220 - How Claude Code Just Replaced All My Coding Tools with Elizabeth Knopf

0:00The reason I put ChatGPT third is just because I think it's the Honda Accord of the models. You know what I mean? It's like, it's good. It's going to last. It's going to get you to where you need to go. You're not driving an Audi, but you're also, you know, you're not getting a Hyundai accent. All right. Flawed code is the only thing that has come out recently where universally people are like, this is mind-blowing. This is incredible. It's like you're back in the 90s here. Yeah, but don't get intimidated, anyone watching. This is criminal. Like legitimately, I feel overwhelmed. I really do. Please help me, Liz.

0:37Liz, I don't know. It's been a couple of weeks since we've spoken because the summer's crazy. So much has freaking happened. And what I want to talk about today, which we're going to get to is Cloud Code. I think it's the only thing that has universally been talked about in a positive light. You told me just before we hopped on here, that's going to blow my mind and that you're kind of doing away with all of your other tools that you've been working with because Cloud Code is that freaking good. So you're going to walk me through exactly how to open it up, install it, put it on my computer, and then like actually how to use it.

1:07But let's set the table a little bit for people before we get to cloud code. Right now, the model war has not only heated up, it feels like it's kind of cooled down. No? Yeah. Yeah. A little bit. I mean, I don't know if it's the summer doldrums and, you know, people were on vacation just sort of chilling, but But I've just felt like people who are both doomsday folks as well as like hype, you know, EACC, whatever, you know, folks that are just like, it's going to be utopia. We're all automated. Both sides have just sort of simmered. That's probably because of a couple of releases and also people sort of actually seeing the adoption or even the lack thereof of adoption within organizations.

1:53So at least that's the vibe on Twitter. Okay. So the thing that was crazy to me, what we've seen over the last few months, we've seen Gemini 2.5 come out. We've seen just in the last week, Claude Opus 4.1. We've seen Grok 4 come out. And now we've seen ChatGPT 5. And what has been interesting about each one of those model releases is I think everybody was holding their breath for ChatGPT 5. Sam Alton was on the press tour and he's like, GPT 5 is the first time that I actually felt like, wow. You know what I mean? And he's like breathlessly describing GPT-5. And then it comes out and we start looking at this.

2:30So this is my chat GPT. I asked it to compare all of these new models. And the interesting thing is if you look up here on the top, we've got all of these tests that they've used to sort of score these models. This over here, this is humanity's last exam. It's a crazy, crazy hard test. And you can see here GPT-5 scored 25 and 26. I don't know what the text or tools only difference is, but let's just combine those two for right now. So it's a hard test. It scores a 25 out of 100. Grok 4 scores a 50. Okay, that's weird. Grok 4 is a better model. And then I start going and playing with GPT-5. And at first I'm excited because they're like, oh, the persistent memory has been improved.

3:13Oh, cool. Now I can remember all of my chats across all of my projects, which I didn't used to be able to do. So that was a really cool improvement. Then they were like, and we've improved the way that we actually process your prompts, the way that we batch them. Now there's no, there's an equal weight given to your questions as opposed to being more focused on the end of your prompt, as opposed to the beginning of your prompt. The processing is supposed to be faster as well, because of this new batching, I guess, technique that they've implemented. And they've taken away all of the 03, 03 mini, 03 mini high, you know, Chatsby B4, whatever.

3:49And it's just kind of like three models. But I've played around with it. It's okay. It's like a little bit better, but it just kind of feels this time like an iteration and not a step function change. And I think that's what you're talking about. Like the vibe is kind of shifted from you're wowed every six months with a new model to, okay, now it's just iterative. Yep. And I think what's going on, so there's a couple, a number of things you just highlighted. So first off, it does feel like a lot of the models and their pictures are just converging there in terms of being, having parody. So, you know, one will have this feature around video generation, the next will pop out with that sort of what it seems like is happening overall.

4:31But I do think with ChatGPT5, like we were expecting, yes, to have it blow our minds. But one of the things that I think people are at least underappreciating is sort of who the audience is and then what they did with the product. So less so about like the technology behind it, but actually the product. So I think one big challenge was that, you know, you had to select the model. Like most people using chat GPT don't know what three means, 3.5, four, five, like they don't know the difference. Then they started giving, you know, more options. And there's the concept of the paradox of choice. when you have too many options, you're overloaded and you don't know what to use for what.

5:10I personally like that because you could get different outputs and you'd get some interesting nuances when you played around with those different combinations. However, for the average person, like they've simplified and actually sort of reminds me of like Google's first search bar, you know, they had just one box and now that's what ChatGPT is. It's like one box, no choices, there's like, here you go, put in your prompt. And so I think from a knowing that ChatGPT has more of the average consumer across their large distribution. They're removing friction, basically, is like they're making it easier.

5:48And they've made it multimodal. And what I mean by that is typically people talk about multimodal as text, picture, video, sound, AI can do all those things and then spit something out. That's not what I mean. I mean, in this, in this context, it can kind of toggle between the chain of thought that you would get with the O3, the LLM based model that you're getting with 4.0 and the image generation. So it's like making its own judgment based on its own discretion of which model to use based on the question that you're asking. So that, that part is cool. I think the other thing that's cool is it does a much better job but multi-step tasks.

6:27Yes. So for example, if I'm like, go do research on this market, then take the research and run it through this criteria, then take the output from there and rank all of them from best to worst. It used to not be able to do that. It would just, that would be too much for it. It's gotten better at sort of this multi-step execution. So those things are all cool. But to your point, they just made it easier to use, right? Like the iPhone has always just been the iPhone. It's just simple. Everyone knows what they're getting out of an iPhone. And I think they took a product step. I don't know if, do you think that's a Johnny Ive thing?

7:00Like, is this a Johnny Ive fingerprint? Or do you think this was in the works? I mean, it probably was in the works because people believe this feedback. I mean, this was all over the internet from news. We could sort of see this. And I think there's that piece of it. And then there's probably the cost element too, because like, you know, if you run a simple prompt for deep research, you know, that eats up a lot of costs. And while there was a lot of rate funding, everything, you know, they're probably doing it from a cost optimization standpoint. You know, they're probably burning a ton of cash on, you know, GPUs.

7:35So I bet that was also a factor in how they came to doing this. So it was probably some users, some of that, maybe Johnny Ive. I don't know. So we talked about this. Claude code is the only thing that has come out recently where universally people are like, this is mind blowing. This is incredible. We have to do cloud code. So we're going to get to that in a second before we do. Now, after all these models have been released, how would you rank these models best to worst? Let's just use the four horsemen, grok four, Gemini, chat GPT, cloud. How would you rank them? So I will qualify this. It depends on the use case, but I will do a net net in terms of actually how I'm using or which ones I most frequently use.

8:18So I'll use that as metric. So Anthropic, Claude. Then I would say Grok for sure. And Grok, I really like for like real-time information. And then I'm just sort of using it with coding. But again, and getting back to the friction point that we'll get into with Claude Code, there's still just so much friction with all the other ones to like build things. And so when I'm just using it from like conversation and research, that's yeah. And then third, Gemini. And then I'll say Tachibiki is my lower. Wow. Yeah. Okay. So here's my ranking. So video is cool, but I have what scientists call a face for radio.

8:57And so what's even cooler is long form audio via my podcast and my newsletter. Nickonomicspod.com. Go there for free. Subscribe to my newsletter. It's one email per week. Super tactical. And then go to my audio podcast. I do three to five episodes a week, depending on how curious I am. And it's stuff like this. It's all free. No sleazy sales pitches. nickonomicspod.com number one grok i don't use grok the most i do think it's probably the best just general purpose model at this point pretty crazy that they become a frontier model in like two iterations nuts two claude dude like the 4.1 update is it's really good and claude has done a really good job i think making it easy for the user their projects are incredible the artifacts now, the way that you can store artifacts is I'm using it all the time.

9:45Like you were the first one to show me. I use it all the time now. It's incredible. Third is ChatGPT. And then fourth is Gemini. The reason I put ChatGPT third is just because I think it's the Honda Accord of the models. You know what I mean? It's like, it's good. It's going to last. It's going to get you to where you need to go. You're not driving an Audi, but you're also, you know, you're not getting a Hyundai accent. You got a nice car. It does what it's supposed to. It's a good thought partner. And frankly, I still use it the most just because it has so much information on me. It can quickly give me good answers.

10:18And I'm trying to migrate to Claude and Grock, but it's going to take some time. So that's my ranking for it. Well, and I think you bring up a really good point is there is a stickiness to all these products. We're probably power users versus, again, the average person using them. I know I've heard people say, oh, I set things up in Chachupiti, so I'm just sticking with it, or that's the one I want to learn because there's a perception that it's hard to get into the other ones or set up all your contacts. And so I do think that's huge value that Chachupiti has. And so even if they sort of stayed at this level for a while, they're not going to have much churn or attrition or see behavioral changes from people switching.

10:59Power users, yes, maybe. But I think you're starting to see these guys sort of focus on a specific type of user, or at least like consider the majority of their base and cater their iterations and development towards that. Yeah. It's very easy to get lost in the moment and think like, oh, chat GPT is falling behind. Oh, open AI. Oh, Sam Altman. Yes, all that stuff is probably true if you're forward looking. But what they've done in the last two years, three years, however long it's been since they launched 3.0. I think it was 3.0. That was the first one or 3.5. Anyways, it's undeniable. Even if they don't grow at the same rate, even if they lose market prominence, they've got a multi-hundred billion dollar company.

11:39It's a legit company. They've got a good product now. They've got a good team. They've probably lost a lot of people to the brain drain because freaking Mark Zuckerberg is paying a billion dollars to people. It's like sports contracts. It's nuts. But they're entrenched. They're a market leader in this space. And to your point, I think we underestimate how sticky these products are and will be moving forward. So, okay. You're going to show me cloud code. What do I need to do? Do you want me to share my screen first? Or do you want to share your screen? Share your screen so that we can go through like step-by-step.

12:10Here we go. Cloud code. So I click on cloud code over here and I get this like settings. It doesn't just come up. So it's actually not in cloud. You have to open the terminal on your computer. So I'm going to do this step-by-step thing so that people can actually follow along and take advantage of this really, really cool iteration that Claude has released to us. So if you're in Claude, you come over here on the left-hand side, click on Claude code. Now, in order to Claude code, you have to have the max or the pro plan. And I think it's, is it$100 a month? I know one of them I want to say is like, yeah, one of them is$100.

12:45It's not cheap. It's not cheap. But I think after you see what we're about to do, you'll be convinced. So you click here. We're going to click on install instructions and this window pops up. Okay. And this is where I got like overwhelmed. Cause I'm like, I don't know what the, I don't know what I'm supposed to do. I'm not a coder. What's this green screen. I'm not, am I trying to hack into the Pentagon? What am I supposed to do from here? Okay. So actually just back up one sec. So people aren't turned off by the a hundred dollars pro is only$17 a month. So it's accessible. Sorry. Sorry. max is a hundred dollars a month but you will get you probably want to graduate to the max plan once you've utilized the pro but okay so okay good call 17 bucks so it's accessible okay okay so now you're here at this little black box scary looking text so what you want to do is in that upper right hand corner is that little copy icon it's within the black box not copy page oh yep yep copy that All right.

13:43Okay. What you're going to do is go to. So you could also come, is it come here? I think I opened this link. Okay. And you're going to, you'll walk me through it. I know. But this was the link that I just clicked on. And this kind of walks you through is like, all right, step one, install cloud code. And this node.js18, what is that? That's just a language. So there's all these things, libraries and languages, which is basically like, Just think of it as sets of code that your computer can access and use. They're not all set up on your computer when you open up or about to open up. So you do have to copy and paste these lines.

14:20You don't even need to type them. Just copy and paste these lines and I'll include those. All right. So you're looking at my terminal now? Yes. Okay. So for folks at home, make sure you have Node.js 18 or newer. Now we can get into this, which is installing Anthropic. So I'll come here, copy, paste. Yes. Cool. Now actually just see if you just type in Claude what happens. Ready? Yes. Okay. So allow, allow. So is this it? This is it. This is it. You've arrived. It's like really like you're back in the nineties here. And yeah. Don't get intimidated anyone watching. This is like legitimately I feel overwhelmed.

15:06Yeah, I really do. Please help me, Liz. There are better user interfaces that we'll get to maybe in a future video, but we'll keep it some. So pick your dark mode, light mode. Pick your. Oh, do I just click on it or like what do I do? Put in the number. Where do I put it in? You just type. Just start typing in it all. One. Okay. Yep. two what the freak what's happening now it connected to it needs to log in to your actual anthropic account so it opened up the browser on your behalf and now you got to log in to whichever account is associated with your cloud account i've never worked in a console before so i just type i don't like yeah so there are some like things that get annoying like copy and paste like to go back like you have to delete and it's like, like you have to, it's not the best interface and you'll get annoyed at certain points, but for the most part, like this is the best starting just to get a bit.

16:06Do I just need to make sure I remember that it's in users, Nicholas Haluski. Okay. That's my file path. So for your file path, you can create new ones from here as well. So once you get into cloud, you can actually tell it, create new folders. So think about where you want to have all your files that it'll be generated. Interesting. So if you want to have it, like right now I use documents as the main folder and then I have cloud code is literally the name of the folder. And then from there I have objects that I'll name and then. But it's stored locally. It's not, I'm not storing this to like the Google drive or anything.

16:40Right now this is not connected. Well, I shouldn't say it's not connected to the internet because it is, it's not like the files you're generating are not going to be like hosted on a website. They're going to like, when you open up what you've done. It'll be on a local site called localhost colon, and then I'll have a number. It's asking me to, how does blank file path work? So I just enter in a document. Yeah. So just say like, say like what, where would you want, say, create a new folder? Okay. This kind of feels MCP ish. Am I off? Sort of, I guess so. A little bit, a little bit. It actually does use MCP in some ways when you need it to, but I want people just to really, you are just speaking English to it once you're in Claude.

17:26I guess what I meant by MCP is like, I can now have access to my local files on my desktop without having to go like copy, paste, import. Yes. Wow. Okay. Okay. So Claude code has been created successfully. Okay. So now we get to build a project. So I don't know if it would be helpful for we could kick off one or I can maybe give a preview of some end states of things I've recently built. So this is it. I'm in cloud code. You're in cloud code. So now let me actually before we get started with some of my stuff, a couple of things that are important to know when you're about to get started. OK, number one, you if you don't know what to do, you can literally prompt here.

18:07What do I do, Claude? Like just ask Claude, what do I do? The second thing that's super important is say like switch to planning mode. Planning mode helps you actually think through and break down what you're going to be doing in your project. So think of them as a project manager who's like, you're the hyped up CEO that like is like, I want to do this cool thing. Then someone goes and figures out like, how do you do that and implement it? So that's sort of what clients do in planning mode. It'll help break things down into smaller chunks to figure out the best way to build what you want to build.

18:40So again, it is using both Sonnet and Opus models, which is another thing to highlight is you might want to switch back and forth between those. And the only sort of like pseudo code you'll need to know is like slash model and select which one. But if you go to planning mode, it'll optimize and delete Opus, and then you'll switch back to Sonnet because your token will get eaten quickly. And so this is just like a little thing to know. There's a bunch of other, as you can see, other things you'll be able to learn. I don't want to overwhelm people. What's a working directory? Working directory, that's basically like your file folder.

19:19That's a directory like as a folder. So like when you're, again, you know, in code language, it would be like seeing and then, you know, making a directory, M-K-D-I-R, like other little things you would say or type in. I can build an agent? Yeah, you can build an agent in here. Like you can build like anything in here. Like it's really that awesome. So people who maybe were like, I tried NAN or I tried make.com and it just seemed too much like that. And maybe like lovable, maybe replet were just a bit more accessible, but then you would get stuck. This is basically that level of easiness from the replet and lovable type thing, but add the full scope of functionality that like an NAN and make would actually fulfill.

20:01So it is really the best of everything and it does it all in code and it makes it just easy because you're speaking, you're typing or speaking to it in English or your native language and it is building the things. All right. So I just asked a question. I said, I would like to create an agent, but I don't know what type of agent that I want. Can you help me think through what agent I should start with? Cool. I hit enter and then I could see it thinking. And then I saw this tokens. It was like tokens being used. when I went in the beginning and I had to select the$5. Yes. It's you. So I'm now paying on top of what I already pay in a membership or a subscription to use this.

20:43So I'm using the max plan where I don't pay attention. But theoretically, yes. If they're tracking my tokens. Yes. Yes. Okay. Yeah. So that's like something you could maybe build to start is like track my tokens and my usage just to make sure. because then as if you're not on the bigger plan, you're going to want to start optimizing for token usage. Like that will, they'll get eaten quickly. There are some tips and tricks to address that. So you don't do that. Number one, you want to be using Sonnet instead of Opus. Opus for planning, Sonnet for building. And then number two, you're going to want to clear your chats because what's happening right now is it's taking in the full context, text meaning every your whole history of your chat every time and then it'll start taking in all your files so when you start clearing the chats and you start building like a new feature you want to make sure you're sort of summarizing things so that there's enough context to be added in where it's not taking in everything from before because when you're putting in everything you're going to be eating a ton of tokens one thing and then i want to then i want you to show me your projects do i close this or do i just always keep this open okay so you want to keep this open.

21:51This is the other thing. Like I've started building my own little tool for like tracking things because when you close this, it's gone. The chat's gone, but your files will remain. So your project and your files will remain there. So you can still access your project and you'll just have to tell Claude to go access the full. And then, but if you don't have a ton of like history or context, there might be things that are missed. The second thing is like, if you're like it made a change and you didn't like the change, like there's not an easy way to go backwards. So you're going to want to use GitHub.

22:24And this is where you're going to have to set up GitHub. And that's a little bit of, you know, a process. Maybe we could do that next week. I don't know for right now. It will share a little bit of time, but that is something that's super important to do. Isn't it called a markdown as you're like writing code? So you save different versions or no? Okay. So this is, that's a.md file is what you're talking about markdown. So you're going want to create something called claw.md which it now sort of does automatically it'll create this file which is essentially your rules file that's like your master thing to say okay you can tell it okay every time i make a change add it to my update github and i actually think it is doing that relatively automatically or very frequently by on its own okay you're putting the rules there like things you want to remember yeah well we'll look at github later i want to now stop sharing my screen and I'm going to let you do your thing.

23:17Yes. If I did close it, where would, how would I go reopen it? Okay. So you would again, go open terminal and literally blank screen, Claude, and it'll pop up. You'll see that little like sweet. Yeah. It'll show you and then say, what do you want to do with basically? And then you just pop in and then you'll just say, go to folder X, Y, and Z. Yeah. Yeah. Okay. Hey, I don't know if you remember this, but when we started this podcast, we entered into a social contract. I would spend time, energy, and money producing this podcast, interviewing these individuals and giving you insights into how to build, buy, start, grow your business.

23:52And you would like subscribe and leave me five-star review. Now out of that, we both get to talk to really cool people and hear really cool insights. We both get a ton of value, but I just want to help you keep your word. So would you do me a favor? Will you go leave a five-star review for me on Apple or Spotify. It would really help. And if you want, even share this with a friend. All right. So let me share my screen. So this is the end state of some projects I was recently working on. So basically what I started with and how I got here. Well, before you do that, will you just describe what I'm looking at?

24:29Okay. So this, this one right here is a data cleanup tool. This is like, it's very ugly, but it's functional, a data analysis tool. And this is basically the output of Claude doing the full analysis for me. And I had it in a couple of different ways. So where am I? So what am I looking at? Am I looking at like a, I'm so stupid. I'm sorry. A word doc. Is this like an HTML file? What is this? I can't see for some reason when I share my screen the – here, let me see if I cop it out if it shows. No, because then it – Did you open that up from the files that were saved locally on your computer? Yes.

25:07So this is something called – you can't see it, unfortunately, but basically you have something called localhost colon 8000. Rather than having a URL to a website up in my browser in Chrome, I have localhost colon 8000. So when you double click, you open that file, it opens in Chrome, the local host 8 ,000? When I was in Claude, I said, can you provide like an HTML site for me to see this tool? It provided me literally this address, this local host address. So then I copy and pasted it and put it in my browser. And I think I even just was able to click on it from Claude and it popped up. So if you don't know where to locate your thing, ask Claude, say, where do I find this thing you just built for me?

25:55Or how can I view it in X, Y, and Z way? How can I see it in a browser? How can I see it in some manner in a CSV file? It will tell you where to go. So if you don't know what to do again. So this is basically the end of a project, backing up a sec with some context of what this was about. I had an Excel spreadsheet from scraping. Well, actually, first off, I was trying to get an analysis on what geographic locations had certain jobs that there were a lot of job openings, basically to see where there might be gaps in supply and demand for certain types of jobs. So what I did was I had initially built out an NAN scraper that would use Appify and essentially scraped Indeed and a bunch of other websites.

26:42That took forever because the tool I was using was just like throwing in one by one URL, scraping it, it would take some time, blah, blah, blah. Then I ended up with a CSV file that was 300 ,000 rows by about 30 different columns. So that was a ton of data. Then I started adding formulas into that and it just got huge and heavy and I started crashing Excel. I then, it was a it was very text heavy. So I was like, I need to classify the data and I needed to extract some information from that text. So I was like, I cannot do this by hand. I can't do this with any of my like Excel skills because it would just either be take too much time.