, and bring our team pull request to review. Think about how you would have done that the old way. 10:37 How many people would have been involved? How much time would it have taken them? What else could they have done instead? , imagine having DOTS do that for everything you can think of. Dots on board with everything they need 10:47 to get to work. They follow your access. They have their own computer in the cloud, their own browser, and they can write and test code as needed. You'll be able to start with one dot today and in 10:58 the future if you'd you can work with a whole team of dots. You also don't need to be in chatbt to talk to your dot. Coming soon you'll be able to send it a text message. It'll 11:08 text you back. It can text you updates, flag when something's changed or ask you to weigh in on a decision. You can even give your dot a call and talk with it directly. 11:18 Dots also work with more than 4,000 apps across Chat GBT. Oh, really? When you onboard your dot, it gets to use the same plugins you've already connected. It can just follow along with your work wherever it 11:29 happens. Through all of this, of course, you stay in control. I don't see anything by Astra most boundaries on app and computer use. 11:40 they bought it, ? For any actions your dot takes. Yeah, they bought it. Open claw got absorbed into into open AI. Peter, 11:50 whatever strip is working for them. He's on their staff. You can see where a lot of this is coming from. Yeah. But you also need a space to collaborate with your team and their dots. 12:02 And the problem here is that almost all productivity software was not built for AI and people to work together. Oh, Marky. we're changing that, too. Next up, we're introducing chat GBT space. This 12:14 is a place where you, your teammates, and your agents can all work together. Space starts with a page the one here you see on the screen. You can write a plan, do research, create 12:25 images, generate data, or build something new. Dots can work alongside you directly in the page. You can bring in a dot and ask it to track dependencies just you would with a senior engineer. 12:35 Oh, this sounds a a flavor at notion. Get to work. Is it does not do this? Your pages all live together in a space. This is notion. They're taking notion on that looks exactly 12:46 Tommy. You could have one subscription get notion and all this. it's going to be what can you migrate out of notion into this? This is across the vowel. This is coming soon. We're introducing a new way 12:56 to build presentations inside. Don't say who not. I've been doing this for a year. It's years for notion. No, it's just the fact that they're stepping on other people's toes hard. that's the part that astonishes me. 13:07 Slides that could be easily read and modified by agents. at this point, slides easily read by agents. Dude, that's how this all works together. That's Claude, man. That's cloud design. 13:17 . Dude, they're copying everything. [laughter] Anthropic launched a a new project here 13:27 recently, which is pretty much how we work. That's true. Yes, that was September 17th. today show you how teams got it out just [laughter] via a fictional music app we're calling 13:39 Blossom. this is Blossom Music. It's a fictional cool app. Mike, you're going have to try it. OH YEAH, LET'S GO BABY. dots have all of your connectors. You 13:50 can add them to Slack. You can orchestrate your codecs. do truly indispensable to me. It's a second set of eyes across I give it access to 14:00 and more importantly a second set of hands to take work off my plate. let's say we're gearing up to launch Blossom Music. It's tomorrow. , things are chaotic. 14:10 They're not making the music. This is lame trash. Boom. Our launch review moved earlier. It already updated my calendar. We shipped 14:21 some new features to test users last night and it's already looked at the feedback. And it looks design landed a last minute rev on our home screen and not only spotted this, but 14:31 sent me the design to take a look at here. , this is all context I'd normally have to piece together myself or worse things I would have missed entirely, but do has it. 14:42 Humans aren't capable. let's take a look at this new design. Humans are stupid. I But this close to launch, I really want to be able to play with it, ? Because Doie has access to this laptop 14:53 and its dev tools, I'm just going to say go build it. And [snorts] no pressure, Die. I love these on this later. 15:03 What framework would you be using? Is this React? Are you going to do Node? How are you deploying it? Where's this CI/CD? Is this in GitHub? Are you doing Azure DevOps? There's a bajillion other questions that are not 15:14 here. And because already knows who I am, what I'm working on. That's . . It has the memory. It's talking to a second brain. 15:25 let's say I'm walk and I just need to catch up on last night's alpha. Your number's not available. 15:35 Good morning, Holly. His number's been disconnected. Hey, I'm just walking. We have business funding for you for $342,000. Pick [laughter] up. Sorry, you're out of 15:51 Oh, all . , how's it going? Internet internet. 16:03 Grockpot just added this feature, too, by the way. You can call your Grockbot. This is for two weeks they've had this feature out. You can call your Grobot and talk to it. I'm going to review with the go to market leads later. Do you want to just 16:13 drop an FAQ into the page? I do this feature though. . Really? Thank you. Yeah. No, I don't need to call with a phone number, but just talking to my 16:23 bots with natural language. I want to go for a walk and just talk to, hey, what's going on today? What are my emails? Let's draft. I want to just walk and talk to it and have it just do things as I Matias, you do this all the 16:32 time already. You're already taking notes with agents and stuff. Absolutely. added. you'll have a conversation with it and it'll talk back to you and then you do it that way. Pretty good. Yeah. Yeah. The Windows app has it too. 16:44 just added when I Interesting. I did see they had that feature. I just was never thought about trying it, but maybe I'll give it a shot. Interactive charts. It's going to get better. It's a little 16:54 rough , but I think this is I where it's going. This is notion and your dots can command. Oh, wow. back slash for 17:04 commands and options. Just tag in teammates but I can also tag in itself and say 17:14 I don't know how not works Tommy this is I'll lean on you for this part of thisart. Do you want me just to share my screen and we'll do the exact same thing [laughter] and the visualizations in here are completely dynamic. 17:24 Can you talk to filter? You can talk to your agent. You can comment on any text or on the page and it'll start working on and my makes total sense because you got a notion plugin with your agents. 17:35 Nope. It's all built into notion. It's all built into notion. And because all my context I can, hands it off [laughter] would really slow me down. It's a 17:46 the same as buying a rock open or anticipation. It's 200 bucks a month for that token limit or credit limit. Yeah, I'm calling it . 17:58 It's going to talk to another dot. He's just going to tell something more mention multiple dots. I'm calling it . You can give me your $20 after. space is a place where dots and your 18:08 show is together and stay in [laughter] how we work is Slack. All of us are getting tagged all day and sometimes an 18:18 innocuous ask reconciling launch specs turn into a hole. No, thank you. I know. I can forward the thread to is pretty good. out of 18:29 a lot of the Microsoft products, I think teams is a pretty good product. And as you can see, my DM is full of these little handoffs. I get to decide what needs my attention 18:40 and what my dot can take care of. And by giving dots their own identity, they become extensions of our team. here's something we totally didn't 18:50 expect. This is the part Microsoft is missing. Treating their dots they're delegates. Microsoft's going to have to respond with their they did their new co-pilot announcement. Gabe and this is not what I don't want 19:01 copilot to be reinvented as a new co-pilot thing. This is what I want copilot to do. Co-pilot should be making dots and bots. This is taking the identity of that agent should be able to 19:12 be used anywhere in any of my systems. This is the way to build it. Microsoft's missing this boat at this point in my mind opens a PR with 19:22 and this is a real example. Our engineers really have you looked into admin portal lately where you've got groups and users. There's an agent 19:32 category as . They already have and I've delegated a lot of things to my but have you tried to use it? Have you tried to give any of your agents access? 19:42 [clears throat] I understand the concept of it but unless you're building a co-pilot agent that sits inside the agent world none of it works for Grobots. None of it 19:52 works for dots from open AI this is all missing. None of this exists in the Microsoft ecosystem. the app and launch. It's the same. , Kentucky does 20:02 have pizza shops . This [clears throat] might take a bit. Don't necessarily understand, but . But the amazing thing is because you have it doesn't mean it works on my phone. Run local execution and get it to be a 20:13 working app the bike in my garage. And then I can even ask to just run with it and ship the PR. My bike is a Huffy and Tommy rides around in carbon fiber bikes he 20:24 knows. Dots are extensions of us. We can give them our own all of our context and memory. and they can take work work off off of our plates. 20:36 They can filter the noise. And when you add all of this together, dots really become extensions of you. It's not just about getting the same work done faster. Here at OpenAI, we've 20:46 felt this firsthand. The dots are changing what we feel capable of taking on. And when this scales to everyone, imagine what you could build together. , I can't wait to see what 20:56 you do with this. Thank you. All . Pretty pretty good demo. The chatting thing is nice. They got to refine that a bit more, but I do that feature. 21:06 I have personally been waiting for this product for years and we can't wait to hear feedback from you all on how you're going to use it. New $500 subscription available to all chat pro business 21:16 premium and enterprise users today. use your dot is included in your plan and conversions don't conversations don't eat into your usage. You can always work with your dot. It's available 24/7. 21:29 One more. Yeah. What did he say there? They don't conversations into your usage. These will hold. One more thing for enterprise users. We're also previewing specialist dots 21:40 today. These are it's that will work alongside your whole team. you seem to have scale intensive work marketing 21:50 you can give them goals and context give them feedback to improve and that'll be we've been testing these inside of open AI and they have been remarkably 22:00 effective in many different roles for our enterprise customers we'd love to work with you to get them set up and make sure they're really great we're also working with Microsoft to 22:10 integrate specialist dots into H365 that businesses There it is. You called it through the Microsoft tools that they already This is This is the missing gap, ? They're building it. We can get it 22:20 inside fabric or inside a 360. That bike is running , Mike. Get much more done. [laughter] Whether you're using AI to expand a small business, create more 22:31 accessibility, or finding new ways to fight infection. We want people to discover that they can do things with AI they never thought possible. that brings us to the second big 22:41 part of today. We want to support you all with a powerful platform for building whatever you can dream of. There are three parts to this that we're really going to talk about. Models. We want you to have the 22:53 best access to the best intelligence at the speed and the cost you need. Second, better tools. We want you to have the same tools we use inside of OpenAI, the freedom to build and run your products 23:04 wherever it makes sense for you. And third, distribution. We want people to discover what you build and connect you with new customers. 23:14 first let's start with the model. We launched Astra a few weeks ago. It's the most capable and aligned model in the market, but people keep asking us for two things. Make it cheaper and make it 23:25 faster. And we really didn't want to disappoint. Here it is. We are launching GPT 6.1 Soul. [applause] 23:35 You don't even know what it is yet. [laughter] You could have said whatever you wanted. We got to tell you why it's good first. this is a powerful model. , it has a 23:46 great balance of cost and capability. It gives you very near Astro level intelligence at a fifth of the price to help you code and build. It's extremely affordable to run. In fact, the cash 23:57 input costs are a 95% discount on standard input. this makes it much cheaper for agents to reuse context, iterate on solutions, and work through longer tasks. We think you will really 24:08 love this model that is smarter than Astra in some ways, and it will become your daily workhorse. we're very excited to get 6.1 Soul launched today. But sometimes you just want the highest 24:19 intelligence at an absurdly fast speed. today we're launching ultra fast across the API, chapter, and codeex. 24:29 Nope. [applause] 300 tokens per second. I think this is cerebrous and we think you are really going to love this and not want to go back. Fast 24:39 already gives you twice the speed at twice the price of standard, but ultra fast takes that to eight times the speed at six times the price of standard. An incredible 300 tokens per second. 24:50 what? It's worth it. , let [laughter] me show you . , this is the same task side by side. We gave both models the same prompt. create and launch a rocket with the dev day 25:01 colors. On the left here, you have ultra fast. On the , you have regular. You can see how much faster this builds. The ultraast rocket is already taken off, and the regular rocket is not even close to being finished. There is 25:12 nothing this on the market today. You can use ultraast with Astra and soon we're going to offer it for 6.1 soul as . Wow. here's what these improvements mean 25:23 for your pro subscription. We're introducing Pro 500 with our highest usage limits and access to ultraast in chatbt and codecs. Pro 500 25:33 will have 25 times the usage of plus and you can use it across partner apps with signin with chatbt which we'll talk about later. We are also reopening the pro 200 subscription today. Pro 200 will 25:44 continue to have access to all of our Frontier models. And with GPT 6.1 Soul, you will have a model close to Astro quality that you can use a lot as your daily driver. Our goal is to give you 25:56 the best combination of intelligence and capability at every price point in the whole market. One more exciting speed update. Today we're previewing our new decisions API. 26:07 The decisions API lets the model respond. Yeah. [applause] Oh, this is interesting. The decisions API lets the model respond 26:18 in a fraction of a second fast. it really feels almost instant. Seriously, they just added a computer moving through these steps. This is not sped up. Yeah, 26:28 this works by giving our Luna model a predefined set of options to choose from routing a request, classify an image, deciding what agent, and by focusing the model on that choice, we 26:38 can make it extremely fast while keeping capabilities image understanding, broad language support, and safety protections. We're very excited to see what you'll all build with this. 26:49 thank you. [applause] OpenAI offers leading models across intelligence, price, speed, 27:02 voice, images, frontier models in every category. We want to be a one-stop shop where you can build. I'm really proud of our research team for getting the models to be this good, usable at this 27:13 scale, available at these price points, into the hands of developers everywhere. there's a question of how the researchers can do all of this. Last year around this time, Yakob and I made 27:24 a somewhat bold prediction on a live stream saying that we would see the first AI research intern in the next year. to say the world did not have a lot of confidence in that prediction at 27:34 the time was an understatement. A few weeks ago, we announced that we hit that goal. We have an AI research intern, a system that can take on a clearly defined research task that 27:46 would take a skilled researcher a lot of time and effort to complete. And we want to show you what that looks and how capable these models have become. for that, I'd to bring up one of the 27:56 researchers working on these models. Please welcome Tel. [music and applause] This is something that people have been touting for a while. that is the AI is 28:06 going to get good it'll have scientific scientific discoveries for us and I'm excited to share how these models are changing the way we do research here at OpenAI astronated models are great 28:17 for math coding and computer use and all of these matter a lot for research being able to use a great model for research helps us train even better and more aligned models in the future we can 28:28 bring capabilities that are more useful to you for example my collaborators and I work on computer use and our models help make Astra a step change improvement for the industry in speed, 28:38 performance, and safety for desktop and browser navigation. Here are some examples, not sped up, of the model helping with legal tasks, designing assets for a video game, making a 28:48 dashboard, or even filling in your taxes. And our model, my son is building video games. We built a 3D game the other day. Wasn't very good, but we could navigate the world and see things. And he was 28:59 talking to it, and it was adding more objects in the world, sliding doors, and he was And we have this is impressive. It's cool. The world of game dev is going to change greatly with this find wins in the 29:10 harness that would improve latency and performance. And then we pulled these into the real prod harness and post trained on them. And this has resulted in already a more than 2x latency win that we've shipped to all of you. 29:22 Cool. I'm not sure what to make of that. She's saying that she they had used research to figure out how to make it more optimized. Our models help make Astro much safer at computer use. 29:32 They use AI to build AI. behavior training to improve refusals in unsafe scenarios and this has helped us achieve industryleading performance on computer use stress tests. our models are much 29:42 less likely to make computer a desktop or browser and much more likely to be user user intent which should really matter for all your dots. 29:52 I think computer use in agents is underrated and that because not every website has an API or an MCP yet our research program. We're not there yet. you need token use within research and you can see 30:02 especially after this summer token use really skyrocketed internally and we're seeing returns on this investment. Wow. that's because your models got we track how often a model can carry out 30:12 a task in research and in January the models were pretty successful on short less than 15inut tasks but failed the vast majority of times on a task that took a day or longer. But as a good good 30:25 ultra fast, our models have taken on over a third of dayong research with no intervention required. what does this mean? Yeah, that chart made no sense. 30:35 AI is to ensure that API community and we want our models to help a number doesn't require a chart. [laughter] Models Astra have already helped solve math problems that were open for 30:46 decades. I think it was how long the test takes on the bottom antibiotics versus the success rate. Taking less time but getting better successful energy and improving energy 30:56 efficiency. She was saying that the longer tasks would fail if you have a test that you give it that's they fail more long. , got it. Yes. even those long tasks can still 31:07 have higher success rates. They're improving it. It's still 35%. That graph sucks. Back to you. [applause] 31:18 Thanks, Tachel. That was awesome. , we have a lot more stuff. , that brings us to the second part of supporting developers, giving you the tools to build what comes next. We want you to 31:28 have access to the same foundations that we have at OpenAI. This starts with the codeex harness. It powers codecs work and dots. We have put a lot of effort into making this 31:39 harness very efficient that you get accurate results at the fastest speed with the fewest tokens possible and it's open source you can look under the hood. Thank you. And build it yourself. 31:49 Oh, interesting. To power products chatbt, we built this harness to run at scale for billions of users and that same scale is there to support your products, too. 31:59 Next up, we're making this sounds something you're interested in the harness. Something I know you all have been waiting for. Codex is fully in the cloud. [applause] [cheering] 32:13 this means you can start a task on your phone, pick it up on your browser or your desktop, where you left off. No more lost threads. You don't have to worry about closing just catching up with everyone else. Your agents just keep working. 32:23 Yes, sweetie. No, this cloud harness new possibilities. This is a me too confl 32:34 that we're launching today. We wanted to give defenders a better set of tools to harden their infrastructure. We see what's coming and the new codec security cloud runs in cloud environments. It continuously finds 32:44 vulnerabilities and prepares verified fixes for you to review. We've added some highly requested new features automatic dduplication, scheduled scans, a new interface, and you can start using 32:55 it today. Next up, we're launching a new agents API to make all of this available to you. The agents API 33:05 core components the harness, hosting, memory, multi- aent controls, and much more. This is the same technology that powers codecs, dots, and all our agents. And it is available 33:15 in public beta you can start building. And we're bringing computer use to agents API. This gives your agents a whole new set of powers. [applause] 33:25 Give us the API. copy a demo of what this looks . say you're building an agent that tests websites. This agent can use a computer to open the browser, click 33:35 through the pages, and test the flow. We also want to support you on your infrastructure of choice. It's been great to see how quickly our partnership with AWS has grown. We've worked 33:45 together with them on bedrock manage agents powered by OpenAI. You can build OpenAI agents in AWS alongside your existing applications and data. Microsoft 33:55 AWS customers can access and chat GBT work too. here's a big one. When you build these agents, they need to work safely 34:06 and protect your data. , today we're previewing OpenAI private intelligence to give you stronger controls over your data while using the most advanced capabilities of our models. 34:17 ZDR with private safety processing delivers frontier model safety without storing your content on our servers at all. And with private inference, we can even provide privacy at the time of 34:27 inference. Together, we think these set a new standard for privacy and Frontier AI. We've worked closely with some of our largest customers to design privacy and safety systems that meet their 34:38 highest standards they can apply this intelligence, and you can too, to their most sensitive work. [clears throat] all of this and everything you're 34:48 building needs both speed and reliability. We've been hard at work on this. We've seen 100x growth in the responses API in the last year. 100x in the last year and we have still kept 34:58 more than 99% reliability. We've cut the time to first token by 45%. And we have sped up tool calls and workflows by more than 30%. We will keep going on this. But these improvements 35:09 make this the most performant and reliable API available anywhere in the market today. thank you. [applause] 35:22 That's a lot of new things for developers to build with and it wouldn't be dev day without a demo from him on stage to show all the new ways you can build with these new tools. 35:33 [cheering] Thanks. I'm just taking a few photos of the waterfront. I'll be there. Yeah, I was just there. That was the that was where they had the Ignite conference. Can you help me? 35:44 Sure, I can help. Tell me what you need. Can you welcome everyone and share a fun fact about Fort Mason? I'll be coming shortly from the left door, get my demo ready. Thanks. Let me get that ready. 35:56 Hello, Deb Day. Great to have you here at Fort Mason. And hello to everyone watching around the world. Fun fact, at San Francisco's 1915 [applause] World Fair, visitors could hear voices all the 36:07 way from New York over the telephone. Today, you've got an AI on the mic, a room full of amazing developers, and a few new tools to play with. Oh, there's Roma. All . Good morning everyone. 36:20 [cheering] [applause] I love your demo building with codeex even when we're not on our laptops and kick off a task this one from anywhere. Last year we did not even have 36:31 an app for Codex. I don't have enough time today to show you half of the things we're shipping. let's dive in. The sunglasses are driving me nuts. It's been awesome to see how all of you have been creating with us throughout the 36:43 past few weeks and you based of the venue [laughter] to recreate the same thing here. , let's take a look. What I really wanted to try is how we could change it together in real time 36:54 with ultra fast. , let's dive in. Talk to it. I know. Give me a second to 37:05 start here. Oh, no. It's the most reliable API. [laughter] 37:16 [clears throat] All righty. Interesting. Send feedback. We might have to type this time. It's . Don't just type. . 37:27 Don't do it again. what? I'll just chat was down for me earlier today. Fly me to the station going on. Yeah. 37:37 Yeah, it's that new faster model eating up all the tokens. what's interesting here is that I can interact with the the room with ultraast. what I'll do instead here 37:48 if we have a little difficulties with the the voice chat I'll go back here and let's do something else with this room . It feels a little empty. I'll say why don't you fly the 38:00 dots to their seats for instance would be nice because feels a little empty far. Time to bring the near the art. we're just changing this world in real time as we go. Pretty cool. , 38:12 maybe we can do something a bit more risky . , and , why don't we just turn the camera to the room to the audience and I'll say, , put the 38:22 live stream on the big screen. What? And let's see what happened there. Where we There we go. Wow. if you want to 38:32 wave at the audience for those of you watching on YouTube, thank you. Thank you for tuning in. Great to have you watching on YouTube. U you can build and change a world 38:43 this or build anything almost as fast as you can think. There are many different ways to build with that Tommy. I want to show you a few of my favorite ones today. The ideas of the speed of thought. 38:53 let's switch to a terminal for a minute. I don't know what he's thinking. we have completely refreshed the Codex CLI with tons of improvements you've been asking for. let's start with this empty project here in React on the 39:05 side. let's see if I can start voice here. Just we might have some voice difficulties at the moment which is pretty unfortunate 39:15 but it's . what we're going to do ultraast anyway but this time we'll do it with simply the the command line. we'll say build an app 39:25 typing last three attendees 2025 at least. I want to give them I've been doing a few demos at next year and Mike can you do the same. 39:35 let's see what happens there. It's funny how my most demos are just talking to an AI bot to show domin 39:46 we have all of these powered by GPT live. the same as in the app. this is not typically dictation. you could just react in real time with all of this. , a few seconds in the making 39:56 here and boom. Our app is ready with ultra fast. let's do something interesting here and let's try to switch things up a little bit. I'll say, what, 40:06 make it six attendees. Why not? Don't blink. It should be fast. And boom. It's already changed, ? , . let's bring this 40:17 full screen. And should we click on this button? Yes. always. All , let's do it. , let's see who's coming back to Dev Day 2027 . 40:27 All , we have six winners. Congrats. [applause] That's cool. 40:37 This is going to be how we expect to build from here and email details later. this is going to consume everyone's thoughts I want to build this audio in the CLI that all of you can try 40:48 later today. And we've added a ton of other new features, too. But let's go back to the app. Thanks for the links. I'm going to put these in the chat that you're sending. Before we do , I will just 40:58 start this project . We'll go back to it in a second. I want to tell you how Astra Adventures came together. It's a simple idea to build a 41:08 game with Astra where you could fly into space in 3D. And it all started from this sketch on paper. , if you take a look, after a couple turns, we got some place, but as you could tell, 41:19 the graphics were a little rough. And that's where having great models across all modalities is helpful. , as you can see, it was easy then to after a few turns get to a place where 41:31 we have these amazing graphics from a sketch that is very textured, very beautiful game. from here quest just got a 41:41 graphics upgrade the game dynamics a little bit and it should be waiting for us. let's [snorts] send a link to I'll put it in the chat. instead of playing this game what I'll say is show us how you play the game. 41:55 What's interesting here is that we're instantiating a browser use connection. that means that Astra will figure out how to control the game and learn how to play it in real time. 42:06 in a second, it should be able to start the game, choose a ship on its own, and if you take a look at the screen here, on the left, you'll see the decisions that the model is making. And 42:16 on the , on the very far of the screen, you'll see the keys it's pressing on the keyboard to figure out how to play this game. And obviously, it's a very beautiful game, but I love watching Astra 42:27 figure out a game that was , you know, designed for us humans, but then figuring out how to play it. Pretty cool, ? Lava Quest 2.0. the idea that it here. 42:37 All , gentlemen. It's World of World of Warcraft Fabric Quest style. Hope it's good. Rest of it. See you guys. Hey, Tommy. Thank you very much for joining us. I appreciate you. 42:48 If I bring a different project over here and let's say I bring I do apologize. I also have a hard stop as . maybe we wrap it here and pull back from this one. what do you say? Call it here. Maybe we'll 42:59 look a little bit a little bit more later in a different episode. Sure. What do you think? app across screen sizes and take screenshots along the way. 43:10 I'm going to pull off the audio feed there. , that was an initial reaction of the open dev day from OpenAI. very exciting. very neat to see 43:21 what they're doing. They're also in that demo, they're using the Duo from Apple. That was the Duo phone that they were seeing that they were demoing there at the end. , very interesting. Very cool. the announcements that are happening here. 43:32 This was fun. , if you this content, let us know in the chat. , let us know what we're If you this content, you want more of these things real people using real tools reacting to the actual 43:42 announcements. If you this, let us know in the comments below. , please give us a thumbs up. If you if you this content as , also please give us a thumbs up on the video as . With that being said, , 43:53 everyone, thank you much for joining us. Ryan Matias, Tommy has left. Thank you very much for joining. This was super enjoyable. I love unpacking these ideas with you and hearing your 44:03 thoughts and ideiation as this stuff happens in real time. , thank you all very much. We'll see you next time. Thanks. Bye-bye. 44:21 [music] Hey, fancy thinking.