0:00Jordan Metzner says the biggest change is multi-agent management, with agents managing other agents. Sam Nadler previews a personal health tool, the new Codex desktop Mac app and news moving public markets.
1:38Jordan Metzner says he gave Claude Code his 15 MB 23andMe file plus blood tests and medications. It reported markers like diabetes risk and lactose intolerance and wrote a report for his sister about a cholesterol drug that worked for him.
4:18Jordan Metzner says he runs it locally with SQLite, not online, and adds new blood work after each doctor visit. He says Epic's MyChart makes it easy to export blood work history and medications.
6:47Sam Nadler says Claude Code is more instantaneous, while GPT-5.2 Codex on extra high reasoning takes five to ten minutes per run but found many bugs and missing gaps when he polished an existing product.
8:01Sam Nadler shows settings for goal, workout days, body focus, experience and equipment. When he logs a set as a struggle, it suggests stepping down, dropping the weight from 20 to 7.5, or ending the exercise.
10:44Sam Nadler says to use a lighter reasoning model for quick iterations and deeper reasoning when polishing toward production. Jordan Metzner says the trainer shows how much easier it now is to build personal software for yourself.
12:19Jordan Metzner says Anthropic released Opus 4.6 and agent features in Claude Code, OpenAI launched Frontier to manage AI coworkers for enterprises, and Perplexity launched a model council mode where a chair LLM synthesizes answers.
13:49Jordan Metzner says people used one chatbot at a time, but now agents manage other agents, making users managers of managers. Sam Nadler says managing four agents already makes him feel four times more productive.
17:08Jordan Metzner says it is too early to tell whether incumbents adopt AI or startups replace them. Sam Nadler expects some giants to struggle within 24 months as companies build software they used to buy.
23andMe DNA in Claude Code: A Private Health Dashboard: the key questions from this episode, answered with quotes and timestamps.
Jordan Metzner [0:00]: Historically, everybody's been using one agent at a time. But I think the biggest change we're seeing now is multi agent management with other agents managing those agents.
[0:20] Theme song: Built this week, breaking it down. Built this week, we show you how. A fresh idea, a clever tweak, you locked in true. Built this week.
Sam Nadler [0:21]: Hey, everyone, and welcome to Built this week, the podcast where we share what we're building, how we're building it, and what it means for the world of AI and startups. I'm Sam Nadler, cofounder here at Ryz Labs, and I'm joined each and every week by my friend, cohost, and business partner, Jordan Metzner. What's up, Jordan? How are doing today? Yo, Sam. How's it going? Another crazy big week in AI world. Lots of fun products to talk about. Some of these products causing havoc in the public stock markets. But for our agenda this week and be actually, before I jump into the agenda, if you haven't already, please like and subscribe. We have new episodes out every week. I believe this is episode 30, so congratulations there, Jordan. But, yeah, please hit the like and subscribe. And for the agenda this week, we're gonna cover a really cool personal health tool you built. We'll also jump into the new Codex desktop Mac app, and then lastly, cover some of the latest news that's making a lot of ripples throughout the public markets.
Sam Nadler [1:27]: Anything to cover before we jump in?
Jordan Metzner [1:28]: Oh my gosh. So much going on. I mean, news is dropping as we're recording this on Thursday. So, yeah, there's just so much going on.
Sam Nadler [1:35]: Why don't we start with your health tool that you built?
Jordan Metzner [1:38]: Yeah. I had done 23andMe a few years ago, and I was using the app, etcetera, and then if you remember, like about a year ago, they were like, gonna go bankrupt, and there's all these things about selling your data. So I went on 23andMe, and I downloaded my data. It's like a one TXT file that's like 15 megabytes, and it's like pretty much unreadable. And so I had that saved in my Google Drive for quite a while, and I was thinking about new projects to do. And I figured, why not give Claude Code access to my DNA, and, maybe I can even add some additional health information. So, hopefully, we can block out anything that's, like, too personal. But, essentially, what I had it do is kind of give me insights regarding kind of this type of stuff that you would honestly see in 23andMe. So, what I have for APOE, my risk for diabetes, if I have any lactose intolerance, which in fact, it says here I have high lactose intolerance, but I actually don't have noticeably a lactose intolerance, at least for myself. Kind of stress responses, aging markers, all this kind of stuff.
Jordan Metzner [2:33]: And then I actually, like, it just keeps going really deep. So you can kind of see all different stuff here about all my different chromosomes. I've included a bunch of health information. So this is personal health information that I've added so I can kind of tie in my personal health directly with my DNA. And so what I've been doing here is been giving information into the system and telling it about what drugs I'm on, and what are my results from my blood tests, etcetera, so I can kind of try to bring it all through together to have this comprehensive system. And I think like one thing I learned is I'm on this cholesterol medication, and it had a big impact on dropping my LDL cholesterol. And so one of the things that Claude recommended was that I talked to my siblings, my blood siblings, to see if that they were on those medicines or if they should try them or not. So I actually had Claude Code build an entire report for my sister based on my DNA and my health information so she could share it with her doctor, and so that they could analyze her health against what might be hereditary or kind of inherited behaviors. So it's an interesting way to think. I think, like, in general, I don't know if siblings share their health information that easily, but it makes me think that this might be like a new mechanism or new way for people to kind of get deeper analysis on their health, like, and how the DNA affects their health and their family health.
Sam Nadler [3:50]: Wow. That's amazing. And how did your sister respond?
Jordan Metzner [3:52]: Well, I asked her if she's on this medication, and she wasn't, And so she had never heard of it, and so I said, like, you should talk to your doctor. And then she's like, well, what should I tell my doctor? And so then I said, like, okay, well, I'll create a report for you. So I don't know. I should follow-up with her. I don't know if she's seen her doctor yet. But, yeah, just like this medication, I was like a super reactor to it, and it was, like, worked very effectively for me. So it most likely means that it'll be really effective for her to take it as well.
Sam Nadler [4:17]: Wow. And in terms of, like, continuing feature build out, are you gonna have I mean, maybe you have already. You added some family contacts, but, as you get health tests throughout the year or throughout, future years, whether it's blood tests or additional tests, do you are you planning on incorporating that into this holistic, health insight dashboard?
Jordan Metzner [4:40]: Yeah. So I'm, I'm trying to find insights and learnings that are, like, not easily available. I'm gonna visit my doctor in a few weeks, so I'm gonna show him kind of everything I've built. But every time I go to my doctor and I get some blood work back, my plan is to add it into the system so that I have this kind of holistic, a system that has, like, all of my information in it. This is kind of some physical traits that came out of my DNA. But, yeah, I think over time, as I get, more and more feedback from my doctor and from just overall my health behavior in general, then I'll add it in here. And hopefully, as the models improve as well, maybe I'll get some more unique insights that maybe weren't easily available previously.
Sam Nadler [5:19]: Cool. And is this tied to a back end database, or is it SQLite? Or
Jordan Metzner [5:24]: Actually, there's no database. Yeah. I think I use SQLite. I'm really I think I use SQLite, and it's all just run locally on my computer. I don't really care to put it on the Internet as it's, like, not as critical for me. So I just run it every time I need it, and I, kinda go through it, and then I'll show my doctor when I see him next time. And otherwise, I'll just, keep updating it until I find kind of, even better results or see if there's any other changes.
Sam Nadler [5:47]: Absolutely love it. I think it's a really smart way to leverage AI tools to stay on top of your health trends. I mean, I've oh, wow. Personally, I get the seal of approval from the doctor once a year, and then I kinda forget about it. I don't really keep track of how, my blood analyses are trending. My blood work's trending year over year. I have a little bit maybe in the last year, but, like, historically, have not whatsoever. So I think this is a really powerful tool, yeah, it totally makes sense to keep it locally. I don't think there's any need to get it online, but super cool.
Jordan Metzner [6:26]: Yeah. I mean, if you, if your doctor's using something like an Epic, like MyChart or something like that, then it becomes pretty easy to export your blood work history and any other history you have. You can add any medicines, medications you're on, etcetera, really just start to build out kind of this dashboard for yourself that kinda gives you insight into your health.
Sam Nadler [6:47]: Cool. Love it. We have a health themed focus episode today because covering our the product that we're reviewing is the new Codex app, which I've been using lately to build a kind of AI trainer. And, to be honest, it's very, very similar to Cowork. But what's been nice is I've been using the GPT-5.2 Codex extra high reasoning model, which I think, is a little bit different approach to Claude Code. I feel like Claude Code is a little bit more instantaneous, and that has its benefits. It makes it a lot of fun. But I would say, I took a previously existing product and then brought it over to Codex to run this extra high reasoning 5.2 GPT model, and I've found it great to kind of polish off some of those things that are like, a, some just some missing gaps and things I wanna wasn't even paying attention to, but just this really deep reasoning, finding a ton of bugs. You know. It usually takes, like, five to eight, maybe even ten minutes, so it's not as instantaneous as my experience has been with co work.
Sam Nadler [8:01]: But, I mean, this is somewhat boring to look at, and I haven't used it all that much, but, just to give a quick demo of what I've built. So this is my day by day personal trainer that's literally dialed in and will give you feedback based off every single rep, based on your goals and your overall plan. So let me back up a little bit, and I'll go to kind of my settings. So my training settings are a plan to workout, a workout duration of forty five minutes. My training goal is to get bigger, not necessarily stronger, just bigger, gain muscle mass, you can do both, and workout days, you have to do a minimum of two days. I chose five and that's based on like a specific split structure. My body focus is upper. You could choose balanced or lower. That doesn't mean you won't do any lower exercises if it's upper. Or if you choose lower, doesn't mean you won't do any upper exercises. It's just gonna be a smaller percentage, maybe 20% versus 50% in the balanced. Experience level is beginner is basically, you have no idea what you're doing versus experience. You've, yeah, you kinda know what you're doing. It's not necessarily an expert. Equipment access, full gym, home dumbbells, basic gym, body weight, and any limitations.
Sam Nadler [9:16]: I don't have any limitations. So, save settings, back to resume the workout. So every day you have a workout, and let's say I was practicing here, but, the AI is today is about executing the plan, not improvising first compound, full control. I've already logged the first set, but I made the I wanted to because I was kinda doing this in my gym already in a ChatGPT thread, and I want it was annoying typing out and letting it know exactly what I did, so I wanted the interface to be super simple. But let's say that I did this and it really felt like a struggle. So it then tells me you've reported difficulty but completed the target reps with acceptable form. On volume days, feeling the work while maintaining control indicates effective training stimulus continue at the same rate. Okay? So let's say, okay, let's maybe do a lot less and still say it was a struggle. Okay. Fatigue check. Step down. Focus on control. Step down at, it's now saying do less if you wanna continue or just end this exercise completely. So let's say do less and log the next set.
Sam Nadler [10:25]: It already pops up the rep, recording kind of element already down from 20 to 7.5 and reduces the rep. So let's log it. Boom. Hand exercise. Energy left. Are you full tank? Feeling good. Half empty. Running low. Still feeling good. Moving on to the next exercise. So, anyway, it's still in beta, but I found using Codex to develop this a longer kind of less engaging approach, but I think, like, balancing between the two or maybe using a slightly less reasoning model when you want quick iterations and then going to the deeper reasoning models when you want, like, to be really begin polishing off and being thoughtful and make it, more production worthy is is, fun trade offs to make.
Jordan Metzner [11:14]: No. Super cool. And are you using it all the time when you work out or what?
Sam Nadler [11:17]: I am not yet. I'm getting it to the point where I think I can begin testing next week. I am just using a ChatGPT thread that it's kinda where I got the idea. I wanted to see what it would be like, and I kinda began updating the thread after pretty much every exercises kinda with my overall goal. And, last week, started building it, and hopefully by next week, I'll be using it.
Jordan Metzner [11:43]: Awesome. Super cool. Yeah. I think this is all part of this, build software for yourself kind of personal software use case, and I don't know what alternative you would have used for tracking or whatnot. Maybe an app, maybe something like a spreadsheet or something, but, maybe now it's easier to build these kinds of things for yourself.
Sam Nadler [11:58]: Well, what I really want, and then we can jump into the news, is like the almost, if you can imagine a trainer after every rep is like, you actually struggled there on that 10 reps. I'm gonna like and the goal for today is volume versus like a heavy push. So like, I actually wanna drop your weight down and make sure you hit those 10 reps, and just having that constant feedback cycle.
Jordan Metzner [12:18]: Okay. Just real quick, let's jump into the news. So lots going on. Anthropic and Claude just announced Opus 4.6, their smartest model, just got an upgrade. They've also announced some additional features, which they call I think they call it a swarm. Yeah. They're they're just launching a ton of new features in Claude Code. Obviously, Opus has been the best model on the market for a while, so I think 4.6 probably, who knows what the upgrade is, but hopefully, they'll we'll see better improvements. I thought, everyone was expecting like a new Sonnet upgrade, which is the lower cost model, but, I mean, I'm excited about Opus 4.6 and can't wait to start and use it. Also, I think, if you saw here in the news, OpenAI launched this thing called Frontier, which is a new platform to help enterprises build and deploy and manage AI coworkers that do real work. So this is kind of an agent management tool. And, again, this is something Anthropic also launched something similar. But, the you see these companies going head to head against each other launching basically, like, feature parity one by one. So OpenAI is launching this kind of, like, multi user agent management and has mentioned they've been doing it with a bunch of other companies so far with pretty broad acceptance. And, Claude basically doing something or Anthropic doing something pretty similar.
Jordan Metzner [13:25]: And then even in the same news, here it says Perplexity is launching a new model called model council for all Perplexity Max users. Model count console mode. Council mode will let you delegate to a swarm of frontier reasoning LLMs. Will they work asynchronously, have a chair LLM synthesize, and more accurate answers considering multiple perspectives. So, this is, I think, what the biggest change we've seen since we've recorded this podcast till now. I think historically, everybody's been using one agent at a time, or, like, one chatbot at a time, one thing at a time. Maybe you ask ChatGPT something, maybe you ask Claude something, maybe you ask Gemini something. But for the most part, it's been kind of one off tasks. Ask them something, get a response. Ask them something, get a response. I think Claude Code kinda changed that in the sense that it would do things kind of in a sequential order, of being able to kind of create a list of tasks. But I think today, in episode 30, I guess we're in February 20 02/05/2026. This will come out on February 6. But I think the biggest change we're seeing now is multi agent management with other agents managing those agents. And I think that's gonna be the biggest real pivot here in using these LLMs is you become a manager of managers, and that manager is managing a bunch of other agents.
Jordan Metzner [14:35]: And so it's just gonna keep escalating where you're higher up on the hierarchy, there's and just gonna be more and more agents working for you asynchronously, synchronously, overnight, during the day, long term. I mean, maybe you have an agent that does, one month of research or something like that or one month of math or something. So you could see that all these different things are kind of coming together, but I think what's most interesting now, both with OpenAI and Anthropic, is kind of launching this super kind of manager of agents type of model where the agents can go off and do autonomous work and come back and bring back to you. So it is a super exciting state to be. They it looks like Anthropic continues to push inside of Excel, PowerPoint, and kind of some of the other Microsoft weak spots, I would say. And, yeah, this is what they call it with Claude Code. They call it agent teams. Set up multiple agents that coordinate autonomously and work in parallel, best for tasks that can be split up and tracked independently. So, this is a head to head battle. I mean, who even has enough time to test both of them right now head to head?
Jordan Metzner [15:28]: So, pick your favorite, test it out, see if it works. If not, act to the other one. But I know our developers are excited, and I think probably all developers are pretty excited. So it's definitely a hot week in the AI LLM news. What are your thoughts, Sam?
Sam Nadler [15:39]: Yeah. I just you know, it makes me think of my own workflow recently in the past couple weeks is literally just managing a few different agents. And I like I really, personally only have the capacity to, like, manage three or four at the same time, but that's kind of how I've been attempting it. But I think if I had some you know? I don't even I can't even having not used agent management platforms yet. I don't really know exactly how it's gonna work. But if it gives me the ability to, like, think through or effectively manage 20 different agents, I already feel, four x more productive with managing four. Like, how what is the productivity gains if I have this managerial level? They all have the same, security access, or I can manage different security accesses. They're having the same kind of the whatever UI, UX aesthetic feel, and they're just, able to share information between them. Seems like it could be extremely effective. I also find it really interesting. It's almost on the same day, within the same week, they're dropping these tools. The pace of development is is insane.
Jordan Metzner [16:44]: Yeah. I mean, they're aggressive. They nobody lets anybody have any free air. So, it's hard to tell if they have these things kind of in their back pocket or they're just like, this is the time to launch. But, we're seeing kind of feature and model parity launching on the same day, which means that there's obviously some type of, kind of alignment towards product development across these enterprises.
Sam Nadler [17:04]: Very cool. Anything else to cover in the news?
Jordan Metzner [17:07]: I you know, obviously, we've seen kind of, the impact of software companies on the stock market. I think, it says AI won't kill the software business. It's its growth story. I think it's, it's still unknown whether companies like Salesforce and Workday and some of these other ones will be able to adopt AI and become leaders, or will kind of a new AI startup kind of take their place and replace them? And I think, like, there'll be some winners and some losers like we've seen through every other cycle, but it seems like it's now still too early to tell.
Sam Nadler [17:33]: Yeah. I think it's I think there's a little bit of panic. I think things will settle down. I do think over the course of, twenty four months, we could see a few, previous giants get smoked or have a hard time, for lack of a better word. I mean, the you know, internally, we're we're building software for ourselves that we'd used to buy off the shelf and, to have ten users, 10 seats, and, that's changing. That, build or buy trade off is is definitely different these days. Anyway, Jordan, it was a great episode. Like and subscribe. Any final thoughts before we sign off episode 30, which feels like a milestone?
Jordan Metzner [18:14]: Yeah. Episode 30, I think, what's interesting on today's episode is we have the launch of the new OpenAI stuff. Sam Altman said there's some new Codex updates. We see that Anthropic launched this 4.6 plus these the team's manager. So I think, like, next week, we'll probably have another crazy episode, but I think we'll see the impact of dropping these products and how they've had an impact on the greater ecosystem. So, again, another fun episode. Like and subscribe if you wanna hear more on builtthisweek.com, And thanks, Sam. That was really awesome.
Sam Nadler [18:41]: Thanks, Jordan. Bye, everyone. Bye bye.
