We Go Hands-On with OpenAI's New ChatGPT Agent LIVE

32:37So it seems like right now what's happening is it's working on the presentation slides using Node. So Node, that would be probably Node.js, I imagine. I would imagine, yeah. And it's coding a bunch of stuff. A bunch. Yeah. So it's doing this all in the terminal. I wonder if that's how it creates all the presentations and spreadsheets when it's outputting stuff. That's a good question. Maybe?

33:05I wonder what it's doing. Let me ask it. Our activity? Oh, yeah. Here we go. Oh, this is cool. You do get an activity view. Okay. Okay. So you can go back and read its thought process because that's important because sometimes, you know, if you see something that's a problem, you can go back through the thought process and sort of pinpoint where it went off track. yeah that's huge and to be honest these aren't like little baby requests we're giving this these are i mean like we're not saying hey i need you to check my email look for anything important and ditch the rest we're saying which is by the way they do suggest they do suggest not doing that um and the reason is because if you just let it go read all your emails there could be a malicious email in like you need to be really specific when you use it in email like that situation Gemini is dealing with this week where fishers have been putting in white text into phishing emails that look perfectly normal and when Gemini summarizes it, it gives you a phone number to call to reset your password.

34:11That is not a Google phone number. So, you know... That is very shady. This is the first time I heard of that. You should be careful with Gemini's email summarizer right now. They're working on a fix for it but my understanding is now Now, by the time you watch this, it may be fully rolled out and you may be good to go. I would check that out before you use it again, just to be sure. They're working on a fix. And, you know, these things are going to happen. Like, these are going to have vulnerabilities. They're going to get patched just like everything else in our technical life these days. So I guess what I'm wondering is, does this kind of negate the need for an N8N style, like agentic workflow where you do everything via Node and sort of stitch it together?

34:56Like, is this truly going to replace that, do you think? For individuals, I think, yes. At the individual contributor level, even in a business, I think it does. what I will say is that there's absolutely still a value to that style of AI agent. And it's really interesting that that came first, but that's another discussion. I think there are still tasks that are repetitive, that you want exactly the same every time, that you want maybe piped directly into your CRM or piped directly into, I don't know, your whatever. And I think those kind of tasks, in any event, solution will still have some value.

35:44But I don't know that it's going to be the level that I would have said 24 hours ago. I know. Yeah. Yeah, we wrote, we spent all this time working on this guide to teach people how to use N8N from scratch and with MCP. And then out comes Agent and just sort of blows that need for that up. I think, though, it's like N8N does have a lot of customization. It gives you a lot more control over the process. Like, instead of just letting this go off and cook, you can kind of see what it's doing on the back end with the nodes, even when you use the Agent node. Yeah, I want to highlight a couple of things out of this story.

36:24I haven't seen that one. That's awesome. Very cool. Did you notice that when it was on the last website, it made sure to check the color filter black so that it would only look for black guitars? I thought that was interesting. Yeah, that is interesting. You know, I think, you know, there's always going to be a need for large-scale, you know, data piping and processes like that on the back end of the business that stuff that makes a business run. The stuff that is right now what engineers do. They're working on that architecture, that plumbing, that keeps the internet functioning. While I think some of that will absolutely be done this way, whenever we see what I would say is the Pro Tools version of this that back-end guys will be using, because, I mean, there's absolutely an impact here on how a codex-type tool works as well, I would say, in the long run.

37:25Mm, interesting. But I think there will be a value for those kind of options still. I still think there's going to be a need for some software. I just think that what we're going to see both as writers, as software engineers, even, you know, maybe SEO people or something at some point in the future, to give another nod to our friend. is a day that looks different and has different tasks than what we're used to with a higher volume of things being done with less effort than they have been in the past. I could totally see that. I want to give this one more example before we jump off here. I think that's a good idea.

38:09It's got a little bit stuck on the slides that we asked it for originally. I don't know if we prompt it differently, if it'll kind of fix that. but I'm going to go ahead and start a new chat. And so I'm going to go to agent mode again. And we're going to run a third one. Yeah. Let's do something that's kind of easy. Do you want to do design an automated financial analysis dashboard? We could. I want to see it give me a spreadsheet. I think is what I want. Let's see. Yeah. You know, I saw Ethan Malik said that one of the things that's really impressed him about this is that it can do really good work in a spreadsheet.

38:47including formulas, which is frankly not a thing anyone's doing beautifully just yet. I agree with you. And this did really well on OpenAI's spreadsheet benchmark. That's apparently a thing. Yeah, that's supposedly one of its strengths. And it scored like, I want to say like a 42 to a 45, somewhere in there. I can't remember off the top of my head. But it was significantly higher than any of their previous models. I don't have that data to compare to other companies right now. It's certainly better than Copilot. What it was was like 45 % versus Copilot XL. Yeah, exactly. Sort of doubled the performance of the standard file.

39:32And it listed people at 70, and I feel like that's a really generous 70 for how humans I know use spreadsheets. I look at a lot of spreadsheets during the week. I mean a lot of spreadsheets every single week. and uh most people myself included are not spreadsheet wizards true um you know there are plenty of people who are um but most people are comfortable functioning in a simple grid a lot of people aren't comfortable with formulas beyond like some you know um so i think there's There's a real cool ability here and some neat stuff to discuss later on down the road about AI-friendly spreadsheets and stuff that we've had discussions about on the side in the past.

40:22So real quick, let's explain what's happening on the screen here to people who are watching. So what I did is I said design an automated financial analysis dashboard. Download the last 12 months of stock data for the top 20 S &P 500 companies. create a comprehensive spreadsheet with automated calculations for pe ratios volatility metrics correlation matrices and performance indicators and then we said here use terminal to run statistical analysis so i wonder if the reason it was using the terminal in the last example was because we asked it to um so i'm gonna actually i'm gonna look into that but it says start by giving me the spreadsheet and then go from there so then uh understood i will begin by gathering the list of top 20 companies, blah, blah, blah.

41:07And we've already seen it kind of work on that. And now it's building this in the terminal. So I wonder what happens if we do this same prompt without the terminal, if that makes it. Yeah, because I don't think we have to tell it what tools to use. My gut says this knows what tools to use, which is kind of goes back to that unified vision of a GPT-5 where you don't have to know. And some of that is already worked into 4.0. I was doing a thing last night where, I swear to you, 4.0 called deep research. I don't know how it did it. 4.0 has been thinking for a minute now. Like, I've been able to get requests through that it spends time thinking on, like it's a research model.

41:50I don't know if that's me being piped through to a different model or what, but I haven't heard much talk about that. Yeah, that's interesting. I really haven't heard anything either other than perhaps the automatic, well, I don't know how to describe it, like filtering on the back end, routing on the back end is sort of pushing it in that direction. So what's happening here now is I gave it the same exact prompt as before, and I just took the terminal stuff out. So I really want to see this spreadsheet in action. Now, one thing to keep in mind with a lot of this stuff, which I think this is a good observation, is that the problem with finding all of this data and connecting all of this data is siloed data.

42:36Like so much of this is on different websites where you have to have certain permissions and access. You can give agent access to your data in a couple of different ways. One, you can let it sign into your account for you. But the other one is through connectors. So if you have connectors, if you see down here under sources, you could actually connect different data sources for it to access. And it's user friendly. Anyone could do it. That is not a thing that's intimidating. It's not a big process where you're doing anything fancy. It's just like connecting your bank account to Venmo or something like that.

43:16Yeah. Check this out, Corey. So we took the terminal part out. it still ended up using the terminal as part of its process. Okay. So now it's saying I need to download. I'll open the Yahoo Finance URL. This is interesting. Wow. I'm going to compare this to the other one. So I'm going to scroll down here. I want to do a project. I'd like to take and give it, get an investment plan from it, put$1 ,000 on something. Not that I have that despair right now, but I would love to one day. dedicated to our guitar. I have$1 ,000. What should I do with it? And what could I do today to make that$1 ,000 become$2 ,000 by the end of the next 30 days?

44:02That would be a really interesting task. Actually, we should do that. Maybe not right now. You've got to be careful with that too, though, because they give very explicit instructions like not to YOLO, like to get it to not YOLO your money, which is very possible. If you give it access to your bank account, that's why like like half of their live announcement for this was warning people like, please be very careful with this. It has a lot of risk attached to it. Absolutely. Absolutely. I just, you know, it's a thing I've wanted to do for probably two years. Like I've thought like like we weren't there yet, but I'm willing to bet this will get closer than anything we've seen before.

44:47Definitely. I mean, like, if it doesn't lose my money, it's already done better than I do sometimes. Facts. Yeah, I can lose it just fine on my own. This is definitely, you know, has the, I won't say necessarily smarter, though I'd argue it is, but it definitely has the ability to process an amount of financial data that I am not capable of processing in my head. if it can do that without hallucinating. Very cool. And honestly, I haven't seen any signs of like hallucinations or those kind of problems in the little time we've been sitting here. I'm seeing something happen, which is kind of interesting with this financial analysis one.

45:34So both of them came to this website called Stook, S-T-O-O-Q, and they got Stook. Like they got stuck. and I don't know if that's because it's a bad gateway or what. I'm really not sure what's happening there. We'd have to dig into it and really look at it on the back end to figure it out. But you see this here? It says bad gateway. So I'm going to try that something. I'm going to take over the browser, say I understand. Yeah, because we haven't done that yet. Oh, I wonder if we could send it to a site of our choosing and say work from here. Yeah, what's a good stock analysis website? I wonder if I get it to use Perplexity.

46:16Ooh, that would be cool. Let's see. Where is it? Right here. There it is. Can you log it in? You don't need to for Perplexity. Is it already logged into yours? You don't need to for Perplexity. It's like Google. Best website, S &P 500, top 20. I'm writing it like a Google prompt. You would want to write this more. Maybe track or two. that's accessible by an AI. Oh, man, I can't spell. All right. The ability to type is going down quick, isn't it? It's a common of decline. I'm using Whisper Flow every day. Yeah, there we go. Now, do you hand it back over to that and let it use that research you've just gathered?

47:06Yeah, let's give it back control and see if they can figure out what I was thinking. Meanwhile, we'll compare me not messing with it. Looks like this one's figured out that that was a dead end. So it's trying something else.

47:25The user is taking control of the virtual browser. Okay. It's kind of mad at me because it was trying to do something, but now it's like... The page changed unexpectedly to a perplexity result. Hmm. I'll see if the file I downloaded is gone forever. Oh, it's passive aggressive. Nice. Yeah, that's hilarious. Some idiot grabbed the mouse.

47:53How sentient do you think this is? I know it's a ridiculous question, but... You know, it's one of those things where that's a question I ask a lot. I'm always reading responses, and I'm like... And it says things like, well, I'm thinking I could approach it this way, but maybe that's not the right way and i'm like it's it's doubting it's doubting itself how do i oh this is interesting uh quick quick interruption so basically i said ignore stuke it doesn't work and then it prompted me it's like are you writing code meet codex and then it basically gave me a link to the codex agent wow the only thing that's kind of annoying about all this is that so far I have not got a spreadsheet, and all I wanted was a spreadsheet.

48:38Yeah. Whoa, whoa, we did get a report back on our guitar search. Okay. Yeah, what's really important to you, right, Corey? Yeah, yeah. Whoa, slow your roll. This is what we're after. Oh, okay, hang on. Look at that 83 there from Queens Vintage. I'm not familiar with them. Should we check it out and make sure it's legit? Let's check it out and see if it's real. Oh. Pretty. Oh. It's out. It's gone. It's sold out. so what it didn't realize is it didn't realize that it was sold out oh it says it it did realize it it told us that oh look right in the center column good catch good catch human error sorry wow dang it was the it was the one though i'll tell you that wow that's amazing a new guide a new guide for for convincing the critical stakeholder it's it's a little different to offer compromises suggest selling another piece of gear explain personal benefits playing a high quality instrument can inspire creativity reduce stress and provide a constructive hobby uh yeah not bad all right cool what i want to check though is i want to check our our spreadsheet um or our presentation i don't feel like let's say i I realize, I say the thing about my wife as a joke.

49:56It's a joke with both of us. And like, I would just talk to her and she would say, that's pretty cool. Go for it. I absolutely would not use this to go manipulate my wife's behavior. I just wanted to say that. Good, good, good to know. Okay, so we got a presentation. So the first prompt that we asked agents to do, it was able to give us a presentation, SEO tools, competitive intelligence. it's got an agenda pricing comparison feature analysis value versus cost swat blah blah blah okay pricing comparison so it's got hrefs okay rush moz pro screaming serp stat i wonder if these prices are 100 accurate i have to go back and check this is absolutely why they're working on a solution like google drive or microsoft office i mean yeah it makes a ton of sense It's like working on, not working on.

50:50We clearly have, or this isn't here, I would say. Yeah, I think something on the back end. What's an interesting challenge for OpenAI is, okay, if they're going to be a cloud provider, they're going to need some of their data centers to be for storage, right? Well, you know, and with the amount, I don't know though, with the amount of data they're going to use and the number of data centers they're going to probably have at some point, I'm willing to bet that the amount they need for that end of things would be comparatively small.

51:28Then, maybe not. Maybe not. Maybe I'm oversimplifying it a little. And honestly, it's not a subject I'm qualified to speak on. I will say that while this is not like a gorgeous spreadsheet, a gorgeous presentation, it's a heck of a start. Like the data's there. The information's there. I mean, I can go through and make it pretty. Yeah, I agree with you. And what's it? I think we got a phone a friend call from our SEO expert. It sounds like the SEMrush callouts are pretty accurate according to our phone a friend. So that's good to know. Wow. That's good to know. So what happens if I ask it, I'll say, this is great, but make this presentation beautiful.

52:17Ooh, there we go. Let's see what happens here. I like how we spend so much time talking about AI, but we ignore the fact that we're not using capital letters, punctuation, or spelling correctly. And it's still figuring it out from there. Doesn't need it. That's why this stuff is really amazing. I judge it when I type into it like a monkey. It's, it's, this is, this is pretty awesome. So it's reviewing the slides. I wonder, I wonder how it'll edit it. Let's see if we got a spreadsheet. All right. So far. Okay. Okay. Stuke is working. So they've gotten this website that they are apparently obsessed with to work.

53:01This website has been so excited about chasing. Yeah. Now here, something's happening. This guy's on GitHub. I don't know what he's doing. Uh-oh. That could go bad. Chrome Overlay. What website is it on? I'm just going to ask it. What are you doing? I just want to take a minute here to say something. There have been a few times since the release of ChatGPT. The first time I used it, which was the first few days in November 2022. Is that right? 2022. When ChadGPT4 dropped. When 01 dropped. When I first discovered Data Analyzer. Mm-hmm. Yeah. And using Deep Research and 03 Pro. This is one of those whoa moments for me.

54:03I'm feeling it. Yeah. I'm feeling it. I'm feeling it. I'm a little giddy. You know, like like it's a moment where you immediately feel like the paradigm has changed again. And like this is what other people have to step up for. And it makes me wonder if Sam Altman wasn't sitting back in his easy chair watching the world freak out about Grok 4 Heavy a few weeks ago. And just kind of, you know, chuckling with a nice glass of scotch. I don't. So I love that visual. I don't think so. I think they're always perpetually terrified, which is why they're constantly streaming ahead and working on something new and trying to push the envelope.

54:46My favorite meme is the one that's the cycle. And it's like, OpenAI releases the greatest model in the history of the world. Gemini releases the greatest model in the history of the world. Anthropic releases the greatest model. And there's always this little dot that says, you are here in the cycle. And that's so just real. Yeah. It's so absolutely how it's going. And, yeah, this is cool. And, you know, and part of the reason I'm so impressed is even, not even just what we're watching it do right now, but like historically speaking in the AI space, something comes out and it's okay. And over the course of the next 30 days, 90 days, 180 days, that six month mark, like by January, every one of us will be using this every day.

55:42Here's what I'll say about what I'm observing so far. I think if you want to do things exactly the way that you want to do them and you like to do them, then you might want a more refined tool than this. You might want to use N8N to design your workflow, to use exactly the application that you want at exactly this step in the process that you want. Or at least you can give the agent node a little bit more freedom by using the agent node, but then you can still kind of give it access to the MCPs and the connectors that you want it to use. If you don't care how the job gets done and you just want the job done, this tool is for you.

56:25The answer may be both. The other thing I would say is how far out of the realm of possibility is it that there will be an N8N connector in here one day soon where you can just say, go build me an agent that'll do this and handle it for me. And it goes and handles building your N8N agent. Like that is not a difficult task for an AI to do for you. It could absolutely go spin it up. So here's where I think the perils of that are is I think it's totally possible from a capability standpoint, but it would have to then go and get the API keys and the logins for N8N. I guess and I'm assuming you would have your credentials shared in there somehow like already saved within your N8N account or for that matter it's getting them from whatever it has access to in other connectors slash MCP connections.

57:17Yeah you're totally right. I think there's definitely a way to do it. I'm sure somebody is working on that right now. I mean just imagine. By early next year. My money's on Q1. Oh by early next week, Corey. Somebody on YouTube is going to figure out how to do this. By the time this podcast comes out, you'll already be doing this. Yeah, yeah. I'm sure I will personally, when this podcast goes live, I will personally include a tutorial video for how to hook up your N8N with an agent because I'm sure somebody will have something like that. That means when the camera goes off today, Grant will be working on doing exactly that.

57:55And before we record the next episode it will be a thing we are able to do and i i'm almost positive that it will be possible it's just the question is so like i was listening to an interview uh with the perplexity ceo ervin srinivas uh on matt berman's podcast and he was talking about how it's the reason why he's developing a browser is because he wants to do as much of this stuff that's happening in the cloud here locally on the user's computer. Because when you have it log in to all this stuff for you, even though this is a cloud environment that's staged and it's not on your computer, so it's not downloading God knows what onto your computer here, you still are giving all of your access to a remote computer somewhere in the cloud.

58:47Yes. The now OpenAI or the agents or anyone who gets access on the back end has access to. So there is a security risk in doing that. Huge one. Of course, you're already giving it to likely either Google or Microsoft right now, too, who have the same tools and all of your data, you know, and don't have a data policy that they're fighting the U.S. government to be able to delete your stuff, too. So, you know, I mean, there's some interesting elements there and a lot of things to consider. Grant, what do you think this means for, you know, we've talked a lot about industries along the way and how these things impact them.

59:27Like we've talked about riders, we've talked about all kinds of fields. But what about, I feel like this one is a moment for engineering almost. Because there's a lot of things you would go to an engineer for that this will do for me. I think that's been basically the thrust of AI development over the past two years, if not the whole time, is that the first thing, the first job that it'll automate is the engineer. Now, keep in mind, I think there's always going to be a role for engineers to come in and fix whatever the AI comes up with and create novel tools. I'm sure we'll get to a point where AI is able to create novel tools.

1:00:13But, you know, the jury's still out at this point if it's able to do that or not right now. But you're definitely looking at a need for smaller staffs and less jobs. Yeah, I mean, this is already showing up in the software graduates having trouble finding a job right now. That's definitely happening. Teams are getting smaller, more streamlined, and they're going to be at some point here. You know, it's still a thing. all companies are still kind of figuring out, frankly. But, you know, we're absolutely looking at a situation where a team will be doing a lot more work a year from now than what they are today.

1:00:51And it's going to be lighter. And the team may be lighter. I mean, you know, there's definitely streamlining in the future of like, you know, what do we really need to do this at the level we want to do this? And honestly, like it or not, they're fair business questions to ask. You know, If you're running a business, you have to be asking those questions. It's your job. I think from the perspective of the worker or the engineer, we'll just limit it to computer engineering because I think they're the only ones right now who are at risk 100%. I would say that I would just be looking at creating either your own company or your own small team, joining a small team to compete.

1:01:29Because what this is showing us here is unless you're like the big tech who has all the cloud servers and has like the AI models, like whether Google or AWS or Microsoft, I mean, we'll see how that shakes out with their deal with OpenAI or OpenAI. Like this gives you incredible tools to compete with everybody else who is not those five people. Just like it gives it to a business, it's also giving you that same ability. That's a good call out, Grant. To compete with giant businesses. So my theory of where all this goes is a lot more smaller companies, like call it millions of smaller companies formed with these stores.

1:02:08Altman's one-man billion-dollar company that he's joked about a couple of times that they've been waiting for the day when there's a one-person billion-dollar company. No, it's not a joke. And I mean, like, it probably exists already and just isn't it a billion dollars. Now, I think the giant risk that this has for small companies is that you become entirely reliant on those five companies I just mentioned. Yes, you do. Whoever creates this tool, if you're using this every day as part of your business, you're beholden to them. The same way that anyone who's on the cloud right now, instead of having their own data centers, their own computers, is beholden to whatever price the big cloud companies want to charge.

1:02:48So that is the risk that comes with this. I'm sure over time the prices will go up. The frog will boil. And, you know, the thing about being beholden is like, you know, Google did a thing this year that's exactly that. You know, they've released Gemini. They want more people using it. And they force added every Google workspace business account in the country, you know, to a tune of many, many dollars a head on top of what they already paid. And, you know, that's the nature of always being beholden to whoever has your data. In my experience, this kind of reminds me, and I felt this way with deep research and thinking models in general.

1:03:28This reminds me of what I observed living with my dad. So my dad was an engineer and he worked on chips and one of the chip design. And one of the things that he would do is he would often, you know, I don't exactly understand the mechanics of this. but he would be working on something, you know, head down to his desk for hours and hours and hours. And then he would have to let it go off and render or compile or whatever it is. And he'd walk away from his computer for like a couple hours and, you know, just let it cook. I feel like this is going to turn work into that to a certain extent, where you spend a lot of time upfront figuring out exactly what you need it to do.

1:04:05And then you like create like 10 of these instances. Like I've got three here running right now. Where your work is, yeah. Yeah. And this is, 24 hours after release. We've got three agents right now. Yeah. Off doing work for us while we sit here and just talk smack and laugh about it and talk about the future. Since we've come back in and we've done all this, is this truly agentic AI in your view today? And if not, what will it take to get there? I mean, I was never the skeptical one about it, about agents being possible. Which is funny because you're always skeptical, one of the two of us. That's so funny.

1:04:45Yeah. Well, don't get me wrong. I'm skeptical about this. I'm skeptical about the actual practical use cases. Like I was watching an interview with Dan Shipper, who's the CEO of Every, checked him out. And he was on the TPBN pod. And he was basically saying like, so there's two problems with agent. One is like there's certain things that you want to be more manually involved in working on. And this is sort of more of like a set it and forget it task. But then the other one is that we're so used to watching like it's such a beautiful UI and the experience of watching it is so unique that you almost get glued to it and you're watching it and you're wasting time.

1:05:24And I'm not going and doing something else. Yeah. So you have to kind of retrain your brain to like do what I was just saying with my dad, like set it and forgot it. Like go tinker, like really get what you want it to do. Set that and let it go off and cook. That's fair. And so I think that that's really interesting. But I am skeptical of this being, you know, like, will it be able to do everything at the level of quality that I expect? Like, maybe this is enough that you could take this presentation here and run with it or, you know, maybe take it into PowerPoint or whatever and add your own design elements to it.

1:05:59Maybe there's another tool that can take this as it is and beautify it. That might be possible, too. That's true, too. So as long as the data here is correct, this is pretty much good to go. But there's maybe other tasks where you do need to be more involved in order to give it all the permissions you need. You have to be more hands on. I think it's fair. So I'm skeptical of if it will be as useful as it appears here. And I think there might be both opportunities there, too, like for tasks where you do want that level of specificity and how it's being done, as well as tasks that you really don't care.

1:06:39Like probably everyone has tasks that fit into both buckets, if I had to guess. I'd agree with that. For my answer, it's agentic AI. It made the decisions on its own. We didn't program anything. We didn't write any code. We wrote sentences in a box and told it what we wanted. And it figured out, okay, I need to go browse the internet. I need to search for these. I need to figure out the financing. I need to consider what the structure of this is. And it called the tools it needed to do the jobs, some that we requested, some it did on its own. It spit out a viable spreadsheet. Not a spreadsheet, but it's still working on that part.

1:07:21But it spit out a viable spreadsheet. Yeah, a spreadsheet is TBD. Like, I have absolutely been in presentations with worse appearing spreadsheets than that. Like, my only real criticism is it's a little boring. Like, I would class it up a little bit. It's probably the biggest thing I would change. What aspect is boring? Just the layout. Layout. And, you know, I would say the charts are a little dull, just, you know, plain white charts. And the truth is, I don't like a flashy presentation, but I feel like it needs a spot of color somewhere or something just to give it a little juj. And it didn't show the whole image on the front.

1:07:59There's a little criticism there, but I mean, I'm nitpicking because, again, we're talking about 24 hours in. Less than 24, well, right at 24 hours in. and it's not going to get worse as is the saying that I say this all the time and I absolutely stole that comment but I think it's going to continue to get better and continue to get access to things that make it increasingly more agentic like scheduling and dispatching. I think that's big. I think when I'm able to say listen I want you to do this every morning. I want to come in at six in the morning and have a report of my most important emails.

1:08:38I want I want the top five AI headlines for the day. And I want you to go through my calendar and gather any documents or notes I need to be prepared for every meeting I'm walking into today. And I would like those at 6 a.m. when I sit down at my desk. Like, you can't beat that. You know, and I want it to know when there's a tool that, you know what, this is a better task for an N8M style AI agent. I want it to know that and I want it to then just go do the thing and come back with the output. Yeah, I mean, if it already had your access to your NAN account with all of your AI credentials pre-saved, doesn't have to copy any of that stuff, which, by the way, they do say that you should take over when you're entering personal information so that you don't give that stuff to the AI.

1:09:29Yeah. Yeah. Just a quick side note. But yeah, like, yeah, it already has all that stuff. There's no reason it couldn't just go set up a workflow for you. Yeah. I mean, like, just go do the thing. That's what I want. I want it to be an interface that just does the thing that I need done. I want it to, you know, to move closer and closer to being that true assistant. And I think I think if you'd showed me this four years ago with no context and no knowledge of tools like Chad GPT, I would have said this was, you know, science fiction magic. or something, you know, I mean, if you go back a few years and I'm willing to bet that what we're looking at in 90 days, in six months, in a year from now, like, like we're 13 months out from when 4.0 first launched.

1:10:16There were no research models. Uh, you know, O.1 didn't officially launch until January of this year. Like there was, I think a beta version in December as part of their, uh, you know, ship, what was it? What was it? Shipmas. Merry shipmas.

1:10:35And by the way, hats off. You know, a year ago, people were making jokes about OpenAI not shipping anything. And now they ship more than I can keep up with. As they said. So, Corey, we have an alarm interruption happen because we got the spreadsheet. The spreadsheet is here. We got the spreadsheet. Let's look at the spreadsheet before we close out. Oh. Okay. So this is a spreadsheet. It's got all of the top 20 on the S &P, starting with NVIDIA, Microsoft, Apple, Amazon, Meta, all the way to J &J. It has the PE ratios of all of these. It has the annual returns. You'd probably want to download this, put it in Excel, and format it in terms of percentages, but it's got the raw data there.

1:11:21We would have to go through and double-check all of these, But I mean, so far it's been pretty accurate, which is kind of interesting. So I wonder how many of these will be right or wrong. I wouldn't want to do this with like 500 cells or 5 ,000 cells because I think the error rate will definitely increase with the scale. But we're going to try one of these days soon if I had to guess. i have a spreadsheet with 166 000 166 000 rows and like 100 columns that i would love to just see what it can do with removes i think from it but the next step here was what i'm seeing is like okay this is just a preview right yeah imagine if if chat to basically builds its own spreadsheet editor uh and powerpoint editor into this thing which that might be what they're working on next to make it really easy.

1:12:17This is the beginning of America's super app era. So in China, they've had super apps for years. They've had WeChat and all these other apps that basically where everything you want to access is in it. It's essentially a browser with a bunch of little micro applications inside of it. This is basically what ChatTBT is going towards in this phase. If you're able to edit your spreadsheets in here, connect it to all your external data and edit all of that stuff in this app, we're looking at the first U.S. super app. I think you're exactly right. And, you know, there's maybe, you know, while that's kind of problematic at the same time, you should see how many screens of apps I have on my iPhone.

1:13:01Maybe that's a refreshing bit of minimalism in my life. Most people, most apps, like most people only use like 10 % of their apps like every day, right? Like there's really like only about 10 % that you're, or maybe 20 % that you're like, these are my main apps I go to all the time. Besides that, any other functionality, if you can just spin it up with code and then have it be working, like the next step is like the code works and runs inside of the ChatGPT application, right? Yeah. Then you're off to the races because you're like, any functionality I need, unless it's something I didn't know I wanted that somebody else created, I can just run it inside the app.

1:13:42There's an awful big mobile app industry that probably has its knees shaking a little bit. I mean, we're not there yet, people, but it's coming. But, you know, we're also not blind. We know the direction it's headed and the speed with which it's getting there. Like every time we think, oh, look, it's slowed down. Something happens that's like, oh, no, no, not yet. And we'll see what happens. You know, right now, there's a lot of companies that are cooking with gas. And they are blowing out some great stuff. You know, I mean, we've had Gemini 2.5 Pro and the various other tools they've dropped.

1:14:22We've had Claude's new tools, which are also sick. uh x came and surprised us with you know grok 4 which is problematic but impressive in a lot of ways nonetheless uh and now we've got this and there's still a big announcement we keep hearing coming in july about maybe gpt5 so you know if you take this and you put you know a new motor in its engine bay uh oh quick quick side note this this does have a new motor in it so swix from latent space and small ai the ai newsletter really great um creator love his content i think they also have the ai engineer summit he pointed out that when you compare the benchmarks uh that open ai released with this uh there is actually a new model under the hood because they compared it separately from O3 when they benchmarked it.

1:15:16So his theory was that this is basically like O4 Full that's under the hood here because they didn't call it GPT-5. So that means there's still another model coming. That's a fair theory, I would say. Yeah, we haven't seen O4 Full yet. And I'll be honest, man, O3 is... Like, there's not a lot left that I need a model to do for me. And that might be a conversation we discuss one day is that the list of stuff I need from it is shrinking till we get robots. I got dishes. I got laundry. I got I got I got stuff I want to pawn off real bad, man. Yeah, that's a topic for another day. Well, guys, thanks so much for watching.

1:15:55We really hope you enjoyed this. And I'm so glad we came back and recorded this because I think this has been a lot of fun, really eye opening. And I hope, you know, just kind of seeing, you know, our first impressions as we tinker with this, too. I hope that was fun and a good approach. If it is, we'll do that with more tools as they come. So, you know, please, if you don't join 575-ish thousand other people who read the Neuron every morning, we'd love to have you, the Neuron.ai, sign up today. We've got a lot of other content, expanded articles based on like some of what we can't fit in the newsletter because every day there's so much more than we can possibly cram in an inbox.

1:16:34Facts. And Grant does a great job with that stuff. So check that out. follow us on Spotify, YouTube, Apple, wherever you get your podcasts. And most of all, though, thanks for watching. We love doing this and we love to see that people are watching and enjoying and hopefully, you know, getting a kick out of learning with us because, you know, we're we're doing our best to, you know, I mean, things change every day. So we're we're constantly in a cycle of learning. So just like you are, we hope we can share what we find along the way. But for now, farewell humans. We'll see you next time.