The Scott Adams School Afterparty - 8/22/26

the executions, Hobbes was facing apparently a bribery case with the AG declined to prosecute, so no justice for Democrat politicians. But apparently Newsom has come out and said he likes Trump accounts. So I think that is maybe a bipartisan win and a great thing that we should probably be touting, because I think that is a really fundamental big change. I think it's going to really benefit future generations. If you do the math, you can see that it can add up to a lot of money for a lot of people. And I think it might be the solution to failing social security, because I think that is on the way out. I think it's a failing system and I don't think it's going to get any better. But if people have these Trump accounts and they start saving and investing from the time they're born, it makes a huge difference. I mean, I learned this in college really took my first finance class and they did the whole compound interest experiment showing like what someone would have at the end of their retirement if they saved and invested in the first 10 years, they were working and then stopped when they were 30 or something or 31. And then they had another case that they compared it to saying, if you don't invest for your first 10 years, but you start after 10 years and then you invest that same amount, meaning the same amount per month for the rest of your career, like 30 years, and then show what that amount is. And the person who did the first 10 years comes out with more money. And that's the power of compound interest. But if you back that up to when someone's born, you know, the generally, I think it's the rule of 72, says that with normal stock market returns, you typically will double it every seven or eight years. And so by the time someone's 21, it's already tripled what they, you know, at least the ones that were put in when they were born. And so you know, you add a few cycles to that and it becomes a huge amount of money, like a relatively small amount starting out, a few thousand dollars or maybe ten thousand dollars can be millions by the time they retire. So I think it's a big deal and I don't think people are talking about it enough. But you know, there's more drama on Iran, lots of ups and downs on the economy we can go into. Thanks up to you. You want to talk about I'll start bringing people up to the stage. If you want to support what I do or participate in our Sunday subscriber spaces, you can find that information on my profile. And we'll start the conversation. Yes, it is going to be good. All right. Well, I think Sean was one of the first ones to request. So Sean, how are you doing? I know. Yeah. Thank you very much. Yeah. Good in yourself. I'm doing great. I'm a, I've been busy with work and with my AI, the Genetic AI system. It's human, man. It's put out 19 pull requests through which you might understand as a new feature in the last 24 hours. And it doesn't really even ask me any questions anymore because I've already worked out what I wanted to build. So it's just cranking. And it's kind of fun to watch it go almost almost like a kids now. It's like you said, set up the system and then you get some nice feedback from it really. I mean, I spent a lot of time with it working out like what would, what would a good system look like for this? Scott, I didn't always talk about systems thinking. And I really thought about this as a system. And I'm like, OK, what, how can I get the best result? Like have it, you know, I want it to build software for me is one thing at least. But I also wanted to do other things like research a topic or, you know, do a task for me or whatever. And that's one of the reasons I started with building a second brain sort of application which I think a lot of people are doing right now. But, you know, I'm like, OK, that can kind of capture all the things that happen like all the emails, all the teams chats, all the calendar stuff I have, all the to do items I have and everything that I need to do on a daily basis. I want an assistant. I want someone who can keep track of everything for me and even do a lot of the work for me and, you know, just make me more productive. And so I set out with that as my first project, which then would enable me to do more projects right? And so, you know, it's sort of a meta thing. But I think part of that then backed up to saying, OK, how do I get this agent to AI development system to work properly? And it's not just like some people do with vibe coding where you just like one shot it like, hey, build me a second brain. I mean, I could have done that. But I think it would have been a much worse result. I wanted something pretty specific because I wanted to work within my company's policies. I don't want it to leak client information. I don't want it to act on its own. I don't want it to record facts and have knowledge that isn't correct. I don't want to put garbage into it and have it just be garbage. And so I thought that through pretty thoroughly. But I also had AI help me think that through. I had it look at all the other people that are building these things and what features they have and all the other research on the topic and like academic articles and examples of open source versions of tools that people have built and pull the best from all of them and synthesize it into something coherent and then spec it out and go through literally dozens of reviews from different angles to work out, you know, what could go wrong? What should we really think through? What should we really do? And I had four different models from four different families of providers looking at everything along the way at every step. So I got lots of opinions and then I'd have one agent synthesize all the opinions and picked out which things make sense and fit it all together. And I mean, it's pretty impressive, I think. And then I have a process to go with that where I say, okay, you know, I'll start maybe by researching a topic and come up with some ideas. And then I'll say, okay, now how do we translate this into a design? And then I put it through that dozens of review process for the whole system once I have a design proposal and then each step of the way as it's building, I say, okay, I want you to build an implementation plan for this step, like how are you going to code it? And then send that to my council of agents and have them review it and then get the feedback synthesize it and then go build it. And then once you've built it, set it back to the council for code review to find any problems with it and then go fix them until there aren't any more problems. And only then do you move on to the next thing. And so that's why it does take roughly an hour for each step because it's going through that process every time. So it's a very thorough process. I don't know how many people are doing it the way I am, but I think it's working pretty both so far. I mean, I'm using the system and it's working. So it's not fully built. Yeah. A lot of the features are what they, what my system decided to call dark, meaning it's there, it's built, but it's not turned on. Apparently I have to go test a bunch of stuff and decide when to flip the switch on all these things. But it's pretty cool. I'm enjoying it. Yeah. I mean, it's, it's, well, feature toggling really the, you know, turns to anything's on when you need it. But I think that spitballing with AI is probably one of the funnest things you can do right now. And you can kind of bounce ideas off it and I have to hash something out and figure out, you know, where, you know, where you want to go with it. But the actually having it build all of that stuff years or so, you know, it's quite satisfying when it all comes together and all works. But having that kind of contentious system that can evaluate and, you know, sort of beat down the idea effectively and sort of get it down to what it actually needs to be. You know, that is probably the most important part of any design process. So the fact you've got, like you say, your council of agents working on it is really cool. You know, there's, I mean, there's, it reminds me of the new Grockbot that, you know, you don't spin touting this week and that is doing a very similar job to some of the agents that you've got there. So it is, I think, but I think the difference is that runs in the cloud. It's obviously a paid service and it's using the cloud models like Grock and related things. I think Grock in cursor, which is what I'm using, which is now owned by Elon Musk or his company, that's similar. It has similar capabilities. I can run cloud agents. I can run local agents. But the local agent part isn't really something that's very compatible with cursor and probably not with Grockbot, just because obviously what they're doing is they're trying to drive demand for their models, right? They want to pay for their tokens. And so I do use some of that and I have been using a lot of that, frankly, because I've, you know, I do want like the frontier model capability thinking for what I'm building right now, at least for this part. Like once I get to, I guess what I would call less important things, I might just say I'm shifted all local and just use my local models for that. But for now, I'm, I am using that for a lot of those council agents, but I'm also using cursor things, which is Grock and something called composer, which I think is a distilled model that came from cloud. But it's basically, you know, an anthropic sort of style model. And I'm letting it do that. And I'm also letting it use Gemini and some other, you know, Chetsupt, the latest solid agent or model from them just because I like all the different perspectives. And again, I consider this the critical system because this is like I want this to be the best software possible because it's going to help me do everything else. And once I have that, then I could probably say, okay, if it's just some trivial application, or even a feature of this system later, I could probably just say, yeah, I can have deep seek where I can have, you know, Quinn or one of the cheaper models do it because once the structure's there and everything else is, you know, it's not going to be such a big deal. But it's a lot of fun. I'm enjoying it. Yeah. It sounds like searching. I mean, if you thought about sort of putting a portal together, so you actually have like your own sort of site that you can go to, you know, maybe as a list of all the stories and things like that. Or with stories, you know, just a minute. So I'll put a what you're doing basically. You mean, you mean of the system or I don't know exactly. Do you mean like telling the story of how I built it or what? Well, no, no, the, so the system that you're talking about, this is not your curation system, I guess, the curation system. No, it's not. Okay, this is your, this is your app, the office office system that you're, yeah, it's basically what I'm calling a second brain, which is kind of like a knowledge base of everything that I mean, it's meant to be like a second me, you know, like it knows what I know. And it can advise me on things and it can assist me on things. But so when you say like having a portal page, and I was, that was for, you know, assuming the news curation system, but this is your internal sort of company system. So yeah, you're already got portals for actually interacting with the system anyway. That would be a valuable for select clientele. Yeah, well, so like the way I designed the second brain is I wanted it to be local and private. I didn't want it to be public or accessible publicly even because I want to be able to put anything in there in my own thoughts, my own failures, you know, journaling like whatever. And I wanted to be able to protect that information. And so, and plus because of the like confidentiality rules at work, I have to protect client work, I have to protect client information, I have to protect internal company information. And so I built in all kinds of safeguards to say like it's kind of partitioning the information by client and by project and it's making sure that nothing leaks out of it. And it is kind of a balancing act to saying like, how do I do that without making the system stupid? Like because in my own head, I can cross index all these different things, right? So I want my system to be able to do that too. And to be able to use that if I tell it, go build a presentation. Like, you know, I wanted to be able to use all that knowledge it's gained. But I obviously also need it not to say, okay, I'm going to take some client detail from client one and put it in a deck from client two. And, you know, so it has to have some kind of safeguards for that. And but that I also wanted it to be just literally just running on my laptop, like not reaching out to any cloud models except the ones that my company has specifically approved, which in my case right now is just cloud code. So it can use cloud code models if it needs frontier capabilities. But otherwise, it's going to just work on my laptop and ingest information and spit it back out to me in whatever form I need it. But it is keeping track of all my to-dos. It's keeping track of like all my, I mean, eventually, I don't think I have this quite big of a big deal yet, but it's going to ingest a lot of my email and my calendar and my team's chats and all sorts of channels of everything that's coming at me so that it can also automatically generate those to-dos. But it can also, again, help me complete those to-dos if I need to draft a document or draft an email or whatever, it'll help me do that. And then, you know, check it off as it goes. So it's kind of a workflow system for me. Because I'm already designing in extra features that are really cool, which is, for example, it'll prep me automatically for every meeting I have that day. It'll tell me like, here's all the history, here's what you need to know, here's what you should say in the meeting, here's what you should ask in the meeting, here's the objective of the meeting, like here's what you're trying to get to and here's how you should do that. And then during the meeting, I'm going to try this. I don't know how well it's going to work yet, but the capstone, I think, at this point, is I want to have some kind of, like, heads up display sort of thing during the meeting. That is advising me in real time, as the meeting is happening based on what people are saying, what I should do and what I should say. And if someone asks a question that means we got to pull up a document, it should go do that and give me the answer.