Ahora todos los agentes de IA tienen ordenador y memoria. Entonces, ¿qué vale $100 al mes?

[0:00] The agent wars are here. Open AAI just launched DOTs, an agent with a computer, a memory, and the ability to keep working after you log off. Meta offers the same basic setup with Muse, including a free option, which Dots does not have. Your first Dot only comes with Chat GPG Pro starting at a 100 bucks a month. So, are you going to pay? Why pay? If everybody can copy the computer, if everybody can copy the memory and the background work, what makes you actually keep an agent around? Why would you hire an agent when there's a bunch of them
[0:31] out there? And my answer is the job. It gets right for you. I'm here to give you the full story on what happened, why it matters, and why do is both super interesting and also might not be the most important release of Devday longterm. I've been using Dots a lot. Here's what it did for me. Open AAI's new agent caught a recurring meeting that kept landing on weekends. Do you really have this meeting on Saturday? And I've been just too annoyed and too busy to clean it up. And no one comes to it. We all know it's a zombie meeting and who likes weekend meetings? And so
[1:01] it just cleaned it up. It just said, "I can take care of that for you." And it rebuilt the invite and cleaned it up. It also spotted a conflict between an existing appointment I had at the good old department of motor vehicles and my travel plans. And these little things, they might seem small, but they're part of the bigger promise of dots. It's the idea that the context you've built with AI starts working for you. And they have a lot of context, right? Open AAI is operating a 1.2 billion user product.
[1:31] They shared that at Devday. They are in a position to be very successful with a product that is focused on AI context. So, I've been using Dots. I've been using space. I've been using the new Soul 6.1 model that they announced. And if you're trying to figure out what plan to pay for, where your team ought to work or what you could build with maybe a smaller team, these are the announcements that matter. The wider announcement list is on the screen of course and you can go through some of those details yourself or watch the live stream which is a lot of fun and they saved it so you can watch it again. So
[2:02] here's what OpenAI announced. First dots are agents that you can give ongoing responsibilities to. Space is a shared place for people and AI to work together. See that context thread coming back with documents called pages that you can edit together. Now we're getting into Microsoft territory. We'll talk about that. GPT 6.1 Soul brings you more demanding work within reach of a cheaper model. So GPT 6.1 Soul is absolutely an efficiency play in a world where we're
[2:32] asking AI to do more and more. So we'll talk about that too. Codeex got reusable cloud environments for software development which is a big deal. Like if you are one of those people who are sitting around carrying your laptop with the screen half open as I see a lot around Seattle around San Francisco rejoice because now you can put it in the cloud right like but that's kind of the idea there and developers get tools to build applications that use agents and operate inside chat GPT which I think is actually a huge entrepreneurial opportunity and we should talk about more so we'll get into that as well
[3:02] there's so much to get into. So if you want the quick what is the dots deal, why are we talking about dots and agents, should I buy a dot, etc. So the quick buying answer is that your first DOT comes with eligible pro plans starting at a 100 bucks. 200 is going to buy you more usage. 500 eyepopping I know buys the largest allowance and of course Astra ultraast which is basically like speeding down the highway on AI. It's insanely fast. You don't need the $500 plan to get a dot. don't think you
[3:33] do. And after using Soul 6.1 a lot, I would seriously look at which model you're running a lot before assuming you need to buy more capacity. I've run GPT 6.1 Soul and I've seen firsthand how it feels a lot like Astra and it barely chews my plan allowance. It's like I can do complicated documents, I can do code work, I can do research work, and I'm dropping one or two percentage points in my overall weekly plan limit for OpenAI.
[4:04] And that's been one of the unsung heroes of this release for me because for those of us on paid plans, if we're heavy AI users, what we do every morning is we see the agent did the work and then we go and check and say, what's my usage? In fact, there's a whole popular series of plugins that are all about checking your usage on AI because that's the question now. And so far, it's been a very encouraging release. Soul 6.1, not six, 6.1 makes me want to move more and more and more work over to it and reserve Astra for jobs that actually need. And that matches the public
[4:35] reporting, right? OpenAI reports that Soul approaches Astra on selected evaluations at a fifth of Astra standard input and output token prices. So, of course, it's not going to chew your plan limits as much, right? It just it's not as expensive. So, we'll get into the details, including a change existing $200 subscribers need to know about that one was less popular. We'll cover that. But anyway, if you put all of these together, you can start to see OpenAI's bet in the space. Fundamentally, they think more of your work is going to stay in Chat GPT over time. So, the AI has more and more context for the next
[5:06] assignment. Now, in that world, a more affordable model like Soul 6.1 or whatever future models they release, it's going to let you give it more work and feed that flywheel further. And a shared workspace is going to let colleagues build on what you've already done without leaving the chat GPT ecosystem. They are building the future Microsoft Office competitor. That's what they're doing. They are going after that same space of collaborative work. And funny enough, as I think about that in the strategy I'm seeing, the distinction
[5:37] I keep coming back to is form factor versus utility. So if you think about form factor, form factor for agents is converging. All the agents are starting to look alike. In fact, there was a little bit of a Twitter fight between OpenAI folks and Grock folks after the launch of DOT because Dots does look similar to Grockbot. There was some beef. But in the middle of all that, I think we're missing the point. Utility in my mind is not converging. And I know that's a bit of a hot take, so stay with
[6:08] me. But just because you have a computer and you have memory and you have background work, it doesn't mean that you have the same harness behind the scenes or the same model, and it doesn't mean you've tuned that agent to serve the same use cases. In fact, what we find in actual agent usage is that different jobs emerge for these models, different competencies. And so the utility of the agent is not the same even if the form factor is. And I don't think we're talking about that enough. These agents are being marketed very
[6:39] broadly. But I think they're going to win through specific jobs they do extremely well. So Muse has a free entry point. I've been using Muse a ton. I had a whole video on Muse. And when I think about Muse and how I've used it and how it's shown and how it's done really well, it does a lot of practical, relatively low intelligence tasks really, really well. Practical help with email, practical help with small shopping. I've used it for date night. And then the killer feature of course is making calls for you. When you don't want to do life admin and figure out how
[7:09] to call the dentist, it will do that. With an assistant like Instinct, it's different. Instinct is more proactive than Muse. Instinct leans into this idea of travel a lot more. In fact, if you listen to the Instinct founders and what they talk about, a lot of what they talk about is transactions related to travel and dining. And they recently shared in an interview that 40% of users add a credit card to an instinct vault within a few weeks of signing up. So, they are
[7:39] definitely leaning into the monetization piece and they're leaning in, I think, a different demographic fundamentally. Now, back to Dots. With dots, I'm watching what happens to the context you've accumulated while doing serious work with AI. Dots are for work. I muse an instinct I shared about because they give us a sense of the breadth of this space. But both of them are personal agents. I believe in a multi- aent future. I don't think one of these is going to be the winner, but I think we have to understand what they do to use them well. So the overlap in tooling
[8:11] with these agents is real. Muse also has a virtual machine. But watch the job that you keep coming back to. Watch your actual habits. And that's what I keep emphasizing. Look at Dots as the first in a longer play and expect that this is an iterative product. Dots is going to evolve as they get data on what users are doing. In fact, we see that iterative process with Muse already. Muse launched. Muse did relatively well.
[8:41] passed 5 million downloads now and they are immediately launching small business tooling because they see that behavior set emerging on the platform. So look at dots as the first release in a series of things that are going to expand the way you can apply intelligence to work context that you probably already have in chat GPT. Now a dot runs on chat GPT6 Astra. has its own cloud computer and its own memory and you can communicate with that dot anywhere. Like I was able
[9:12] to do it on the web through chat GPT. I was able to do it on mobile. I was able to do it on Slack. I was able to do it in codeex. You connect applications you want to and you give it responsibilities. I'm a big believer in the value of context for useful AI work. I've talked about it a ton and I want the context I've built with AI to really be easier to access for useful work. and OpenAI is making this play essentially to make context intelligent and that's the part I think has huge potential and right now the small jobs I'm getting
[9:43] help with feel like the beginning of that story and so one of the things I'm going to be paying attention to as I work with my dot more is essentially how does D dots the product start to distinguish salience in context which is a fancy way of saying if dots knows a lot about me and it does because I use the product a lot How does it know what really matters to be proactive about? And that's something where I'm seeing a lot of recency focus from the dot in what it's thinking about in the proactive suggestions. It wakes me up to and yes, it really does do that. And so
[10:14] I wanted to be better at getting to the important things. And that's something where I think it's going to have to either evolve as people use it more or I'm going to have to feed it more. I'm going to have to work with it more. Some of these product features, it's not as simple as saying make it that way. It's actually more about figuring out how you work with the model capability you have to maximize the value you're looking for. And that's why I emphasize skills so much as like agent working skills are the future. Like it is the meta skill we
[10:44] all need to have. Now speaking of context, there are boundaries to what DOT knows. And so even though I've been saying the larger play is intelligence and context, that doesn't mean that dot immediately knows everything I did in chat GPT. In fact, does it? All it does is it receives your chat GPT memories and it can form memories from connected applications. And so I think that's a very interesting choice. It's actually a privacy favored choice. I think they did
[11:14] a great job here because it means you can disclose to a proactive agent what you want to from your context in chat GPT and I think a lot of people we look for agency we don't want you know a agents to just suck up all our data and so having the choice allows us to use dot in a way that works for us and I think that's something that makes me want to trust it more like intuitively as a human I'm like okay so there's a boundary here and I can choose when I walk across it and share information and so if you're getting started with dot I would basically Look for something that
[11:45] is going to keep you coming back and keep you conversing that is an area of responsibility. If you use chat GPT or codeex at work, I would focus on a work application because frankly you're paying for this and so you want the ROI on this and you want it to be valuable to you at work and yeah, if the personal stuff comes and people do use it for personal stuff too, that's great, but you want to make sure that it actually works at work. And so in that sense, part of what you're doing when you connect it up to Slack and other things is you're picking where you want it to
[12:16] focus. Do that deliberately and make sure you pick something that is going to be recurring so you can start to push your dot to learn the task space and get better over time. OpenAI's launch example is useful here, right? It's a product launch. It changes scope. It's a situation that I also talked about when I talked about Claude Opus. Like these are the kinds of situations agents can engage with whether they come from OpenAI or Anthropic or other places. In this situation, the dot understands the audience and the direction because it's
[12:46] got the context because we've allowed it to in Slack, etc. And it prepares revisions to launch materials proactively. And I think when we talk about intelligent context, I see people's eyes roll back. I want to explain really clearly what it means. If you have a delayed feature, the announcement needs to change, the sales deck needs to change, the presentation needs to change. Essentially, all of this stuff cascades downstream from an unpredictable trigger. In that world, the value is knowing which promises depended on a particular feature that
[13:18] changed and having an always on agent that can read that and intelligently respond and proactively develop adjustments. That's really where DOT starts to shine. And that's also where I think we're going to start to see differences versus the autopilot feature. Remember I talked in the Microsoft video about this idea that intelligence has saturated out business use cases. It is notable to me that OpenAI chose to put their strongest model Astra on dots and we'll see how
[13:50] they evolve that over time, but that's where it is for now. And that means that they believe there are frontier intelligence use cases you can give your dot that would not be as well served by an earlier generation or lower intelligence model that they have in the stable right now. I am curious what those use cases are. I would love to hear from you what you're using your dots for. If we circle back out, if we look at that application, how a proactive agent knows how to add value. There's a tremendous market out there. If you know AI, if you watch this video,
[14:20] if you are excited about what you can do from a small business or side gig perspective to implement AI for others, there's businesses out there for this. If you know an industry where outdated customer promises cause expensive problems, helping companies keep those promises aligned with reality is a service you can do. And that's just the simplest and most obvious one. DOTs basically allow you to sell proactive intelligence into companies that often have chat GPT but don't have the time to get the value out. And if you look at
[14:51] the macro picture of where the industry is going in the last 30 to 60 days, you look at Dreamforce, you look at Dev Day, you look at the launch of Muse, everyone is realizing that we need to get value out, not just activity in. And so much of the story when agents were not as smart was about can we get people to do activities slightly more efficiently and now it's about value out. And so that's where I would encourage you to start thinking if you're thinking about dots and you're thinking about plugins which was also a major launch at dev day and
[15:22] you're like there's an opportunity here to get people using stuff and get value out of the product that I'm selling or out of the service that I'm selling. Think about it from an outcome perspective. That's where the market is going. Now talking about Microsoft, we don't want to forget the most direct competitor to the knowledge work category that OpenAI launched. Think about spaces. Think about pages. Now these are literally collaborative digital spaces where your agents and people can work on a document that's live and on ultimately a space for a
[15:53] project that's live. And so in the launch example, the salesperson can flag a promise made to a customer and the product lead can explain why the feature slipped. I love that they make the product person explain the delay. That is such a product thing. Uh the next revision can use both pieces of those information together. The next change arrives, the plan, the discussion, and the people who can correct it are all there in the same space. I think spaces is going to be a really interesting product to watch. In these kinds of situations, you almost don't want to
[16:23] look at the feature set. You want to say, okay, the feature set is fine. It's great. What organizations are going to use this? And what org types and team types does this imply? Like think about spaces and pages. It implies a very flat organizational structure. Something similar to the labs. It implies a place where people feel really comfortable doing asynchronous collaboration in digital spaces and having agents involved there as well. It's a very AI fluent space. space feels like it came
[16:54] back in time from 2027 and is talking to us now as a launch. It feels like space is essentially presupposing AI native and AI fluent organizations that I do not see very often outside the valley. And so I think it's going to be really interesting to see who adopt bases and how quickly they're adopted. And in this situation, you know, Microsoft has a natural advantage. Autopilot may not run on a frontier model, but autopilot is going to be trusted because it's Microsoft. And I think that the idea of having agents and humans just
[17:25] collaborating in project spaces is going to give IT administrators at a lot of other companies question marks that they don't fully know how to answer. That doesn't mean it won't do very very well. And that doesn't mean that OpenAI doesn't have great answers to those questions. It just means that the no one got fired for Microsoft argument is not going to apply here. Enterprises are made of people too and when they make these vendor decisions they also think about what is the least risky option and so I think that the question of spaces is going to be can you in that corporate
[17:57] strategic environment make it clear that spaces represents a new way of working together make it attractive by continuing to invest in it which has been a challenge for open AI at times because they you know Sora didn't last etc etc these are bets that they're making and they've been pretty transparent that they can make these bets because they're AI powered and they can put a lot of bets on the table and that's great, but then you have to be able to double down on it to make it stick with the enterprise. And I think that's the question I have and that's what I'm going to be excited to see because I think there's a lot of potential. Essentially, if you've been
[18:27] doing a ton of work in chat GPT by yourself, OpenAI is giving you a reason to bring your colleagues into it, right? So, space and pages are available right now on pro plans. So, if you're an individual, they still are available to you. Enterprise plans, which totally makes sense. Collaborative slides and spreadsheets are coming next. So if you're like, does it do slides and spreadsheets? The answer is very soon. And that's where I think the business and enterprise piece is going to be really interesting to watch as they continue to add to that feature set. It makes it more and more compelling to say, we sell frontier intelligence. We sell this space where you can work. We're the whole package. So business and
[18:59] enterprise will also get team tasks, which is a really interesting feature. And you can do recurring assignments that way. And you can do chat GPT participation of course in Slack or in Teams. And by the way, if you're like, this is definitely a competition between Microsoft and OpenAI, I would be remiss if I did not call out that there's also a partnership between Microsoft and OpenAI. And so you position it as tension, but you resolve that tension by partnering together and distributing together so that the whole solution is available to a company. They can choose whatever AI they want and it can be available wherever they want it to be.
[19:30] Because ultimately companies want outcomes and results as I keep saying which means if they get a simple package that is something that they can just go to and say employees you get everything and you can pick and you have to be productive they may go for that. Now of course no work is complete without meetings and OpenAI leaned into that with the meetings plugin at Devday which allows you to save summaries. It allows you to save action items into space. Initially it's in a MacOSS beta for pro and business users. I'm sure they will expand it soon. And really the idea is you start to connect the meeting, the
[20:01] conversation, the document and the work all together. So you can see a world where the meeting is happening, the transcript is developing in real time, the dot is there and your agents are there in the space and they're starting to develop the assets as you talk, right? That feels very futuristic, but all of the building blocks are in place for that to happen. And I think that's going to feel absolutely magical to a lot of companies outside the valley who are used to using co-pilot for email. Like I think there's a huge quantum leap
[20:31] in what AI can do that is about to hit like the jaw-dropping factor of it can do all of this while I talk is absolutely there. Now if you're wondering do your private chats do your own memories just get shared with the team that is not how that works. Like they keep a clear boundary there. So spaces is only for shared work. Now, that presents an interesting challenge and an interesting new skill for all of us who are AI users as we get into and this is not going to be the only collaborative AI native piece of software out there. There's going to be a lot more of these coming shortly. And
[21:02] as I argued in my piece on AI native workspaces, somebody's reasoning, the reason you make your decision needs to help the next person doing the job. So if your colleague is the only person who remembers why the decision was made, just storing the latest dock in spaces doesn't solve that rationale and that tribal knowledge problem. Like we have to think intentionally about what information we bring in from our private AI work into a team space in a way that
[21:32] would be useful. And so you should be thinking about is there a decision making that you made in a chat that you need to bring in? Is there reasoning about something that you want to bring in? How do you want to bring that in? That is a really big emerging category. We don't have norms around it. We don't really understand how it works. We don't have expectations set. We don't have tools for it. All of that is something that's going to be a huge emphasis in the next 6 months or so. Now, if you're wondering where Anthropic is in all this, they are absolutely in play because Anthropic, of course, is a
[22:02] heavily enterprise company. They're an AI native lab. They're frontier company. So, what are they doing here? Well, they're combining chat and co-work into a single cloud experience and they're rolling that out to pro and max users. Cloud co-work sessions have this that that same really handy cloud capability and for prepping documents and spreadsheets and presentations. Claude absolutely is in the conversation. In fact, I have been loving using uh a combination of Fable 5.1 uh for really thoughtful, intuitive work and Astra for
[22:33] chewing through stuff. And I've also been using Opus 5.5 really heavily. As I keep saying, I don't pick favorites. I use a lot of AI. And Anthropic even has something a little bit like dots meet spaces already. So for teamwork, there's Claude Tag, which launched in June and is in beta for team and enterprise customers, and it works in Slack. It remembers relevant channel information, and it can schedule work or follow up proactively when enabled. And so if you compare that with kind of where team tasks are going, where Slack integration is going with OpenAI, there's some
[23:04] overlap there. Like product shapewise, they're attacking similar problems. Again, I'm going to say these all assume a degree of AI fluency I don't see very often outside the valley. And one of the interesting questions in the space is does launching with much fanfare a great product help to juice that adoption and help to get people thinking differently and maybe reconfigure their teams a little faster or is it going to take a bunch of forward deployed engineering work, a bunch of forward deployed change management to actually get it done? And
[23:36] it it's looking to me like essentially these kinds of events, these dev days, these pops in popularity, these high-profile events, what they do is basically get you to understand the the high level and then all of the dirty details do have to be worked out with actual implementers because it's complex and every organization has a unique team fingerprint. And so what if you're a more traditional team? What if you're not ready for this full-blown collaboration? OpenAI may be developing an answer for that. We have to sort of see how this rolls out. But they are previewing specialist dots that are
[24:08] designed just for organizational workflows and jobs. And they're working with Microsoft, see I told you they were partners to integrate them into Microsoft's agent 365 controls. So that allows OpenAI to compete for employee work while accommodating the systems their employers use to manage agents. And I think that's going to be really important. So a company is going to end up being a multi- aent company. It's just going to happen, right? And that's where most companies want to be. They don't want to be locked in. Now, you're probably saying, Nate, you've talked about all these models. How do you
[24:38] actually allocate right now? So, I'm just going to give you the tidbit. I get asked this all the time. Right now, as I said, I'm using anthropic models and I'm using OpenAI models really heavily, both every single day. So, we start there. I am using Astra where I need a tireless, extremely high intelligent worker that just chews through problems and it handles a high degree of technical complexity and it's also extremely steerable. and I have a bunch of custom skills for it. And so I find that I get extremely predictable results that I can
[25:09] just do longunning work with. Soul 6.1 is somewhat similar. So with Soul 6.1, what I'm finding is I have to shade the complexity of the work down just a little bit cuz it's not Astra, but it also has that persistence that I love and it just goes after problems and it's very very capable at computer use and is able to synthesize efficiently when I'm trying to do multiple data sources. And so it's just great for I have a big piece of work. I have a lot of data. I have a lot of files. There's going to be a lot of contact tokens. I don't want to
[25:40] spend all of my weekly allowance. Then we move over to the anthropic models. Fable 5.1 is phenomenally intuitive. And I don't even know if intuitive is the right word. I gave it a problem set that was across six months of complicated transcripts, docs, like all kinds of insane stuff. hundreds and hundreds and hundreds of pages. And I said, "There are patterns here that I am not seeing because I don't have all of it in my head at once. I can't as a human, but I'm going to
[26:11] formulate a question for you about where I want to go next based on all of these conversations and documents, and you need to figure that out with me and I asked that question of Fable and Fable was able to articulate novel insights that were sourced and inferred directly from the docs and the transcripts. It was right there in front of me, but because there was so much of it, I couldn't see it. And it was expressed beautifully. Like there was an intuitiveness in the way it expressed it
[26:41] that I absolutely loved. And so when I have those kinds of problems, I go to Fable right now. And then with Opus 5.5, Opus kind of like Soul 6.1 is extremely efficient and I like both of them together because they have different personalities. So, Opus 5.5, I find I can push it to not use what's called mannered pros. And so, I can actually nudge it and it will easily be steered, which is something that has been really refreshing with this Opus model. And it will write really clean and it will
[27:12] produce a tremendous amount of work over time. It is not as fast as Astra. It's not as fast as So, ultra fast mode. I give it work that I can take my time with, but it does really, really high quality work. And by the way, if you want even more efficiency on the anthropic side, of course, you have sonnet. Sonnet is actually priced like soul. And sonnet is a very solid model. Sonnet 5.5 is efficient. It takes well- definfined work really well. It chews through it. It handles complexity pretty well. And so sonnet is another piece
[27:44] there where I find I have a lot of confidence if I give something to a sonnet model, it's going to get through it. It may not be quite as fast as an ultra fast mode, but it absolutely gets through it really well. That's just some initial thinking. I hope that helps you if you're trying to pick between the two. Like I said, I wouldn't want to pick. I love them all. All of them have jobs in my workloads. And that's really the larger point. I want you to start to learn that skill set because that is the skill set of the future is being smart about where you assign stuff to AI. AI delegation is what we're all going to be doing. So the question is if you're a
[28:14] free user, if you're a 20 buck plan user, do you upgrade to pro? Do you upgrade your plan? Is the DOT worth it? So your first personal DOT is included in all three eligible pro tiers. So you don't have to go to 200 or 500, you can go to 100. And of course, DOT's availability is somewhat geographically limited. So if you live in UK or Switzerland or the European Economic Area, it's not available yet. And you're going to have to be willing to wait for that. I there may be a regulatory issue there. I'm not sure. But there is a catch here for existing $200
[28:44] subscribers. eligible subscribers, and this is a new change, they retain their previous allowance through October 29 if you're already subscribing, and after that, they are going to move to a lower allowance at the same price. So, you're going to get less usage for your $200. And that has not been popular. Now, the answer, of course, is effectively soul 6.1 is super efficient, so you don't actually lose out on your tasks. We will see. In my experience so far in the AI revolution, people want more and more AI and they're not happy when they get less
[29:15] of it, no matter how efficient the model is. But we will see. Now, if you're signing up new today, that new subscription is already going to have the lower allowance. So, what $200 covered last month is going to be different from what it covers next month. Now, if you go fancy, if you go to the Ultra Fast, the Pro Max, whatever it is, the $500 version, Astra can generate tokens really, really fast. You can go to eight times faster in Codeex. And so that is going to really put a premium on your ability to steer correctly. Do you know how fast you can
[29:45] get a wrong answer at a pricey price with that? And so I think of this kind of as like a progression of skill levels. If you are not confident in your ability to steer AI, well, I would not be paying $500 because I don't think you're getting value out of it. I would really look at these as economic propositions and I would say am I saving time? Am I saving money? Am I making money enough from this AI that I am purchasing that it is a good investment
[30:15] for me? And I think that we're not used to doing that. We're used to subscriptions as consumers where we just say we get this cuz we want it. These are getting expensive enough now that if you choose it, you should have a plan to make the hundred bucks back. And Dots may be able to help you with that. Uh there are other features that may be able to help you. Maybe you're going to find that uh you're able to do so much with the soul 6.1 that you're able to chew through that really efficiently. But that's really my mindset, not just with open AI plans, but with any paid AI plan. You get so much great free AI now.
[30:45] You get free AI with Muse. You get free AI that is coming with uh a lot of other plans. Gemini has sort of a generous free offering that a lot of people use. ChatGpt of course is free for most of their 1.2 billion users. You need to be at a point where you are ready to say this is worth it. And if it's me and I'm upgrading, I would not upgrade just for DOTs. I think DOTS has a lot of potential to shape the future of work and shape how we think about autonomy and proactivity over time as a mental model. Dots is going to be a part of that future. There will be other
[31:16] dot-like products in the workplace that are similar. It is the model that matters to me, not this particular product. Now, if you use chat GBT a lot, as I do, I also find it useful as a particular product. And so, like I said, it immediately saved me time back on my calendar, which is valuable to me. It immediately sorted out a conflict issue with the DMV and I didn't have to touch the DMV website. And so, there's some hassle saved. It absolutely adds value. But you need to think about it more as I'm investing in my future. I'm getting ROI back on my time. I'm getting ROI
[31:47] back on how I use this product, and it's worth it to me on that basis. Now, I would be remiss if I didn't touch on Jev and Decisions API. Decisions API is basically OpenAI's answer to Jev. So I did a whole video on Jev. It's a type one model system. It classifies stuff for you. Well, decisions API from OpenAI does the same thing. It uses Luna to choose among answers you define. It can classify changes. It can classify things as pricing. It can classify things as release dates. It can do categorization like Jeff. So if you watched my Jev
[32:18] video, you'll recognize why the constrain decision is useful. Now, decision starts in limited preview, but like Jev, it's looking really cheap and really fast. And so, to me, again, I follow these ecosystem trends. Classification is going to be a mega trend in the second half of this year. And I agree with Diego, the founder of Jeff. There are so many business problems that are classification shaped. It's going to be really exciting to see this turn into a whole ecosystem that is going to actually enable businesses to do a lot of automation and really enable you and I to do automation. I have been
[32:48] using Jev in the background do a bunch of email classification and I had Codex write a little Python utility so I didn't have to use tokens to call Jev and it's been super efficient and if you're like what is a Python utility you don't have to know in the year 2026 finally I want to talk about plugin extensions incredibly exciting opportunity plug-in extensions give us a chance to participate in shaping chat GPT software and it gives us a chance to build
[33:18] interactive tooling powered by our customers APIs or our customers open AAI subscriptions and it allows everyone to essentially contribute to the chat GPT ecosystem. There is a marketplace product just about every time I turn around in AI. Marketplaces are notoriously hard to start. Certainly, OpenAI has a tremendous number of users to jumpstart this and we are seeing anecdotes already of people launching plugins and getting tremendous uptake because there's so much passive usage
[33:50] around solutions that require specialized tooling that right now people are just asking questions of AI for. So, if you're asking AI for a dashboard about voice of customer, imagine being able to just have a plugin that can pop up in the answer and say, "Hey, would you like this plugin from so- and so, do voice of customer really well? It seems like it applies to your records." And you say, "Yes, please." And that's how the marketplace starts to get going. So, we'll see where that goes. But, I think that there's a lot of potential there. And that's one of the most interesting opportunities I see right now. If you're like a side gig
[34:20] person, a builder, build a plugin, build something to solve a problem and make sure it's tuned to the kinds of questions people ask. It's going to be a huge market. And that ties into the login with chat GPT and use your chat GPT subscription that they launched for other companies as well. So if you are launching something and you don't want to use your tokens to make people use your AI service, you can have them bring their chat GPT subscription to you. And so they can log in with Chad GPT. This is available to limited partners right
[34:50] now, but they're absolutely going to scale this. It just seems like such an obvious idea. And then you the entrepreneur, you the builder, you the company, you the service don't have to figure out token fraud and don't have to figure out how to serve tokens to people and token pricing. You can just offer the service and say bring your own AI, plug it into my service, and use it with within your subscription and it will just work. So I think that's going to be a fairly significant category and we're going to have to think about building for it differently because you're essentially building around a particular product shape that someone is bringing
[35:21] in from chat GPT. That's going to be an interesting one. Ultimately, if you step back and look at the dev day story overall, it's a story of enabling small teams to get much more ambitious. Remember when I talked about teams and how team shape is going to be impacted by these releases by dots, by space, by pages. This is where small teams get more ambitious. This is where we start to see a lot of push to get aggressive small teams going that have missions that are really clear. Before a service used to require a large engineering investment, more ongoing human attention
[35:52] than a customer could afford and that may now be in reach, right? You should raise the sights of your ambition. A lot of our job in 2026 and 2027 is to find ambition to match the intelligence we have. I reflected recently that by 2036 I think we're going to have a thousand times less agency and ambition relative to intelligence than we have today. And I suspect that's conservative. And I'm talking about the ratio here, right? How much intelligence is going to be available compared to our willingness
[36:22] and ability to do something. Intelligence can grow enormously while we keep attempting the same size of project. And I see that as a failure mode, right? And taste matters here too. You have to decide whether the thing deserves to exist, whether the customer needs it, whether the result is good enough. My favorite really, really high taste experience for consumers right now is Muse because of all the thoughtful details. And we can circle back to agency in the story of product releases from OpenAI and DevDay. So, you may
[36:52] recall that OpenAI retired the standalone browser Atlas. I actually talked with the builders of Atlas on this channel a while ago and they ended up moving a lot of those browserbased agent capabilities into chat GPT and codecs. The product as a form factor may have not been something they chose to prioritize, but the ability to work through a browser was a correct primitive for AI and it just found the right home. And so I am really really bullish on the ability of teams if they have a good sense of taste and high agency to try things and then converge
[37:23] on the right form factor over time. And so with dots I would watch the jobs that people keep using it for as packaging changes and evolves. I expect dots from an agent perspective to be a durable intelligence over context category. How does D dots evolve? Well, that's a question I'm going to be curious about for next step day. Ultimately zooming out the question we have coming from dev day coming into the rest of 2026 and 2027 is what jobs can be done right and achieve outcomes. Remember how I've
[37:54] traced the story of outcomes through this whole video. That's the question. That's a question not just for OpenAI. It's a question for Microsoft. It's a question for Anthropic. It's a question for other major players in the space. And I believe it's a question with a T for trillion. Opportunity to unlock value by actually doing jobs is massive. And that is where a lot of our focus is going to be in the next year and a half. All right, you tell me what your favorite feature was, what your least favorite feature was from Dev Day, how you've been using DOTS. Sound off in the
[38:24] comments. I'd love to hear.