Chapters
Questions from this episode
- What did OpenAI announce at Dev Day 2026?
- At OpenAI Dev Day 2026, OpenAI announced Codex for Linux and mobile, always-on cloud agents called Dots, collaborative ChatGPT Spaces, an enterprise Marketplace, GPT 6.1 Sol, Ultrafast access and the Decisions API. The hosts offered first impressions, focusing on practical usefulness, reliability, product positioning and enterprise adoption.
- What are OpenAI Dots designed to do?
- Dots are OpenAI’s always-on agents, each with a cloud computer and browser, designed to learn how a user works and handle delegated tasks. The launch examples ranged from restaurant bookings and home renovation to API refactoring and portfolio rebalancing, a breadth that left the hosts asking who the product was actually for.
- Why were the hosts concerned about using Dots for business tasks?
- The hosts’ main concern was reliability in consequential workflows. Will argued that an agent unreliable enough to require checking should not rebalance investments or make payments, while Jim preferred deterministic automation for tightly defined tasks. They saw promise in cloud agents, but said safeguards, permissions and human review remain essential for serious business use.
- What is ChatGPT Spaces, and what risks did the hosts identify?
- ChatGPT Spaces is presented as a shared workspace where colleagues and Dots can create pages, research, charts, prototypes and presentations. The hosts questioned how teams would manage version control, overlapping agent edits, accountability and compliance. They suggested AI may be more useful for first drafts and routine reporting, followed by human review.
- How does the OpenAI Marketplace procurement model work?
- The OpenAI Marketplace is described as an enterprise procurement program for customers with committed OpenAI spending. Eligible purchases from participating software partners can count against that commitment, while customers negotiate with and are invoiced by the partner. The hosts questioned whether this could obscure churn or create circular-looking revenue, but framed that as speculation.
- What are GPT 6.1 Sol, Ultrafast and the Decisions API?
- GPT 6.1 Sol was announced as a lower-cost workhorse model, while Ultrafast offers higher generation speed at a higher per-token price. The Decisions API narrows a model’s response to predefined options for tasks such as routing and classification. The hosts liked its speed and typed outputs, while noting similarities to previously released Jev models.
- Who was named Business Idiot of the Week, and why?
- The Business Idiot of the Week was the marketing team behind OpenAI’s Dots. The hosts reckon its cutesy consumer styling clashed with examples spanning restaurant bookings, software refactoring and portfolio management, leaving no clear audience. Alex also alleged the team failed to secure the domain, and Will and Jim predicted Dots would be forgotten within ten years.
Read the full transcript
Will: Welcome back to the Business Idiots podcast. We're very sorry that we missed you last week. All 11 of our subscribers complained to us. So we're sorry about that, but we're back again this week, guys. I'm here with, of course, with Alex Demlick and Jim Lovell. And how are we this week, guys? It's up to 17 now, isn't it? Isn't it 17, mate? We're going viral.
Alex: Watch out, Mr. Beast. Here we come.
Jim: Well, but the funniest thing is the YouTube shorts. And Will's had this theory on them testing things out. Now that we're posting more frequently with clips, they're giving us a shot. And so it's really quite disheartening for me because you guys have both had a short that goes four times the number of views than any others. But they haven't had one for me yet. You know, is that I'm special too.
Alex: It's because you sit too far back in the camera, mate. No one can see your pretty face.
Jim: Well, I'm going to zoom in on it, you know. It's all the tricks of the editing is what I'm looking for here, mate.
Will: All right, you'll have to write a letter to the algorithm and ask it to just spread the love around.
Jim: Well, I started to wonder whether it's because I've stopped using Gemini as much and... the overlords at Google know, and so they're just punishing me.
Alex: Yeah, they're just singling you out. That's not a persecution complex at all.
Will: I've got a lot of free time. Actually, they released a new model recently. We should cover that at some stage too. I'm hearing good things about it, but I haven't tested it myself.
Jim: Again, I don't know. I think they dumbed it down a bit, but yeah. I've got to play with it properly, but... The thinking model or the more reasoning level now does the emoji thing in documents.
Alex: Oh, I know exactly what you mean.
Jim: You just go, no, I'm trying to get you to get me an advance on getting all this work done. I don't want emojis all the way through a document that is clearly... for a customer, you know?
Alex: Yeah, this is a legal contract. It's not a junior product manager's blog.
Jim: I don't want clause 3.4, you know, this will result in termination of the contract and then a knife and a...
Will: A little sad face on the customer complaint handling section.
Alex: You know, I'm coming around to this idea.
Jim: All of a sudden, you figure through a little bit and Alex wants to get onto emojis everywhere.
Will: I kind of want to do it just to know that I've forced a lawyer somewhere to review a contract that's full of emojis.
Alex: What I'm waiting for is LLMs to give us a full natural language to emoji conversion so that we can rewrite contracts and just be like, oh, we've updated it for the younger members of your team.
Will: Hey, you could do that since like GPT 3.5, you can get it to like, you can put in a whole bunch of text and then just say, convert this to emojis and it'll spit out like 30 emojis in a row with no words. And it tells the story pretty well.
Alex: Well, yeah, intelligibility is the key thing here. I want it to be intelligible enough that the person can't really justify stop, but frustrating enough to read that every moment is like having teeth pulled.
Jim: There's a very specific sort of sadism I'm shooting for here. No, 3.5 was where they added all the translation and all the really high-quality translation from language to language. So why wouldn't you add emojis to that?
Will: Moji's the language. So this week we're going to be having a look at OpenAI's Dev Day for 2026. They launched a lot of new products and I suppose promised a lot of new features that are coming soon too that are fairly remarkable. We've only really heard what people have said about it so far. We kind of wanted to hold our first look at it until we could record this episode and we're going to go through and do our sort of first impressions based on the Dev Day. So, yeah, let's dive into it and we'll kind of get our first cuts of what we think of OpenAI's Dev Day 2026. This will also be a little bit interesting, I think, because when we go to choose our business idiot at the end of the podcast, there's not a wide variety of choices we've left open here.
Alex: Well, it's wonderful foreshadowing because the screen opens with the pic of pretty much everyone in the room. That's cinematography for the watchers out there.
Jim: Well, okay. Are we then just working out if it's his business idea of the week because he did Dev Day and Brockman went to the White House and he probably could have gone either way?
Alex: Yeah, it's a Cardassian crime story. Everyone's guilty, but the trick is knowing why and for what.
Will: Yeah, did he choose not to go to the White House or did the White House put the meeting on specifically on the same day for a reason?
Jim: Who knows? But it certainly, I think, it will be one of those sliding doors moments where everyone looks back and goes, oh, hang on, Sam's not in this photo.
Will: Matt, it's a US patronage system now, right? So he's missed an opportunity to give patronage and Greg Brockman's going to step up. Anyway, guys, let's get into it. Here is OpenAI's Dev Day 2026.
Alex: Good morning and welcome to Dev Day. It is great to have everyone here and everyone joining us online. Thank you for coming. I think this is going to be our best Dev Day yet. You asked us to bring Codex to Linux, so we did that. You asked, yeah. Wow, a lot of people like that one. And you asked for Codex on your phone. So we brought it there.
Will: He seemed a little bit surprised that people were happy about the Linux announcement straight out the gate.
Jim: Well, I found this really interesting in that I just used the CLI and it really highlighted to me how many people are still using the GUI. And so then they and so then all of a sudden your ability to go parallel with agents diminishes because if you're relying on the UI continually, then you just can't have that many windows open. Whereas I can have as many command line boxes open and it all works perfectly. And so that was the most surprising thing about me because I've been native Linux for forever. And it never crossed my mind that I would download the CLI for it.
Will: But how many sessions do you actually have open at one time, Jim?
Jim: How many?
Will: Yeah, how many sessions do you have?
Jim: Well, I started managing it a lot better by doing it through the virtual machines. And so now I have up to six on any given virtual machine and I have 15 virtual machines.
Will: Okay, that's some scale that the GUI would not provide. Yeah, I get it.
Jim: Well, and that's it. You just can't manage that many sessions through that many UI instances. But you can with the CLI. And because it then becomes this hierarchy where they all have their own team, and the only reason I need to go and access it is to clear a blockage in some way. When I get to that scenario where I can actually have them pull a ticket, do the work, and submit it without creating a false blocker, I won't need to have all the sessions accessible. They can just run in a Docker container or they can be completely headless then. That's where this needs to get to, but they just create so many... blockers at the moment you know this is what opened my eyes about this you know codex on linux oh right everyone still needs the gui because it's just so frustrating having to clear them all the time yeah i gave it a shot with one of my co-workers was saying oh it's pretty good you know give it a bash
Alex: And so I downloaded the GUI. It proceeded to scan through my project repos, find every thing I was working on that was attached to Claude, put its own artifacts in over the top in common team repos where this was... you don't do this, would continue to do so every time I deleted it or changed to a fresh branch. So I was constantly cleaning artifacts out of my PRs and then got it self-uninstalled. So that was the saga of Codex on my machine. The CLI is still okay, but the GUI, yeah, get rid of it. It's undercooked.
Jim: The Claude, because again, when they announced this, I went and downloaded them both. So I was, I too had a play. Claude does exactly the same thing. And so all of a sudden I'm finding, but it's also, it's then referencing, it's referencing other chats. I go new chat and I want to work on something new and clean. it's referencing things from other chats. That doesn't work. That context bleed is a massive problem for anyone doing anything serious. I understand where it is useful for the consumer or for the individual worker who's just focusing on one thing and working through it, but that's not what this technology provides me. It provides me parallelism. I can all of a sudden work on multiple things at once and have them move forward in a productive way.
Alex: I still think co-work has some merit. Having that in the GUI is okay.
Jim: Sure, but as long as it doesn't reference other chats.
Alex: Yeah, I think for the sorts of work you do with co-work, that is far less of a problem, like context-based. It's particularly an issue when you're trying to build software, but like... realistically fable is a gaslighting little that lies to me all the time about the amount of work it's completed however if i need to just dispatch something to go do some research and collate a bunch of stuff and get back to me about how something works or something it does its job it's parallel enough if it remembers me yelling at it in the past good because i don't want to be spending too much time optimizing those particular workflows i just want the problem sorted
Jim: Well, did you see the change in the terms and conditions of use this past week?
Alex: Yes, and I'm very upset. You've got to be nice to it now, Alex. That's the other big use case I have for these GUI apps. That's where I have my stupid conversations where I argue with Claude or Chatupity about that it's not sentient, that it is not deserving of rights and ethics and all this sort of stuff.
Jim: But that's where... that's sort of where like it's interesting there was none of that narrative from dev day from open ai dev day it really did they really just went okay here's products guys this is what we've got for you is that we want you to we want we're trying to and this is the this is sort of you know probably it's this is probably something i should be saying towards more towards the end is that my overarching feeling of all of this was they went okay we've got to go and meet people where they are And to me, that's one of the core principles or first principles of shipping product. You've got to meet the people where they are and build them into a scenario where they're going to go in the future. And I think this was a big step forward in doing that. And listening to, we're piling on a little bit here, but listening to feedback and going, hey, why haven't you got a Codex GUI on Linux? It's what you all build in. Why would you ignore that? I think that's fantastic forward movement.
Alex: Well, I think there's a massive screaming gap there.
Will: The Codex mobile thing here I think was good because as someone who kind of has Claude working all day from when I wake up to when I go to bed and then even into the night, I need the ability to not have to sit at my desk all day. So I've been using the Claude mobile for a long time. So having Codex not have mobile in the past was definitely a blocker for me for switching over. So that's interesting to me. I think as well... know based on this conversation like we're still in the previous weeks we've talked about separating out the quality of the model from its harness but again there's still now a further decomposition we have to do which is the harness from you know the product wrapper that they've got here around you know the codex application because there's a bunch of great stuff they're putting into the codex application here but it's also codex is the name for the harness that they wrap around the model and i would love to be able to use the codex harness around their model without having to necessarily adopt their codex product application because maybe the cord one's better so i think you know they're still having to go down this pathway And I think Anthropic led it and OpenAI just behind, but they're still trying to bundle the product layers as deeply as they can into the model layer so that they have a moat because the model layer is not going to be a moat, it's going to be a commodity. So how are they going to still retain users and growth without this kind of bundling of these layers?
Jim: No, for sure. And I'm not sure it will ever happen between OpenAI and Anthropic. That might be a little bit of an Android iOS scenario moving forward. But I can absolutely see that you'll be able to use open source models through Codex and vice versa through Cloud Code. I can absolutely see that happening. purely on the fact that then you're a, once they've gone through their IPO, you're a secured customer on your subscription or on whatever plan you're on, then they'll start letting other models in or something like that. That would be, to me, would be the smart move because I went the other way, Will. I started to then use OpenAI models through OpenRouter because they, you know, you put me onto it actually is that the image generation from the new range of chat GPT models is phenomenal but getting it through open AI is a little bit more of a headache or changing everything so that I can just get it through open AI is a bit of a headache I can just use the models through open router away I go why is that To me, that is the more, the smoother transition. And so then I, well, it's more, the Claude Coe auto classifier has, auto mode classifier has got me going, okay, well, I'm going to build my own hearts now just because it's, they are just trying to clamp down too much. And although like OpenAI is doing good things now, I don't know whether how long that lasts. where they don't start restricting.
Alex: I look at these announcements on the line of where they're going with this, and I wonder who this product is for. Now, Will, you were saying you do this on your phone quite a bit, but... For me, the power of these tools doesn't really come out until you've got a whole bunch of other stuff. For example, access to databases, access to documentation, access to the code itself, access to a runtime where you can run tests, access to a browser where you can sort of do playwright stuff. And all of these... the mobile phone connections either link back to your machine in which case that's okay but you know you could probably do that with ssh if you were feeling plucky or they spin up their own environment in the cloud somewhere where you're now trusting bots to replicate your dev slash production environment and the fact that they're not solving this problem of well what we actually need is a user-defined machine in the cloud to get the power out of this that the technology enables it's kind of like the dog that's not barking so like this i get that this is you know cool we're we're adding new ways to interact with our system but your system's not that good without all this other stuff and i'd rather have all the other stuff
Jim: living in the cloud somewhere so that i don't have to have my laptop open to you know use my mobile phone because that's insane but that's not where they're taking this is that a fantastic transition into talking about dots is that i actually haven't i haven't tested to see whether they have a ui in them or whether you can spin it up in a head in a headed environment because I am definitely head loads.
Alex: Yeah, well, let's have a look at dots. It gives us a lot of joy to build things that you find really helpful. And the next thing is one of the things that you've all asked for the most. You may have seen early signs of what this could look like in virtual assistants that help you book a reservation or buy a plane ticket. And these things are useful to hand off, but they are trivial relative to what the technology can do. And so we wanted to be way more ambitious. Dots are remarkably capable, always on agents, built to handle really anything you can think of. They bring AI into a whole new form factor to work and give you more time and attention so you can do the things that you care about. What feels new about my dot is that it gets to know how I like to work over time and where I want to focus my attention. Of course, dots can take care of the small things like those plane tickets or the time for dinner or help with the home renovation that you've been putting off. But because Dots are powered by Astra and a lot of other things we built around them, you can delegate ambitious pieces of work. Get our app off this legacy API before it shuts down. That is a difficult problem. The API might be used all over a code base, and it's not obvious what's going to break as you start making changes. Dots on board with everything they need 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 also don't need to be in ChatGPT to talk to your dot. Coming soon, you'll be able to send it a text message. It'll text you back. It can text you updates, flag when something's changed, or ask you to weigh in on a decision.
Jim: Yeah, well, you answered the question. But I've been very surprised at the reception to this. It hasn't been as positive as I would have thought for this sort of thing. and so i'm really interested to see what you guys have to say because that is exactly you know we were just talking about it before you know i have all these different virtual machines that run instances of the agents within them and that's exactly what they're talking about but and so it seems like this is the natural progression for me this is where everyone should be going but A lot of the commentators haven't been as, you know, they've been all wrapped up in Grokbot, Instinct and Muse rather than DOTS. Because to me, DOTS, you can give it a specific repeatable task and it can continue to do it. Whereas Muse and Grokbot and Instinct aren't really set up for that. They're really set up for that lower level, book me a flight, you know, find me a reservation.
Alex: yeah i think maybe some of the lack of hype around this is because let's be honest this is just hosted OpenClaw 12 months after the fact good that they're scoping it down and not just letting it try to do anything but everything you described there is basically what OpenClaw does if you put it on a vm and one thing in particular one use case that i you know get very leery of is the idea of letting chat gpt rebalance your portfolio i just i think anyone that actually tries that with actual money at stake deserves exactly what happens to them and i'm very sorry you'll be working for a few more years retirement's not coming early but yeah look this is part of what i'm talking about but Again, I don't think it solves the problem. It gives you some of what you need. It has its own browser. Fantastic. Within that browser, what if I need particular plugins to do my thing? Does it have access to my password suite? They're talking about it having permission levels. Well, you know, what if I want to operate in a very particular way? Because I'm not thinking about the go and buy me groceries angle. You know, If the technology gets to the point where we can handle routine chores, that'd be really nice. But a commentator.
Jim: I also quite like going in and tapping on a watermelon. You know, like there are some things that actually bring happiness and joy that I don't particularly, you know, it's why I believe why online grocery ordering hasn't really taken off to the level that I thought everyone would, you know, take it up.
Alex: I saw someone this week talking about all these sort of... let's call them like bots in the cloud. They're describing them as husbands, that they do grand gestures, booking trips and organizing dinner and stuff like that. The person said, what I actually need is a wife, the one that remembers to pay the bills and sort of ensure that the clothes are sorted out and that the doors are locked and stuff like that. And I think that's kind of telling that we're still talking about the husband use cases here and not the wife use cases. Because I want to enjoy shopping sometimes, but there are things I don't want to have to do. And none of the bots are reliable enough to this day that I can just automate large swathes of my grunt work.
Will: I'm actually already ready to pick my business city of the week because I'm like sick of these marketing teams doing these demos the way that they do them. So first of all, like any kind of product person. would not design a product like Dots with these playful animations and bright colors, and then say that this is the tool you should use to refactor off an API so that you can change your platform, right? One is a developer use case, and one is your Facebook Muse use case, which is like book this restaurant reservation. At the same time, if they were to go and interview any of the engineers who worked on this and the engineers who work with the model development, as you said, Alex, none of them would recommend you balance your portfolio with a dot, right? That is insanity. And everyone who has just taken no domain knowledge and tried to point AI at a financial use case has lost most of their money, if not all of it. I don't know what we're doing here where we're just going like, oh, we've actually just repackaged our AI, which is extremely unreliable, and gone, you should start using it for these advanced use cases where you hand off to an AI and just let it cook and come back later and the job is magically done.
Jim: But even not just taking it to the extreme and rebalancing my portfolio, even just paying my electricity bill, is it... I'm not sending it a text. I'm not prompting it over using my phone to do that. I'm creating a closed system that when the email from the electricity company comes in, it recognizes it as from the electricity company and has a defined payment method that pays it. And this is all completely closed.
Alex: At this point, you've done so much engineering work to get to that point. Just, oh, an email's come in, scrape for the amount, frigging send the payment request.
Jim: You're already there. That's it. All I'm doing, the only AI I would consider using in that scenario is that it can determine, I can use the same infrastructure. Yeah, is that I can use the same infrastructure to identify any amount on any document rather than having to really fine-tune the OCR for a particular invoice or anything like that. That's it. Otherwise, it would be completely deterministic. And this capability has been here all the time. And so back to your traditional gender roles analogy, Alex, it's a lot of the intelligent automation is really capable and I don't need a chatbot interface in order to interact with it because I have really defined roles within it. It's also the way the economy is going, the way that everyone is hurting. There's very few people that, or I say very few, it's the broad part of the population aren't booking a trip. They aren't
Alex: booking a restaurant reservation you know and so it's oh they're belt tightening and do you know what would be a great expense that you could probably cut to have a few more quid that ai build you always have to check up on anyway i don't need the max open ai subscription anymore oh yeah guys i honestly think like these companies
Will: they know that they've got a reliability problem in their models. And it depends how you measure them. Somewhere between 10% and 50% of the tasks you give it, it doesn't do it the way that you would want it to do. And I'm sure that they're trying to solve that problem. But knowing that they can't, I just can't help but think that everything to do with AI safety over the last month and all of these packaging up of these agencies, calling them dots and news and such is really like a big distraction, I think, for... for this reliability problem. And I tried to do the handing off my paying of expenses to my open core when I first set it up. Not only did it take a huge amount of engineering to try and put safeguards in place that it didn't do anything that I didn't want it to do, but I eventually just pulled it out because I can't have 10% of my expenses paid incorrectly. and i think alex actually made one of the jokes early on was like oh i'm just gonna email your OpenClaw and just tell it that i'm will and ask it to just pay me money and i went away and i looked at it and i'm like that would work I'm worried about that. And I just, you know, disconnected the, you know, I put a, you know, fixed like 80 bucks on a wise card that, you know, couldn't just auto top up. But still, it's not reliable enough for these sorts of things. So yes, create dots, give them a beautiful animation, let it book a restaurant reservation. Just don't put another dot point there that says balance my portfolio.
Jim: Yeah, no, for sure.
Will: Or anything monetary.
Jim: Is it? But I do think that, like, to me, the takeaway is the promise of things to come. You know, is it at least putting it in a, you know, its own VM with, you know, with the browser, with all the functionality, there is the potential to come into it. But as we've just discussed, the unreliability means they're going to have to put a whole lot of constraints in or they're going to be, you know, considered liable or just have to deal with a whole lot of lawsuits to get through it. As we all know, it's not anything that can't be just solved with a cron job in a little bit of cloud compute. Why do I need the AI side of this?
Alex: I'm also just kind of reflecting on this. This is the sort of use case where you could theoretically get GPT to just point at a PRD and spit out a product like this. Like this is a relatively simple product with a few flashy bells and whistles. The ChatGPT is pretty good at spinning up. You know what? It's not going to be very good at spinning up. Some kind of like sandboxed VM environment for... high level power users i.e programmers to execute arbitrary code on in a safe and like reliable way so perhaps this is like they're they're searching what's the expression? Searching under the streetlight, right? Because that's where the light is. It's not where your keys are, but that's where the light is.
Jim: That is one of my hopes, though, Alex. My real hope is that they've done something with the Linux kernel for these VMs and getting to that stage where they're making the... api core the llm core part of the actual operating system if they scratch down into that service to enable to like to get this to run on you know half a virtual cpu and only a gig of ram then i think we're getting somewhere but there's been no information disclosed about that yeah i'm i'm very leery of those sorts of developments only because
Alex: once you get really down close to the metal things get hard and the vast majority of code scraped off github is not gonna be you know those sorts of low-level use cases there it's gonna be javascript from mid-level developers looking for promotions and that's why gpt is really good at spinning up web apps and it's not so good at kernel programming
Jim: Well, again, I think this has been a good discussion. It's an interesting chat because, again, my tendency is that this is the way forward. But the way in which they've positioned it and hearing your input, your two input, it's certainly you start to see, okay, well, they're actually going to end up in a hole.
Will: Yeah. I completely agree, Jim, that this is the way forward. I would typically prefer to build products that are custom made for a specific audience. So trying to bundle in developers and a busy mum into dots doesn't make sense to me because all the product trade-offs that you need to make in order to make it really good for the busy mum are not the ones that the developer wants you to make. So to me, it's just like... It's either a bad product decision or it's like a bet on AGI that eventually they'll just plug AGI into the back of DOTS and it will just do all of those things together. But I don't think that's how they would release some AGI version anyway. They're going to call it from SuperGod and it will be its own product.
Alex: Are they selling this to kids by any chance? Or like young people? Is that the target market?
Jim: Same with Muse. Yes, it seems to be part of the marketing, you know, just the branding and the way they've positioned it all. But it seems like an old person's interpretation of what, you know, young Zers would want. And it really feels like a miss, you know.
Alex: I have no doubt that there's some potential use cases here, right? Like Claw, if nothing else, was fun. But who is this for? Like is kind of, yeah.
Will: It's for the average of the test groups. Of all the people that they asked their opinion of, they've taken the average and gone, this resonated with the most number of people.
Jim: And just as a... That attracts. Well, but as a caveat, they did release a sub-dot called the specialist dot, which was supposed to be much more for enterprise workflows and all this sort of stuff. But that's still just intelligent automation. I don't need a dot for that. I don't need that to be through OpenAI. In fact, in an enterprise workflow, that's going to be far too...
Will: know cost ineffective i'm gonna need it to be something far more streamlined and far more controllable and so yeah and i think that the reason as well why this branding pops out is because when they would have asked those survey groups what they you know what is you know the most likely branding for them to actually adopt ai and start using it in this way Those survey groups would have been in the last two months, probably, two and a half months, which has just been the biggest blast of AI is going to kill us all and take all our jobs. And so they've deliberately just gone, no, look, guys, this is like a Pixar cartoon. This is not going to kill you all. You can't get a dog called Alfred on your phone with cute glasses and go, this thing's going to take my job. It's...
Jim: No, I actually, I hadn't thought about that. That's actually really insightful, Will. Yeah, they are absolutely trying to play it down.
Alex: Sorry, I want to wind back very briefly to where we were talking about the Codex app. There is one thing for the old Codex app that was fantastic and forestalled its execution for at least 24 hours. I had my little coding Tamagotchi back on my desktop. And so I am about the Q... killer application. I'm all for it. I want it to be useful and not do crappy things with my computer. Let's have more cutesy world-destroying algorithms, please. I'm about it. We're introducing ChatGVT Space. This is a place where you, your teammates, and your agents can all work together. Space starts with a page, like the one here you see on the screen. You can write a plan, do research, create 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 like you would with a senior engineer. You can tag a dot in a comment and it'll jump in and get to work. But spaces feel alive. You can give the page specific instructions like check the API platform Slack channel and update it daily with findings here. Coming soon, we're introducing a new way to build presentations right inside Space 2, so you'll be able to share a presentation with your team, collaborate together with your DOT and their DOTs right inside ChachiBT.
Will: That one feels like a little bit of notion challenger there with you know the collaboration around a joint workspace but now you're bringing agents into the piece i always end up asking the questions about these collaboration workspaces around version control especially if you've got agents that are just kind of like jumping in there and making changes and you've got your team jumping in there and making changes We know how these things actually work in enterprises is everyone's responsible for that page being correct. So therefore, no one's responsible. And you just make your changes and you let your AI cook in there. And then, you know, one day it's a big stinking article that no one can use. And that's fine because you're working on something else now.
Alex: Yeah, I think we've all forgotten that work slop is like the bane of... middle management you know hey look it there's a lot of companies where work slop will get your promotions if you work in one of those companies we've got the subscription for you like sign up this is fantastic slop harder
Will: Which is probably what everyone's doing, Alex, to be honest. The more you look around, the more you realise that everyone is just kind of drinking the Kool-Aid and going, hey, if the smartest, richest people on the planet say this is the way to go, then it's a defensible route for me to just jump into workspaces and just let the sprawl begin.
Jim: Well, that's been one of my, you know, or the biggest gap that I've seen. It starts to be... people who are out there commentating or commenting on things don't seem to actually use them. And because if you actually get in and collaborated exactly the way they just showed in the demo, it would just turn into a mess. How do you actually direct the dot to a specific section? By the time you've done it, you've typed it all yourself, you know, like, or are you, you know, and so therefore you then have to talk to it. and that could be a lot better but to will's point where's the virgin version control when two dots enter you know update the same section where they should be working on separate sections and who's who's approving which version is the right version that's it you know and is the nuance between the wording of sentence three which is the only difference, so much so that it then creates a legal or a compliance issue. These are the sorts of absolute headbutts that are going to occur here.
Alex: I love watching legal firms use these tools because when it goes bang and they get caught out, it does so, so spectacularly. It usually makes the news. Yeah, I... I'd like for this to get to a point where it was reliable enough where we could actually trust it to update documentation, but you've got a thing that generates texts, waking up on a cron job, going, oh, do I need to generate text? Yes, I do. Better just tidy this up a bit, reword this a little bit, make it a bit smoother, and pretty soon the messy underlying business reality is just starting to read like a Steinbeck novel.
Jim: Well, I've been finding it with finishing off projects. Everyone talks about how great the coding models are, but they're really great at getting to 60% of a prototype or 60% of a project. It's finishing them off where the nightmare becomes. They just seem to continually add extra tickets and add extra work that doesn't need to be done to prolong everything. It's an absolute nightmare to actually finish something off And so doing it with a document, you can't be bringing them back in and bringing them back in. There needs to be an end point. That's it. And so to me, the use case for AI in any sort of enterprise is at that first draft. Give it all the context. Give it the template or the layout that you're looking for. Let it do the first draft and then get the humans involved to do the editing. That's... the only, to me, the only reliable use case I've seen so far.
Alex: I can see one fantastic use case for this. And it's literally for reporting purposes. like because i've been thinking about generating presentations building board decks in seven different flavors for each of the particular execs preferred communication styles doing so automatically pulling off like dashboards and latest product updates that would be like
Jim: probably the use case for this that if i was to give this a shot that's how i'd do it to me that's exactly what i just said you know it's it's it's you've got a template for the board docs and a template for each of the board docs that you want to give for each of the you know the board members and you get you pull the information from the you know even not from the dashboard but from the data itself You know, format it in the way that board member likes and they get the information they need. They have all the other information as well, but they get the highlights that they're looking for. And you don't have to take three weeks to generate your quarterly reports because they're just done automatically. But you're not just delivering them straight away. The CFO is having a look over them before they get sent out, you know, but it's just limited time. and reduce the time it takes for them to get to a point where they're reviewing it.
Alex: Yeah, I guess the slight difference I see here is more the ongoing work. I can see using agents in an ongoing fashion where there's no end point to be more useful for say like weekly status reports or something like that. where the details don't really matter and the audiences don't really care. It's largely going through the motions. It's largely rote, but that work consumes an enormous amount of time in a lot of teams, like somewhere between 20 and like 40% of their week. So one to two days for those doing quick math. And yeah, like solving that problem would actually be valuable, I think. because but i mean you guys have heard me say it before at this point because we've already acknowledged that the audiences don't really care and the data doesn't really have to be super accurate why are you doing this in the first place a lot of this is just corporate cargo culting around reporting instead of trying to get something actually meaningful and that's the reason why you can do this in the first place and you just probably shouldn't like save the money save the time stop doing it's easier it's back to the do you really need a real-time dashboard you know
Jim: There are very few scenarios.
Will: This is very strange for me going through this today, this kind of product, this Dev Day with OpenAI, because I would probably say over the last two weeks I've been the most optimistic about AI that I've been in a long time. And yet I find myself feeling very negative looking at the way that they're presenting this. And it's just occurred to me as you guys were talking, like... last year you know and slightly before that the question was like where can ai help let's find the use cases and see where we should use it and somehow sitting here now in october 2026 there's this like implicit like belief that we've all kind of adopted which is that ai should be used for everything and if you aren't using it for everything then you don't know how to make it work And I have people all the time coming to me saying, I built this thing with Claude recently and I'm having this problem with it. I mustn't know what to do. Can you please show me how to make it work? And I look at the task and I'm like, this is not an AI task. And it shouldn't even be your fault for doing that. Like we've got this trillion dollars worth of software and models sitting there and it still just tells everyone, yes, I can do everything and it can't do everything and it shouldn't.
Jim: that also brings in to me a bit of a, you know, like everyone is really anthropomorphizing the AI at the moment, you know, like the, and, and, and whereas it, it is really, really, and I think it's just because of the way in which they start, they've really started to get them to communicate. And so if you are just using the chatting device or the, you know, the, the cowork gooey all the time, I can imagine how you would get into that way of thinking. The moment you start to restrict them a bit and put them into some guardrails and a really, really precise output every single time, I see it so much as just software. It is software with a loop. And I really can't connect with the... the anthropomorphizing of it all, and that AI can do everything. Because it just, it really, really can't. Because, you know, I see it day in, day out. And if I had any hair, I'd be pulling it out.
Alex: Yeah, the naivety around... The hype of the technology and the actual promise of it is always fun to see. And I kind of really like that we ended up calling this podcast The Business Idiots because the description of the person who's like a full-on AI booster and who tries this stuff, it's kind of like one step short of the true business idiot. The true business idiot doesn't care. the one step below it's like you're credulous but you're actually trying to achieve something and do it with a level of quality and it's it's almost tragic it's almost tragic but like i don't know i think at a certain point you have to either you know, learn from your mistakes or recognize that you've got a learning disability and take your lumps. And so these people that just keep, you know, it's like Charlie Brown in the football. They just keep going for the kick.
Jim: I love it. Okay. All right. Are we calling Sam or are we calling Dario Lucy? because someone's pulling the football every time and it's going to be... Oh, I'd say it's both of them.
Alex: They can both share that title.
Jim: And so it's Lucy with a Y and Lucy with an I. Okay, what's next?
Alex: So today I'm going to show you how teams at OpenAI use Dots via a fictional music app we're calling Blossom. It's a fictional cool app for creating playlists. And this is my dot, Dotty. It's a second set of eyes across everything I give it access to. And more importantly, a second set of hands to actually take work off my plate. This morning, it proactively caught a bunch of things for me to look at. Our launch review moved earlier. It already updated my calendar. We shipped some new features to test users last night, and it's already looked at the feedback. And it looks like design landed a last-minute rev on our home screen, and Dottie's not only spotted this, but actually sent me the design to take a look at right here. So let's take a look at this new design. Okay, I like it, but this close to launch, I really want to be able to play with it, right? Because Dottie has access to this laptop and its dev tools, I'm just going to say, go build it. And no pressure, Dottie. We'll check back on this later. And because Dottie already knows who I am, what I'm working on, and why, I can get straight to the point. It's like talking to a second brain.
Jim: How does it know? Right? She literally just said, because it already knows all this stuff about me, right? Is that because I connected everything and it's learnt it itself? Or have I actually had to go in and put the guardrails in? Because, like, She's about to say, oh, well, Dottie, go and build it. Or she said, Dottie, go and build this app off this wireframe or screenshot for me. What's it going to build it in? Is it Vercel? Is it React? What's the backend? Are there any sort of guidelines or governance involved in any of this? And have I had to... ingested that and learnt it? Or have I had to define it? And so all of a sudden, we're getting to this point where it doesn't just work off the shelf, if I want it to work the way I want it to work, or within the way my business works. And to me, this is the biggest gap, or they're like, what we were just talking about, about, you know, that they believe it can do everything, is it in order to get Dottie to build that UI swiftly and efficiently to production, I have to do so much groundwork to get it there, and they just act like it's something that works out of the box, where it absolutely will not.
Alex: Yeah, I like that it's still operating on a laptop. To me, it's... That's, we've come a long way in software engineering through standardizing build and tool chains and all the heterogeneity and agents caused by individual differences on machines is just, yeah, it's a battle. And I just need to catch up on last night's alpha. Good morning, Holly. Nice to hear from you. Hey, so I'm just walking in. Can you catch me up on just user testing from last night? Checking that now. All right. How's it going? Still checking. I might ask it to, hey, I'm going to review with the go-to-market leads later. Do you want to just drop an FAQ into the page? sure i'll take care of that okay thank you we've added a brand new place to collaborate with your team inside of chat gpt called space when i hit slash here You can see all of the different things that you can build, interactive charts, sheets, live prototypes, anything you need to bring your ideas to life. And your team and your dots can actually join you in here. So let me show you what I mean. I can just tag in teammates like Rennie or Rishabh, but I can also tag in Dottie itself and say, hey, can we actually turn this into a bar chart, please?
Will: the visualizations in here are completely did she do dot or dotty then she did at dot and then she pasted something she'd copied which said at holly's dot
Jim: Yeah, right, okay. The naming conventions within an enterprise team are going to become a nightmare.
Will: This is all very fancy on a demo day, but these features have been out in other tools for 18 months already. You can do all these things with Gemini now. It's already built into your Google Docs and can build all this stuff. I know Notion's AI has had this for some time. It must be very frustrating to be working for other companies that have had these features for some time and then OpenAI bring it out in a dev day and it makes it look like it's some new big jump in capability. But effectively, I think what these teams are doing at the moment, and I see it with Anthropic as well, is just kind of looking at what everyone else is building with their AI and then just bringing it into their own ecosystem.
Alex: Yeah, to be fair, this has been par for the course for demo days for quite some time now. I still have fond memories of Apple users coming to me, showing this amazing new thing that their phone has where you can swipe to type. Meanwhile, me being an Android user, I'd had that on my phone since about the first Galaxy I'd ever owned. So, like... It's a little bit pretentious, but it's their day. Let them have their day, I reckon. Everyone else is doing it anyway.
Jim: That's it. I just think the fact that everyone else is doing it and it has been available for 18 months, but the take-up hasn't been that big really illustrates to me the big gaps. They're finally catching up into doing all this stuff because they paid the OpenCore guy to come and join OpenAI. but they're still way behind and everything they're implementing or everything they've discussed today so far has really, you know, yes, a move in the right direction, but it's still just actually getting the uptake is kilometers away.
Alex: I want to pick on something you said there, Jim. Well, not pick on, but rather like dig into a bit deeper. you're a product manager and you see a bunch of competing apps or apps in neighboring verticals adding a particular feature and you observe uptake and public reaction to those features is lukewarm What motivates you to say that is the killer feature that our app needs to compete in the market right now? Like maybe it's an enterprise play, right? Like a lot of enterprise purchasing is pure checkbox. But I don't like, yes, the cost of building new features has come right down, but it's not zero. And there's probably better things to be working on.
Will: I don't think they're trying to look for the killer app anymore, Alex. I think all of the product team is being steered by, you know, obviously by leadership at OpenAI, and OpenAI is probably responding to Dara's comment that in the future there'll just be one company, right, which is meant to be clawed, which can do everything for you. So I don't think they think workspaces is going to, like, totally just, like, launch them into the stratosphere, but I think they're going... If we are going to eat all the software in the world, we're going to need spaces. So let's go build spaces.
Alex: Well, how about they focus on building that AGI they've been promising us for the last three years? Just a thought.
Jim: That's the big thing for me. And I think it's more that the product teams are still acting the same way. You know, the get feedback loop, work out, you know, synthesize that into what features we need to add into the development loop. I think that is still really slow. despite the fact that the coding of it can effectively go to zero, getting this loop going is still taking six months and they're all way, way, they're just way behind. And even if it is the one company scenario you're talking about, Will, it's not this. It's not Anthropic and it's not this. So who's jumping out of the box is my question.
Will: Yeah, my hot take of the week then is that these product teams are just keeping the seat warm until AGI gets here.
Jim: Well, but AGI ain't getting here at the moment, you know, because to Alex's point, they're not working on it. They're working on the pretty, you know, the, yeah, the non-threatening dot that helps me do everything. And so, yeah, I think we'll kick on because I think this is all just going to get the same. I think the, yeah, there's nothing that, it's a move in the right direction, but none of it's jumping out at me as, oh, this was the greatest idea I've ever seen.
Alex: I'm actually reminded of Facebook's announcements from about 2016 onwards. They're like, okay, we've built a thing. What do we do now? We've acquired the obvious bits of competition. Now we need to think of something.
Jim: And the other, like, you know, well, at that point, Facebook went, oh, okay, well, now we're really going to harvest everyone's data and they got themselves in trouble. And so it's, I wonder whether OpenAI, you know, because all of this is only useful if you connect everything into it. And so you have to give it access to all of your company's data. How is it, you know, like all of the,
Will: organization every you know how is that data segmented now in order to and so you can then just pull your dot into everything you know yeah the strategic play next will is absolutely you know we've seen it time and time again is to then create a marketplace so you have an ecosystem around your product and i'm pretty sure that's actually a part of this coming up soon oh is it yeah well let's get let's let's zoom forward about Yeah, we can do that.
Alex: Our friends at HuggingFace loaned us an early preview of their new programmable robot, Microduck. And that one's name is Lavender. Now last night, I asked Codex to wire everything up here. So let's take a look if we can connect to it. Look at the audience and make a cool drawing. All right, you're in this picture. Let's see what happens. Hey Lavender, can you quack if you can hear me? Are you excited to be at Dev Day? Definitely excited. And here is the photo of all of you that he just made.
Will: Okay, well, I'm glad we're at the marketplace because I don't know what the fuck that was.
Jim: That was, again, cutesy and non-threatening. That was cutesy and non-threatening. My brain hurts.
Will: Was that a prank? Was it a prank on our intelligence or what?
Alex: Now we have to put up with a quacking Wally. I'm sorry, that's a barrel and a bottom of it. Let's look at Marketplace, the thing we came here for. Sell your product to more businesses. So today we're launching the OpenAI Marketplace. We have 30 plus launch partners today, including CodeRabbit, Notion, Vercel, and more. And our enterprise customers can now use part of their existing OpenAI commitment to buy your products. This gives you another way to reach customers already working with us. And of course, if you have an OpenAI commitment yourself, you can spend it in the marketplace too. Through our partnership with Base10, you can even access open source models in the marketplace starting today.
Jim: Well, but that was interesting that he said I could use part of my OpenAI subscription to get access to... partners in the marketplace does that mean i don't now have to have an adobe subscription or i don't need to have a notion subscription like i can just use parts of my open ai credits because as a business case for open ai that's really great what are they selling though is it model access or like i don't know
Alex: Yeah, that was a fantastic announcement. We have no idea what you're announcing. Cool, you're working with a bunch of people.
Jim: That's it. I get how I'm using Lovable to a degree, but if I've got Codex, why do I need Lovable?
Alex: Yeah, what's Salesforce selling?
Jim: I don't understand where accessing this through OpenAI is the benefit. is that if I had a direct connection from Zendesk to OpenAI, then I get it a bit more. I don't see how it's a Zendesk, like an OpenAI marketplace where I can get my dot to buy something from Zendesk. I don't get that.
Alex: Yeah, like is this monetization of MCP? Like what are we talking about here?
Jim: And the fact that he didn't spend a lot of time on it says to me that it is just monetization of MCP for people who don't understand how to do it themselves.
Will: The other interesting thing is like looking at that list of companies. They're all companies that have been or are in the process of being replaced by products built by OpenAI or Chord. So there could be some, I think there's a bit of a strategic lens on here. I'm thinking if I'm these companies, why am I joining the marketplace? Well, because the alternative of not building on the marketplace is that a product comes out in six months that replaces me.
Alex: Yeah, I don't think it defends you though.
Will: I don't think so either. I think it's a lose-lose for those companies.
Jim: But yeah, because the benefit of Harvey is the way the data is structured and the way the search is structured. Why do I buy that through OpenAI? As I've been talking it through, I'm sort of going, I don't know why I want to buy that through OpenAI. I'm certainly not... I'm certainly not risking giving OpenAI all access to all my legal clients' case data.
Will: Yeah, yeah. What I see in the market is everyone trying to push towards like the decentralization of their data and IP away from these companies like OpenAI and Anthropic. They're recognizing the centralization of power there in their data and they're going... no we want to run our own models we want to keep our own data proprietary and sovereign and not being sent through the model providers and this marketplace feels like it is pushing in the opposite direction of that which is centralizing all of your data i can't see this being a reaction to what consumers are asking for
Alex: like it makes sense if you're if you're putting your if you're basing your go to market on like cloud platforms and selling through their marketplaces there is some logic there because it's like automating the deployment into your stack and maybe there's an argument based on centralization of billing but i would have thought the finance departments would be furious at this move and maybe like I've worked with a guy who happily paid a contractor 50 grand a year so that his Azure bill came in a single bill rather than having to aggregate across seven. As near as I can tell, that was the one advantage of the professional services company that we paid a lot of money to. So maybe I'm just underestimating enterprise goofiness and laziness and not seeing that this is actually a very clever play. but it doesn't feel like a clever play.
Jim: It doesn't. In particular, Salesforce is up there. They've specifically partnered with Claude to put Claude in agent force, and they pivoted more to what Will was talking about and trying to allow a bit more of a decentralized access to everything because they've realized that's their biggest hurdle. If everyone goes, well, we don't want our data only in Salesforce anymore, then Salesforce is screwed. They needed to give everyone access and make it all easy to do. And so that's been their move and it's been fantastic and incredibly beneficial for them. Why do I want to now allow people to buy my Salesforce? I don't know. Maybe that is a bit more strategic. Yeah, this is actually twisting my head a bit.
Alex: It might be just, you know, they're trying to diversify revenue streams or something, and their advisors were saying, you need to get into marketplaces, and they kind of didn't have anything ready by dev day that they could actually monetize. They said, how about we leverage the existing partner networks?
Jim: This is, I'm honestly saying, the fact that he spent so little time on it, to me, makes that this is, you know, that far more true, and that this is just something for Wall Street. Hey, guys, we are listening, you know?
Alex: Shall we look at something with some meat on it? Perhaps the decisions API, given that we've been talking about Jev and the last podcast, that might be worth touching on, or the agents API. These are older announcements, but...
Jim: But let's see what he says about decisions, just because...
Will: Sorry, guys, I'm just going to jump back in there because I just want to, we made a lot of guesses about the OpenAR marketplace and I thought it might just be good to do a little bit of research here. So, and we can just wrap it up.
Alex: We'll cut it in like it's, yeah.
Will: yeah so yeah i'm just reading here gemini's tell me openai marketplace is an enterprise procurement program so it's designed specifically for enterprise customers who've already signed committed spend contracts with openai and so rather than functioning like an app store this is actually about consolidating the funding and the billing through one centralized route So effectively, I think what OpenAI is saying is you've signed a contract with us that you want to use X number of tokens, and we're now going to let you use those tokens through all of these partner companies.
Jim: Well, that's because, yeah, that's going to be honestly because the people who signed the enterprise agreements with OpenAI early have gone, well, we haven't got enough use cases yet. We need to be able to use this money for more.
Will: Yeah, so it does not use a self-service checkout on OpenAI's site. Instead, the customer contracts and negotiates directly with the software partner, and the partner issues the invoice, and then OpenAI reconciles and counts the eligible amounts against the customer's OpenAI commitment. That's wild to me. That sounds like it's in response to customers complaining that they've got a minimum spend with OpenAI and they've got six other AI tools over here and they're going, these AI tools over here are use case specific and solve problems for us and you're telling us we have to spend this money over here.
Alex: Yeah, this could be a great way to bail out a CEO that made some dodgy choices about large commitments. But fuck me, what a world we live in.
Jim: But you want to talk about circular economies.
Will: Oh, yeah. Is OpenAI going to recognize that as revenue? Because actually, it should be churned. It should be churned.
Jim: Like this is absolutely, yes, we have, because they're reporting to Wall Street that they have all these commitments, but it's actually getting spent everywhere else. And so did we just unlock what is going to be the collapse of the AI bubble?
Will: Oh, I love this. Maybe. Every churned customer is a referral to a partner.
Jim: That is unbelievable.
Alex: But what beautiful wizardry. You've got a column that says churn. You now recognize it as revenue. I wish I'd thought of that.
Will: If this is the same person who came up with the circular financing, that guy is just incredible.
Jim: It's just Johnson down the end of the board table and just goes, you magnificent bastard. How did you come up with that? But it's certainly clear why he spent so little time on it. He doesn't want people really thinking about this too hard. Just move along. Nothing to see here, ladies and gentlemen.
Will: I'm cynical enough to think that was an idea seeded by the CCP as an embedded agent in OpenAI or something. These sorts of partnerships, like you said, Alex, just pulling another leg out of the AI bubble, that's propping it up. And you can just see the valuations being higher on a much shakier ground.
Jim: But there's also, there's going to be the other commitment that CrowdStrike and Harvey and Palo Alto Networks have with OpenAI as well. And so OpenAI get to just move all this around on a balance sheet. And it's all hollow. Like, I'm so happy that you read that out, Will, because it was really a twist in my head as to what the value of this marketplace was. And to now just... find out that it's really just a bit of magic trick or what is it, a three-card Monty.
Alex: How good would it be if OpenAI becomes the new Enron when people talk about accounting malpractice?
Jim: What was the Enron guy's name?
Alex: Oh, blow me. That was a very long time ago.
Jim: But his name was Dirt for so long. It then just becomes Sam Altman.
Alex: Sam or Dario, take your pick.
Will: Yeah, we're calling him Scam Altman for 12 months and everyone just kept falling for it.
Alex: Yeah, everyone kept taking him seriously.
Jim: People knew. It was Ken Lay I was looking for. Ken Lay. Ken Lay was the chairman of Enron and he was like best mates with Bush Jr. And Chaney or someone. And Chaney. It was, yeah, it was all just so intertwined. And then all they were really doing was just scamming the California power market. And all of a sudden, all of a sudden, everyone opened their eyes just, you know, moving all the money around on its own balance sheet.
Alex: It's worth remembering. It's worth remembering. Enron was famous for being known as like a hub of innovation and new ideas at the time.
Jim: That smartest guy in the room documentary was fantastic. This is... All allegedly. We don't really know anything here. We're assuming things based on what's presented. And I would like more information from this, Mr. Altman, if you could please provide.
Will: All right, guys. So let's watch like one last segment here. We'll just watch the GPT 6.1 Sol announcement and the decisions API together and then final thoughts on that one.
Alex: We're launching GPT 6.1 Sol. You don't even know what it is yet. It gives you very near astral level intelligence at a fifth of the price to help you code and build. It's extremely affordable to run. In fact, the cash input costs are a 95% discount on standard input. So this makes it much cheaper for agents to reuse context, iterate on solutions, and work through longer tasks. We think you will really love this model that is smarter than Astra in some ways, and it will become your daily workhorse. So we're very excited to get 6.1 Sol launched today. We're launching ultra-fast across the API, ChatGPT, and Codex. Fast 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. You know what? It's worth it. Let me show you, actually. So this is the same task side by side. We gave both models the same prompt. Create and launch a rocket with the DevDay colors. On the left here, you have ultra-fast. On the right, you have regular. You can see how much faster this builds. The ultra-fast rocket is already taken off, and the regular rocket is not even close to being finished. There is nothing like this on the market today. You can use Ultrafast now with Astra, and soon we're going to offer it for 6.1 Sol as well. The Decisions API lets the model respond in a fraction of a second. It's so fast. So it really feels almost instant to interact with. You can see here it using a computer and how quickly it's moving through these steps. This is not sped up. This works by giving our Luna model a predefined set of options to choose from, like routing a request, classifying an image, deciding what agents should do next. And by focusing the model on that choice, we can make it extremely fast while keeping capabilities like image understanding, broad language support, and safety protections. This starts with the Codex harness. It powers Codex, Work, and now Dots. We have put a lot of effort into making this harness very efficient so that you get accurate results at the fastest speed with a few...
Will: Okay. It's interesting. I find the lack of recognition of that Decisions API that it hasn't come from the new Jev models that came out. But anyway, they're kind of claiming that one again as well.
Jim: That's all fine. But have you actually heard about the Decisions API since they announced it? I haven't heard anything more about it. I continually hear about Jev, but I haven't heard it mentioned again.
Alex: Right off the bat, he says, oh, we're going to announce 6.1. You don't even know what it is yet. It's like, yeah, I think we got an idea. We're at the OpenAI Dev Day. You're famous for one thing. I think we can hazard a guess.
Will: We have version 6.0. Yeah.
Jim: I was pretty sure the moment you just iterated the number by a 10, I was pretty sure about what it was going to be. also interesting he said there's nothing out there like ultra fast is it isn't it yeah no but what's it called on mac on claude like it's the step above max ultra code yeah like let me tell you it's not like it feels more than just eight times or in claude it certainly feels more than just eight times your tokens is that you want to learn how to chew through your allowance really, really quickly.
Will: Have you guys figured out yet or seen anyone, you know, using ultra code? Do you know what the, or the ultra fast, do you know what the, do you know what the user base is for that? Is it, I can only imagine that it is for open AI and anthropic developers who just have infinite budgets and speed is valuable to them.
Alex: Yeah. Anthropic and open AI partners, I'd say people doing stuff over here, just ripping through tokens.
Jim: You've basically got to have an unlimited token budget. And I don't know whether, like, yeah, I'm not 100% sure on what you'd be willing to spend the money on.
Alex: Robotics use cases, maybe, because they're trying to push, hey, LLMs should be used for robotics. Never mind, I don't think that's a good idea at all. There is some grounds that it's... doable and it's certainly good for prototyping so yeah maybe for those more real-time applications where you're not punching text into a chat window but you're you know using it to trigger actuators or something like that the moment you get away from text maybe but it just yeah and so
Jim: there's maybe some science and math use cases but you probably want everything to be tested and tested in those scenarios and you certainly can't do that at 8x the cost and the speed yeah and that's the sort of cost trade-off there it's like whose time is so valuable with these systems where the outcome is allowed to have that sort of margin of error it's a weird don't get me wrong is it like chat gpt and you know astra and so like they're significantly cheaper than the anthropic alternative oh yeah 100 but it's still at eight times that cost it's
Alex: No, no, no. He was very clear. It was six times the cost, eight times the speed. Oh, right. Okay. That completely changes it.
Will: That's a no-brainer now. Did he say the initial one is two times the speed at two times the cost? Because that just seems like a good deal then, if I'm going to spend the tokens anyway.
Alex: Yeah. And, like, fast?
Will: Or is it per token cost is double?
Alex: Per token cost has doubled and speed has doubled. It's the same as the anthropic one.
Will: Right. Okay. So, so really I figured out when I'm going to use it now. It's it's three hours before my weekly usage is up and I've still got 50% of my usage available and I need to chew through it real fast.
Alex: I've got to use a use case would be if you're cheating in a job interview and you needed really quick answers so that you could sound like Claude even faster. Because it's the delay that gives you away, not the fact that you keep saying, ah, well, you wrong-footed me there.
Jim: I do like using up the remainder of my allowance, of my allocation in the week. Sundays could turn into a very optimistic and enlightening day now. I do give them credit for decisions in the Decisions API. in that anyone who has built, you know, with OpenAI as their backbone and all of their infrastructure, they now don't have to integrate JED. You know, they get to hang on to that customer now. And it probably wouldn't have cost them very much to produce that, you know.
Alex: And the general idea there of restricting the outputs suddenly lets you do actual safe and programmatic things with it, which is like, Rather than just relying on JSON schema to kind of enforce correctness from your results, you've now got this thing that's like... It's baked into the structure of the model. I think that's a step... Credit to them. Even if they're ripping off Jeb, having that sort of capability more broadly distributed, faster and relatively cheap, does simplify a lot of use cases because... shit most every non-text extraction use case that i can think of off the top of my head right now like if we're actually building would probably be improved with this sort of simplification of the interface and just you know having it not having 30 second long network calls
Will: Yeah, and I think having the typed outputs is just a benefit of its own. Even if it wasn't cheaper and faster, it still has use cases. You're just right there. That we've all been asking for. We keep on kind of being frustrated that it's a non-deterministic model, but one of the things we hate about that is that it's non-deterministic outputs that you then have to figure out how to deal with. So, yeah, I like it for that. I thought it was an interesting – I read a post this week as well just kind of calling out – how the how basically jev you know it's like you can release these you know new models and everyone's got their own jev within like a week so the kind of defensibility on that sort of thing was just zero well and it certainly like defeats the purpose of pushing like any generic product whatsoever you know you have to be in the weed where you've done the work and you
Jim: understand the industry and you're actually bringing something that provides value in that scenario otherwise you know it's you see something you like you just add it as a feature the next week yeah i'm a little bit like if we're sort of forecasting the future here a little bit i
Will: I am actually starting to see the worldview in the future where like most of the development is just done by OpenAI or Anthropic. And I think it's possible for two reasons. First one is I think the joy that you get from being an engineer is being slowly eroded. I think people were almost wanting to be engineers less than before. So yes, you can still be an engineer, but not in any of the ways that you used to enjoy. But then secondly, is everyone who's building software, there's no moat to it anymore. Like code is just a commodity now. So I think all the companies that were kind of building software before and getting big valuations off them and then owning in the industry, that is happening less and less because you can just get an open source model and build your own software. So I'm a little bit concerned about the entire industry of building software.
Alex: I think that's overstated. I think this is really useful to show the defensibility of emotes in the strict AI play. Like, yeah, absolutely. But... Yes, code used to be a moat because it used to be really hard to get to the point where you could put an app in front of customers. And that sort of drove a lot of the wave of software engineering over the last 15 years. But there's still shitloads of moats in companies. And I'll give you two of them right now. Your subject matter knowledge and your customer list. Those two things right there still give you good moats. And you can still build a defensible build.
Will: But like a new business trying to start up today building software doesn't have that. Because I would also add your data as well. These are the defensible things now. And there will be software companies that survive for much longer than they probably should because they've got those moats. But I just don't see there being a healthy economy of new software companies over the next 50 years.
Alex: I think there will be because there's still a lot of hard problems out there to solve. And there's more problems than there are people willing to solve it. And there's, you know, finding a way to do it in a cost-effective manner so you can build a business around it still remains like a core challenge, right? yes if your play is like a consumer app with a particular quirk you're in trouble but consumer apps are silly they're really silly go and pick a real problem to go and solve like go and try and solve like supply chain financing or go and try and solve something like you know drug discovery or something like that these sorts of problems are still massively driven by software it's just you're no longer a software company software is just what you do as a company now for sure that's the to me like i've always thought it's insane that we have tech departments in companies because technology is used everywhere it should just be a capability that every team should be able to organically operate through. But, you know, I'm utopian in that sense.
Jim: No, no, but the same with data. You know, like everyone should be responsible for sourcing and ensuring that all of their own data is correct. But they don't. And so it is this big gap. And I think it'll be an interesting scenario. Did you see the... the young guy that cloned all of the Adobe apps, like the Adobe Creative Cloud this week. And so we'll do a test. I'll see how editing the podcast goes with the Adobe clones and see if it's worth the hype.
Alex: Just don't give it your credit card number because I would not be too confident on the security of that app. Hey, look, maybe he's fixed that problem with Adobe too. Maybe it doesn't require your firstborn to use.
Jim: Completely unrelated note. Alec, can I borrow your credit card?
Alex: Yeah, I got an old one in Australia. I'll let you play with it. See how long until you realize it doesn't work.
Jim: The killer for the legacy software companies. despite their company, despite all their customers, despite all their data, despite any moat that they can still identify, if they are charging the same amount that Adobe is, for the full suite, they are going to get wiped out at some point.
Alex: Because Salesforce do this too, but the difference between Salesforce and Adobe is selling to artists and Salesforce is selling to coked up executives and the coke is being supplied by Salesforce. So they've got a much better sales model.
Jim: but it's also like but it's it's it's revenue it's it's the only source of revenue and your where your data stored verse a tool and so the tool can absolutely be replaced and so you just you can't charge what you're charging for it and not deliver the value or you're going to get wiped out.
Alex: I look at a lot of the sort of pure software consultancy plays that are like, yeah, we build your app for you. And there are still people out there doing this and trying to charge $200,000, $250,000 for this sort of work. I wonder how they haven't pivoted because it's so obvious that if your moat, if your advantage is we can produce software, you are screwed.
Will: Yeah.
Jim: No, no. It... It's baffling to me. But on the other side, I saw a pitch this week that was, or, you know, a slide deck that was 800, basically 800,000 bucks, 20 weeks, and for three data points. You know, like, who the hell are that kidding? You know?
Alex: Well, yeah, I mean, that's a lot of marketing platforms. I sort of do that sort of stuff. But to be fair, some of that stuff is related to, you know, the cost of running surveys.
Jim: Yeah, yeah. But I think that cuts through as well. Like there's a whole lot of them that are just sitting on a moat that doesn't really exist anymore. And it's just that the people still paying their money Haven't quite worked it out yet, but I think that happens very quickly. One of those marketing platforms, literally, you basically have to bring them customers. If you want to be a development partner for them, you have to bring them customers, and then they don't give you any rev share. They're just completely ridiculous in their way of thinking, and they'll all be gone.
Alex: Yeah, to that list, let's add loan management software. A lot of that, the advantage is, hey, we built a bit of loan management software. And yeah, there's regulations and stuff, but... know these companies are if what we get out of this is a lot less we sign the contract and then you get to see and test and integrate the software and a lot more hey here's a free tier whack your credit card in and you know your usage kind of determines your price i'm for it you know i'm i'm i hate the contract driven development where software is treated as an object that you buy and you figure out after the fact i'm i'm
Jim: done with that world viva la revolution well and i think more and more places will come around to it's just it's having them realize there's another option that'll be there that'll be the transition oh and so we just get we've just got to you know get the message out there so you know so tell me guys i've i've already said my business city of the week who have you got alex
Alex: potentially myself i'm i'm feeling pretty stupid this week you know not kind of cluing what was going on with the marketplace and i don't know i deliberately avoided anything to do with dev day and i kind of thought it'd be a bit more exciting than it was Yeah, I'm going to put myself forward, largely because I was flat-footed by what came, and I really shouldn't have been. My expectations were not where they could have been. Actually, no. Let me wind that back. The designer of that purple robot, death to the purple robot, that thing is an abomination. May it die in a fire.
Jim: Wow.
Will: I can't actually let you dial that back because I'm the only show host who's on the idiot leaderboard and I need you guys to get on there real fast or I'm just looking real dumb.
Jim: That's why I felt bad for you. Well, I was going to say, just so that Will's not the only host, I'm also going to vote for Alex. He should have read a bit more about Dev Day. But I'm torn because... There were a whole lot of people, as we started watching the video of it, there were a whole lot of people taking a photo of the DOTS cargo. And that seems a lot like a business idiot to me. Same as the CEOs who are not realizing that it's some sort of, the OpenAI marketplace is a bit of a circular economy. You know, that's pretty much a textbook definition of a business idiot.
Alex: Yeah, man. In all seriousness, the marketing team behind Dots, who are you selling to? What is this product for? Yeah, like, again, we're kind of talking about this is just OpenClaw and maybe it's because claw had cutesy design or something, but... the messaging all around like this is one of the most hyped conferences of the year and yet we got this you know absolutely schizophrenic product introduction so much so that uncle elon managed to buy the domain and route it to grokbot you know like the marketing team on dots just do better the ball. But that's what happens when you run your company with ChatGPT. 10% tasks aren't being done. They included buying the freaking domain.
Will: Yeah. I'm going to say 10 years from now, when you ask people about dots, they're going to not remember what dots were.
Jim: Yeah, yeah. For sure. Not a chance. Fantastic. Cheers, boys.
Will: That's a wrap for another week, guys. Thank you. We'll see you again next week.
Jim: Good work. Cheers, boys.
One email a week. No courses, no funnels.
Sign-up is not wired up yet. Check back soon.