Claude Code’s Shining Moment, ChatGPT for Healthcare, End Of Busywork?

0:00Cloud Code is on a legendary run and becoming useful beyond programming. Open AI rolls out chat GPT for healthcare. Prediction markets have an insider trading problem, or do they? And is the end of busy work a bad thing? That's coming up on a Big Technology Podcast Friday edition right after this. Fiscally responsible, financial geniuses, monetary magicians. These are things people say about drivers who switch their car insurance to Progressive and save hundreds. because Progressive offers discounts for paying in full, owning a home, and more. Plus, you can count on their great customer service to help when you need it, so your dollar goes a long way.

0:42Visit Progressive.com to see if you could save on car insurance. Progressive Casualty Insurance Company and Affiliates, potential savings will vary, not available in all states or situations. Welcome to Big Technology Podcast Friday edition, where we break down the news in our traditional, pool-headed, and nuanced format. Matt, we have a great show for you today. We have filled our document with a ton of stories to go through, including what's happening with Cloud Code and why it might be applicable for folks beyond programming. Maybe it will end up impacting much of knowledge work. That's at least what Ethan Mollick has to say.

1:18We'll go through his post on the matter. We're also going to talk about OpenAI's foray into healthcare. It's actually an official foray into healthcare because millions of people have been using ChatGPT for healthcare already. We're also going to talk a little bit about prediction markets and the end of busy work. Joining us as always on Fridays, the dude is Ranjan Roy of Margins. Ranjan, great to see you. Welcome back. Good to see you, Alex. Excited to be here. Let's begin just talking about CloudCode. I don't know if you've seen this, but for me recently on X, CloudCode has been a meme. It's almost like those Chuck Norris memes where people have been talking about how CloudCode could basically do anything.

1:57and it's got these superhuman capabilities. And I was waiting for a post to kind of illustrate what was going on with Cloud Code that had made people so excited about it. And I think we got one from Ethan Mollick, the Wharton professor, a friend of the show. And he writes the one useful thing substack and his post is called Cloud Code and what comes next. And I think he really illustrates just the amount of autonomous work that Cloud Code has been able to do and the fact that it's not simply a code autocomplete anymore. So let me just begin with the story that he starts with. He says, I opened Cloud Code and gave it a command.

2:38Develop a startup idea that will make me$1 ,000 a month where you do all the work by generating the idea and implementing it. I shouldn't have to do anything except run some program you give me once. It shouldn't require any coding knowledge on my part. And so you make sure everything works well. Malik writes, the AI asked me three multiple choice questions and decided I should be selling sets of 500 prompts for professional users for$39. Without any further input, it then worked independently for an hour and 14 minutes, creating hundreds of code files and prompts. And then it gave me a single file to run that created and deployed a working website.

3:17Right. Let's just talk about, you know, we're going to go through what the ideas that Malik brings down in terms of why this is an improvement over previous generations and what the implications are here, why people are excited about CloudCode. But just to start, are you impressed with the output that Cloud spit out here? I am. This specific output, I'm actually not that impressed. I'm interested, but I still think, and I'll definitely get into this in terms of like overall what I'm seeing with this kind of world of autonomous agentic. But I think again, like writing 500 prompts that you can package up, maybe even creating a website, taking all these steps is actually incredible, but it should be table stakes.

4:04And more and more people are realizing every day that it is table stakes, that the technology is already there to do this level of work. And I think, remember, what was one of my predictions from 2026? Agentic AI is going to be real. And I think Claude Code is a lot of people's first entry into this world of truly agentic AI. And even Malik goes on to write, he was like, the latest AI is capable of doing work more autonomously while self-correcting many of their errors. But then the AIs are being given an agentic harness of tools and approaches they can use to solve problems in many ways. That second part I loved because this is what I've been talking about for six months now.

4:47Agentic AI is not like a process flow diagram. It's giving it a set of tools and letting it go figure out what to do and actually go do it. And I think Cloud Code, the excitement around it is everyone's finally getting a taste of that. That's right. And so the way that Malik frames it is the combination of these two things that the AI can do autonomous work and self-correct and then have this agentic harness, which is jargoning to me, but basically be able to call tools. I like it. Agentic. Oh, come on. It is a terrible. No, bad. Very bad. What do you like about this? Okay, go ahead. No, no. Hold on.

5:23I like, okay. I have been saying like a predefined set of tools and data connectors. That doesn't sound exciting. agentic harness. It's like kind of just, you know, holding it generally in place, but still letting it go. I don't really know much about horse riding or how harnesses work in any kind of situation like that, but at least in my mind, that's what I'm thinking. I like it. No, it's not. It's terrible riding. I mean, I really like the concepts here and it's bad. It's jargony. You're too close to it, Ranjan. Here's all you have to say. It can do autonomous work and self-correct and it can use tools.

6:00That's it. Why do you have to make it more complicated harness is shorter agentic harness i'm using right agentic okay we're going to move on after this but we are not accepting the term agentic harness in the show it's just it's not working not going to work agentic harness of tools and approaches they can use to solve problems in new ways why do you have to use the phrase agentic harness why can't you just write they can they are the ais are being given tools and approaches they can use to solve the problem so so it's unnecessary jargon. This is where I think like, and I brought this up on this podcast like six months ago at Writer where I work, like, and we're enterprise AI focused.

6:42We launched a tool and I'd started testing it in June. It was, and we joked that you were like, you know, now you're at an AI company, you're feeling AGI. But it was the first moment I actually felt like, holy shit, I gave it some direction and now it's going out and doing all this other stuff because it had a predefined set of tools and data connectors. One might call a harness. But anyways, like feeling that autonomous AGI-ish type action, I think like what Claude Code has been the breakthrough. I've been seeing this for like six months now. Like this is happening. This is no longer us talking about can I book a flight and have it, you know, search and go do it for me.

7:23Like this is going and doing tasks for people people are doing already. And Cloud Code is the first time a lot of people are experiencing with that. And that's why I honestly think this year this is going to break through in a big way the way coding assistants were 20, the story 2025. Right. And that's sort of getting past our differences on languages. I think that's why I thought this was important. And that's why I think this was worth bringing up here is that I think that the common conception is that, you know, AI for coding is just that like it's code autocomplete. Or in some cases, you know, you put a vibe code prompt in Replit or something and then you get a website.

8:05I think what's interesting here is that you can give just the prompt and it will end up, you know, in a way that that Malik has specified without him having to touch the code. it can end up building a website for him, which includes a sales funnel, by the way. And basically all he has to do is put it live. And the way that it's doing this is the long autonomous tasks of coding and then also being able to use the tools. We won't argue anymore about the language. which he says the result of these two factors has led to big leaps in the latest AI tools made by the big AI companies. Now, listen, if you're at this point in this discussion and you're not a coder or you're not building a website, you might be asking yourself, well, okay, this is very specified and what's in it for me.

8:58And this is, again, like to go why we're leading this show with this, it starts to broaden out. And if anything, this might be just the beginning of what's coming for the rest of knowledge work. And I'll just continue to read Malik because he's very good on this. He says, unfortunately for most of us who want to experiment with AI, these new tools are built for programmers. In a lot of ways, this is a shame because these systems are actually broadly useful to knowledge workers of all types. And by seeing what you can do and experimenting with them yourself, I think you can learn a lot about the future of AI.

9:36He goes, to return to the example of the startup company launched by Cloud Code, it was only touching a small part of the capabilities of what the tool is capable of. If I ask it to do user testing of the live site from different personas and give me a report, it deploys another tool connecting to a web browser on my computer. It takes control of the browser, goes to the site it created, and scrolling through as a human would. He said he asked the Cloud Code tool for a critical report of the UI, and he said it did a better job of nailing potential issues and spotting some sketchy fake reviews on the site.

10:13And he says, as a next step, I could easily ask it to implement its suggestions, continuing the process with minimal input from me. This to me is really the interesting part is you can not only have it go create things, you can have it test things, and with natural language, then go and optimize. And if you think about, maybe it's a website, but maybe it's also, and we're gonna get into this, a folder of documents on your computer. And you have it go in and start to give you feedback on it, make data visualizations from it. One example that he gave is his credit card statements. So you could just kind of download them onto your desktop browser and obviously do this with caution and then set Cloud Code loose on it.

11:00and maybe makes visualizations, spots anomalies, gives you a report on your finances. And you need that more advanced capabilities. You need the ability to code and to call tools to be able to go and do this type of work. It's pretty crazy to me. I mean, it's real. It's happening. I'm living in the future, Alex. I'm already there. But I can tell you, this is where I think the big shift I saw is early 2025, when people talked about agentic AI, again, these thinking of like these big, heavy processes and how do we automate them in some way and bring an LLM's intelligence to make a decision in one part or maybe create some copy in another part.

11:45The big shift was completely rethinking like what agentic AI is. And again, won't say the word harness, but allowing it to start to take these tools. And I think what he showed there, like around this, like creating an entire business, launching it at 500 prompts at 39 bucks a month. Are you really going to do that? Maybe, but probably not something. It's these kind of like smaller multi-step processes where I am calling by the end of the year, all of us are going to start incorporating in our life in a lot of different ways. Go check my credit card statement, find any charge above like$25, come up with a summary, like email me, stuff like that is going to become routine for, especially at least any early adopter, I think by the end of the year.

12:35And it is interesting. So we've talked, remember like last year we were talking about like, is AI going to hit a wall? And there were all these different things about like, well, is pre-training still continuing to work? And I think, you know, there's another bad jargon word, which is like the scaffolding, right? Which is something we've used like what do you build around these tools uh that that enable them to sort of cheat their way through the limitations of you know the bottles topping out or for instance the context window and context window of course is the amount of text that this model the model can work on before it sort of runs out of its ability to like keep it in the like sort of memory window so to speak uh and and it's just interesting to hear how this stuff works because uh because malik even talks about how like the sort of Claude code cheats its way through like the limits of the context window.

13:22He says, when Claude code runs out of context, it stops and compacts the conversation so far, taking notes about exactly where it was when it stopped. Then it clears its context window and the fresh version of Claude code reads the notes and reviews the progress to dates. And then it gives it, those notes give Claude everything it needs to keep moving. This is why Claude can run for hours at a time it carefully notes what it's doing along the way produces interim work and um you know like pieces of software and reports that it can refer to i mean this is this is very interesting the way that these companies are architecting stuff yeah i mean this is where anyone who was using claude two years ago year and a half ago and would hit that dreaded like you've run out of uh credit even as a paying claude pro subscriber and have to either wait 12 hours to run a prompt on the leading model again like the architecture side of how people are approaching these kind of agentic workflows is getting solved i think that's the big thing and it's not at the model level that models are helping but and it's yeah it's the it's not the product actually you know what instead of product versus model architecture let's throw in there as well as another layer that can be optimized or innovated on to actually to really move this stuff forward so are you team architecture i think i am team architecture i mean this is something you're on stage with microsoft larry or sergey that you talked about architecture well sergey talked about algorithmic improvements okay and so i think that under architecture yeah Yeah, I would call it architecture.

15:02And it's starting to show up in other places, too. Here's Cloud Desktop. So this is a relatively new thing that Anthropic has released. You can put Cloud on your desktop if you're in the$20 a month bracket of the paying users. and Malik says, just give the AI access to a folder. Remember that Claude can do anything to the files in that folder, so be careful if it's sensitive and make a backup and you can start working with AI. Have it research and write reports. Give it access to your credit card records so it can put them into a spreadsheet and tell you about anomalies, ask it to do a data visualization, whatever else you like.

15:43All right, Ranjan, so as our resident AI agent expert, I'm going to ask you a question about this. Do we really need to be running this stuff through Cloud Code or can you just upload it into the context window of the Cloud chatbot itself? What does going through Cloud Code give you? Is it just more power because it can build things on the fly for you? Tell me a little bit more about why this is important. Yeah, it's the ability to, again, it's multi-step. It's not one action. So it's a set of, like, especially giving that general guidance and action, thinking of it from an outcome perspective, it's no longer telling Claude, here's what you should go do.

16:27It's saying, here's what I want. and to get there, like that's not how a typical chat back and forth works. The chat back and forth is let's iterate together. Let's explore. Let's try to understand versus I know what I want. Go get there for me. So I think and that ability to combine whatever the actual kind of like reasoning side, if it needs to actually create like hard coded Python scripts that are deterministic and are going to be used, it can do that. like not limiting it to that chat window is the biggest difference here. And again, increasingly, and again, talking about my own company here, like we see at writer people, the biggest change is the routine side of it.

17:10It's these things are no longer going to be one-off things you do for fun and kind of try it as an experiment. You start to set routine. So this kind of work is being done on an ongoing basis. So it's not, I'm going to go in and upload a document. It's actually happening of the background and like the the claude ui gets abstracted away like maybe you just get an email or a notification with whatever results so so i think it's a big difference concrete examples give some concrete examples of like the type of routines that i mean like not to get too into the enterprise sales world but like right now at work we have like salesforce we have gong which is like a call recording and data type thing.

17:53We have something called Zoom. We have all these different systems. I literally have a list of accounts. In the past, I would go through every morning. I would either have like a report in each of one or a dashboard. I literally was able to create something where it's like, go in, here's a list of accounts. Did anything happen in the last 24 hours? What happened? Summarize it. Email me. And like I built that. That's a whole app. There's tons of code in there and API calls and MCP connectors. And I can do that stuff now. Yeah. So this is where, and I think I'm a reasonably technical person, but anyone, especially simpler levels of this, will be able to start doing.

18:33Okay. And that's a good segue to the end of the post here from Malik. He says, today's AIs are capable of real sustained work that actually matters and in turn is starting to change how we approach tasks. He says, it is starting unsurprisingly with programming. One of the more famous coders in the AI world, Andre Karpathy, recently posted, I've never felt this much behind as a programmer. The profession is being dramatically refactored as the bits contributed by the programmer are increasingly spare and between. I have a sense that I could be 10x more powerful if I just properly string together what has become available over the last year or so, and a failure to claim the boost feels decidedly like a skill issue.

19:16Malek concludes, Don't let the awkwardness of the current cloud code or specialization for coding fool you. New, oh God, new harnesses that make AI work for other knowledge work tasks are coming in the near future. And so are the changes they will bring. Oh my God. Sorry. Okay. Hold on. One, I love this quote from Carpathie. So I'm trying to contain myself or regain my composure after seeing harnesses again. But okay, Carpathie. I loved this quote. I think it came out over the holidays when I was at home, caught it on Twitter, like started seeing it more and more because like this is how I feel as a fairly technical but still knowledge worker at heart.

20:01Like I feel when I'm seeing what I'm able to do, I feel like, holy shit, I think I'm falling behind or like I don't even know if other people are getting this faster than me that they're going to be way ahead. And so that feeling was kind of like lingering in me the moment you start to see what's possible. And I think like when I saw him say that, especially about programming from one of the leading world's leading programmers, like it starts to make us feel like this stuff is so interesting right now that even someone like him can actually feel like I better get a handle on this right now. right and look i'll just say this i'm not completely sold that this is where we're going i know you feel firmly you're also like you know you're invested you have a built-in bias you're yeah because it's where you work i'm not but here's the thing i've seen enough uh energy around the increases in capabilities the type that we spoke about uh at the beginning of this conversation the type that uh malik listed as he explained what's going on uh and i think that that has been clearly drawn out by Malik in this post that I think it would be foolish to ignore it the same way that Malik is basically saying like don't be fooled by this being just a coding specific thing soon we'll probably have the interfaces that will make this type of work autonomous work things that self-correct things that call tools available for for many many more different types of knowledge work professions and that's why I think that this is important to highlight I mean we are getting of obviously we're getting, you know, a year is just a delineation in time, but we are getting started here in 2026 after what I think was a lot of movement in this space.

21:46Obviously, a lot of money went into it. We've been very critical about the financial picture of some of these companies. But I think that shouldn't blind us to the fact, and again, we're here for like nuance about this. That shouldn't blind us to the fact that this technology is progressing in some very interesting ways. And our mission on the show is also to be able to be like, hey, pay attention to this because it's something that we think has the potential to be big. And I think that's the case here. I think so too. And I'm going to make the bet here. Right now, every week, Alex and I have a collaborative Google Doc that we paste links in and kind of copy and paste text into.

22:20I think we are going to not be working in that way and have something more interesting as a process in a few months time. I'm making that call. I will take the, listen, I may have just talked about how I believe in the power of this technology. I will not bet against Google Docs. I won't do it. We're going to be doing the same thing. And I will not change my process. That's my bet. Love to paste a link. It's actually the act of pasting the link is the most human thing you one can do. I'll agree. Well, also, I mean, I think it's sort of just to not to over index on this Google Doc discussion, but why not?

23:02I mean, this is the problem, too. People have processes. They like their processes. And it takes an unbelievably better process and system to be able to get them to switch off. And that's probably the biggest thing that will hold this technology back over time is just the fact that, like, people like their processes. and if you say here's this brand new agentic way to do your you know show planning or a take tasks a b or c uh their their immediate reaction isn't going to be like oh hooray let's let's switch over their original reaction is going to be like over my dead body you you got i mean honestly this is the biggest thing the technology is there like i i've been saying it for a while the technology is there it's a change management person human ui whatever problem but like exactly as you're saying we have we have a process it's a beautiful one it brings us to this show every week and and it gets it done but yeah people are tied to process so yeah i think that's going to be the most interesting part of this year but like the way i see it already kind of going is the more people hear about there are better ways to do things, the more you, especially early adopters, you're like, I got to at least try.

24:20And maybe technology will reinvent the world, or maybe AI will reinvent the world's oldest technology. And that is, of course, email. And maybe it will change processes there. This is from TechCrunch. Gmail debuts a personalized AI inbox, AI overviews and search and more. Google has unveiled a new AI inbox for Gmail that's designed to provide a personalized overview of your tasks and keep you informed about important updates. Google is also launching AI overviews in search and a grammarly-like proofread feature. Additionally, Gmail is bringing to all users several AI features that were previously available only to paid users.

24:57The new AI Inbox, and this is I think where the most interesting news is coming, the new AI Inbox tab features two sections, suggested to-dos, and topics to catch up on. The first section displays summaries of top priority emails that require an action, such as a reminder that you have a bill due tomorrow or that you need to call your dermatologist to confirm your mailing address so they can ship you your prescription refill. Under the topics to catch up on, you'll see updates such as your Lululemon return has been processed or the end of the year statement now available from Wealthfront. What do you think about this?

25:29Is this basically the same personalized inbox that we have from Gmail, I mean, Gmail has primary and updates. Is this the same thing? Or do you think the AI layer on top of it is gonna actually change the way we email? So I love this article and I'm glad we just got to it right now because I created a system taught to me actually by my coworker at the time of the Financial Times, Stacey Marie Ishmael. With Gmail flags, you can create like customize instead of just the star or not star, you can have, there's like five different options and like creating a tiered ranking system And I have my system.

26:05It's from 2011. I'm still using it in Gmail today. I turned off primary inbox and all of that. So I am curious to see whether I start to use this. But on the other hand, like email has to change. We all know that email, like what's the value of the actual email? Is there a better way to do it? So I'm glad they're at least trying something here. well this is kind of exactly the version of what you were talking about before that has been so useful is that it sort of sorts through uh yeah yeah i guess the action in your inbox and prioritizes it for you and summarizes it so you don't have to like similarly you don't have to go through every system this one maybe you just don't have to go through every single email yeah no no so so that's why i was laughing because like the fact that i probably will stick to my 14-year-old system, even though this has just been offered to me to give you a little bit of credit from your earlier point there.

27:03But yeah, I think the other interesting part of this to me is it's such a rich opportunity for Google, for anyone who has not been using Gemini to kind of introduce AI to them or introduce this kind of more agentic AI, because this is in their inbox. Email should be a very straightforward. It's just it's mostly text. It's very structured. There's subject lines. And like it's something LLMs should be able to do a very good job of handling. There's a lot of context. And you imagine if this starts to be the way you get the next. I don't even know how many hundreds of millions of users Gmail has. Is it billions?

27:44Billion plus. Billion plus. Suddenly. I bet it's probably two billion. I mean, it's absurd. It's ubiquitous. suddenly all of these people will start to over 2 billion users active worldwide suddenly all of them like as an entry point to bring gemini to all these people and start to make them think here's how to use like slightly more agentic ai adding more like now that we've summarized this for you do you want to go do something actually given our earlier conversation about manis and metal last week as manises they're kind of like a trojan horse into consumer maybe this is google's entry point into consume like true consumer agentic okay i actually got an interesting uh email from a reader just to like put a bow on this manis conversation that i think is just worth reading.

28:41The reader says, keep in mind that this is, we were talking again last week about a meta acquired Manus, the AI agent company. And I was like, oh, maybe that's actually going to be the thing that they start to bake into their consumer chatbot. I got a lot of like, well, not a lot, but some reader feedback being like, no, this is an enterprise thing. So let me just read that. Then we can take our break. It's a great message here. The reader says, keep in mind, or listener, actually, listen to mail when you do a podcast. Keep in mind that Meta's customers are in the business of advertising on its platform, not the users.

29:15At the end of the day, it matches businesses with customers and the more businesses it can pull into the platform by making business operations, reaching customers easier is the better. Zuck has previously said our goal is to make it so that any business can basically tell us what objective they're trying to achieve, like selling something or getting a new customer and how much they're willing to pay for each result and then we do the rest. I imagine agentic workflows will be super important in achieving that vision. Meta is attempting to capture and automate the entire funnel from the product ideation to sale.

29:47Okay, that's an interesting alternative perspective that we didn't cover last week that I think makes a lot of sense. Yeah, I think that's reasonable. Again, like in any of these areas, yeah, business model that drives the entire company is going to kind of influence things. I guess if you think about it, like Facebook workplace, remember that? Or does it still exist? Yes, I think they shut it down. Okay. Like that actually is a perfect example of there is no reason Facebook and then Meta should not have been able to do something really interesting in that space. It's a very lucrative space. There's no reason like they should not have been able to do that, but they didn't.