Do you guys want to start out by introducing yourselves? Oh, why don't you go first Joanna? I am Joanna Stern. I am the Senior Personal Technology Columnist at the Wall Street Journal. I am Ben Cohen. I am the Science of Success Columnist for the Wall Street Journal. Ben and Joanna aren't just colleagues. They're also friends who frequently text back and forth. Well, I often send Joanna annoying text messages about technology and I started bothering her with texts about Claude Code, asking if she had played around with it, and also asking like, when are you going to write about it? Claude Code is a coding tool made by the AI company andthropic. The tool allows users to create websites, apps, or anything that requires coding just by typing your vision into a chat box. It's a process known as vibe coding. It's coding with your vibes, right? You don't know how to code. You type it into your chatbot, describe the thing you want, and create the thing. Late last year, Claude received a major update. And it worked so well. It started blowing people's minds in the tech world. And a few weeks ago, Ben started playing with it and texting Joanna about the stuff he was making. His first one was he was bragging about his new personal website that he made with Claude Code. And he was like, hey, senior tech columnist, look at this website that I made. Aren't you proud of me? And I think my text back was like, that's really awesome. Welcome to 1995, Ben. Like, that's super cool. It looked old and crappy. No, actually, it looks really good. Everyone should go visit bzcoin.com, is that right? Absolutely. Ben and Joanna have both been spending a lot of time with Claude Code. And to say it's a development, they could have big implications for jobs across many different industries. It makes vibe coding even simpler for people who don't know anything about Code, but also do know everything about Code. I mean, that's the amazing thing about this product, is that people who have spent their lives as software developers and have risen to the top of their field as coders are now increasingly using Claude Code for their code in addition to people who don't know a lick of code like Joanna had me. Welcome to the journal, our show about money, business, and power. I'm Ryan Connickson. It's Wednesday, February 4th. Coming up on the show, why everyone is freaking out about Claude Code. Aka My Cloud, GPUs for Agentec AI, bring AI inference in closer to users everywhere. Get started at akamy.com slash GPU. There once was a woman who lived in a shoe. A size two snug book, what could she do? But that's not where her story ends. Thanks to a little help from her experienced friend, she got her score into much better shape and relocated to a box fresh new place with room to grow and a mortgage to suit. Now, she lives in a space of four bedroom cowboy boots. Better your experience credit score to help get mortgage ready, experience better your score, better your story. So I've been reading a lot about Claude Code and I've been seeing a lot of people who follow AI saying that this could be the biggest thing to happen in AI since chat GPD. Why are people making that comparison? Well, I think since we've had the chat GPD moment, we've seen all of this progression in models. And so I think all these big leaps we've had with AI products, whether it be image generators or video generators, these have all been small unlocks to things that computers can do for us. Right? We're like, wow, holy crap, it made that image. And it looks so much better than the image it made last year. And I think that is what this Claude Code moment was for coding and vibe coding was, wow, this is getting so good that people that are in the tech world are using this fully to write their code. All right, tell me the story of how Claude Code was created. Okay, so the origin story of Claude Code goes back to September 2024, like the ancient days of AI like 18 months ago, with Boris Cherny, who is the father of Claude Code. He began tinkering with this coding tool as a side project, one day not long after he started at Anthropic. Cherny was just an engineer at Anthropic, among the company's rank and file. He created Claude Code as a tool to help him with his own job. It only took him a couple days to build. And when he shared it internally in a company-wide Slack message, he had no idea how big of a deal it would become. It was celebrated with two raised hands emojis in the company wide Slack. But as his colleague started using it, it became pretty clear it had a lot of potential. What he found interesting was that whenever he looked at the screens of like data scientists within Anthropics and not like software developers or coders, they were using Claude Code, even when it wasn't actually all that easy to use, like a primitive clunky version of Claude Code, they were getting value out of it. And then he starts looking around the office and he sees non-technical people on the sales team using Claude Code to analyze their calls and summarize meetings. And he thinks, oh, this is actually interesting. People who are not just pure coders are using this to their advantage too. Anthropic release Claude Code to the public early last year. But it wasn't until a major update last November that it started going viral. I think that timing is really important. It comes out right before everyone goes home for the holidays and is off. And I think that like part of the momentum and how this thing spreads so quickly in Silicon Valley is that people who work inside these companies were just at home and it was the holidays and they were screwing around on their computers like they always do. And suddenly they had time and inclination to play an experiment with Claude Code. I think a whole lot of them were like completely blown away by what they found. If you're not using Claude Code in 2026, you're going to fall behind. I'm switching from cursor to Claude Code today based on all the hype. This is a wild. This is actually wild. Yeah, it definitely is a game changer and I'm loving it so far. In early January, Ben started playing around with it too. And that's when he sent those text messages to Joanna. And as they were chatting, they got an idea for a way to really put Claude Code to the test. And so we were kind of sending screenshots back and forth and we were kind of like we should probably just do a story on this together. And then as I looked at our chat, I was like, we should just make this the story. We should vibe code this conversation about vibe coding and see if we can make the website an interactive column and put it on the Wall Street Journal.com. Joanna and Ben wanted to see if the two of them, two writers with almost no coding experience, could create an interactive article on the Wall Street Journal's website just by using Claude Code. They wanted the web page to include screenshots of their chat messages with buttons that would allow readers to toggle into different formats. So the conversation might initially look like messages on an iPhone, but click a button and the chat box would now look like Aile Instant Messenger from the 1990s. And to make this, all Joanna had to do was describe her idea to Claude Code. I just put in this prompt to Claude Code. I'll read you part. He said, I'm writing an article for the Wall Street Journal with my colleague Ben Cohen. The idea is that Ben and I go back and forth in the story to debate the merits of vibe coding. The whole story is done with little chats back and forth, looking like I message chats with our photos. Please design a web page for this article. And then what happened next? It did that basically? It did that. It designed a Wall Street Journal page, what it thought it should look like. It took some liberties with our logo and everything, but everything that sort of came beyond the top header was honestly 50% of the way there of what we actually published. Ben and Joanna wrote all the words in the article and the chat messages between them. Claude wrote all the code, and it did it really, really fast. And I think my reaction was like, oh my god, not only are we doing this, I can't believe it did this already. I mean, this was like two minutes after we were talking about this. And then eventually we brought in some actual humans who know how to write code of the Wall Street Journal to fix our code. So when we sent this to some of our in-house folks, Brian Witten, who's one of our computational journalists, Audrey Valboina, one of our designers, they had some notes. They weren't as impressed as we were. Coming up, we talked to one of those humans who knows how to write code and contemplate what this all means for people's jobs. When the tax year ends on the 5th of April, valuable tax allowances may be lost simply because people left things too late. Thankfully, Vanguard is here to help you make well-considered decisions, not rushed ones. Their tax year end hub is full of clear guidance, helpful tools and timely reminders to help you understand your allowances and give your investments the best chance to grow. Search Vanguard Investor to learn more when investing your capital is at risk, tax rules apply. Could AI help you do more of what you love? Workday is the next gen ERP powered by AI that actually knows your business. We help you handle the half-dose, so you can focus on the can't wait to do. It's a new workday. Brian Witten is a computational journalist at the Wall Street Journal. Sometimes I have to call it journalism at scale. I'll build a lot of tools for reporters, investigative tools, reporting tools, things to help them, analyze stuff that we can put together. Sometimes I'll build interactive things for readers to click on, which is nice because then I could show my parents something and be like, I built that and they'll understand what I do. Brian is one of the people Joanna and Ben showed their code to. What did you think when Joanna and Ben brought you this code and this little project they've been working on? What was your reaction? I guess first I was just super impressed and glad to see more people messing around with this kind of stuff because these coding agents kind of turn everybody into a developer, which is exciting for me because that means you can be doing more things and all of the responsibility doesn't necessarily fall down to one person when it comes to building tools or building things. But as Brian dug into the code, he saw that not everything was perfect. At first I saw a lot of outdated practices in the code looking like it had been coded in the late 90s. There are significant problems with accessibility. You couldn't really use the keyboard to do things. If you were using a screen reader, you'd have no idea what was going on. The styles needed updating. It would have clashed with stuff on the page. It did stuff that would have kind of made the rest of the page look like a mess if we'd left it in there. Oh and there was a bug. There are a few bugs. So if you were to give it a percentage of it just going from like was it 100% done, 90% done? Where would you give it a grid? It's like 80% it was pretty close. 80% sounds like a lot. Yeah. Here's our colleague Ben again. This is also like a really instructive example. Like code might get you 90% of the way there. But for that 10%, humans are not just valuable but kind of essential. Like this whole thing doesn't happen without that last 10%. Still, the process was a lot more efficient. I have built many of these types of projects year over my years at the Wall Street Journal working with our development team, working with our great graphics and design team. And projects like this can take weeks. If it's a really big project where you've got a lot of deep reporting and you need it to look really perfect. This as we were saying took seconds and two days to get from start of idea to finish. I asked Brian the computational journalist if cloud code made him worried about his job. Not really as someone who uses cloud code all day every day. I feel safe because I'm maybe I'm being short-sighted but it seems like the more I use these things and the more I build with them, there's a limit to how satisfying the result is without a person seriously directing it and adjusting. What it's producing. There's a lot of technical know how when it comes to knowing what to tell it to do. Brian thinks it will have an impact on jobs but it's unclear exactly what that impact will be. It's lowering the bar for coding projects. It allows us to be more ambitious. And so if you think of a company that has maybe one or two engineers and they want to do an ambitious project. Before this you'd need to scale up, staff up, 5, 10, 20 people working on both coding agents. You can get to that level with only one or two people. So I think the optimistic case is you'll have maybe companies with lower headcount with fewer engineers working for them but more companies with engineers building things because they can do it with these coding agents. This is how Boris Cherny at Anthropic is using the tool he created as something to dramatically increase his productivity. A year ago, 10% of his code was coming from Cloud Code. Six months ago, 50% of his code was coming from Cloud Code. Now over the past two months, he has not written a single line of code by hand himself. A hundred percent of his code is coming from Cloud Code. Not only that, Cherny now starts his day by spinning up multiple Cloud Coding agents at once, effectively creating a small team of robot developers. He calls this multi-clotting. It's the idea that you can have these agents doing various tasks for you at the same time like you would have a team. He has a sort of good system that works across all of the devices. And he told us he launches it is like fix this issue or make this new feature. And so these agents, these Cloud Codes are just working on that while he's getting ready for work and while he's commuting to work, he's constantly managing his multi-clot team. His little team of agents. Yeah. There's something crazy and also profound about this, which is that he is no longer a coder. He is a manager of this fleet of robot coders that are working on his behalf. But like in the past, he wouldn't be managing a team of robots. He'd be managing a team of junior coders. So doesn't this mean that this could take out like a whole layer of entry-level jobs for people? This is the biggest question. This is the biggest question right now. All others in the AI industry are wondering the same exact thing. So what's your sense of how disruptive this particular version of the technology is going to be in the workplace? Like do you think people are going to lose their jobs as this starts to become more widely used? People are losing their jobs, first of all. We need to accept that there are certain industries where people are already losing their jobs. It's clear that like there are certain professions and occupations where people are going to be able to take advantage of this. And the implications of that are like very uncertain right now. Like what the productivity implications are, what the workforce, displacement implications are. Like I don't think we really know what they are going to be yet. What we do know is that like there are fields where this is coming if it's not already here. Here's Anthropic CEO, Dario Amade. Talking at a recent Wall Street Journal event. There are whole jobs, whole careers that we built for decades that may not be present. And you know, I think we can deal with it. I think we can adjust to it. But I don't think there's an awareness at all of what is coming here and the magnitude of it. And the thing that is both thrilling and like deeply terrifying as Joanna has written many times over the past few years. This is the worst and dumbest that these AI models are ever going to be. Right? Like what we are playing with now is only going to get better in the days and weeks and especially months and years to come. In the last few days, Anthropic added new features to Cloud Code, including a tool that can review legal contracts and perform other industry specific functions. It's also released tools for finance and customer service. Then on Monday, open AI updated similar tools. In response, investors started dumping shares of companies they worried could be most vulnerable to disruption. And how much did investors lose? Some $300 billion in value. Software names getting hit again today after the sell off yesterday amid these fears of AI disrupting the industry. Piper Sandler downgraded a few stocks. They are Adobe, Freshworks, Vertex. The concern is that this could be an existential crisis for some companies. Why pay for software solutions when you can now build your own far more easily using something like Cloud Code? In one or five or ten years looking back on the development of AI, how big of a moment would you say the release of Cloud Code will be thought to be? I think it goes to one of the quotes that we had from Boris Chirney which is this is a major democratizing moment in coding and in AI development. And this will have been the moment that some company whether it's Google or OpenAI saw as crap. We got to make that thing right now because everyone's going to want to make a new app or a new tool that's going to better themselves. And so this, that is that moment right now. Also, OpenAI had a ChatcheePete moment. Google really had like the Gemini 3, Nanopanana moment a few months ago. This is that moment for Anthropic. Like this is like the breakout moment for one of the top frontier labs that you know at this time next year might very well be one of the most valuable companies in the world. That's all for today. Wednesday, February 4th. The journal is a co-production of Spotify and the Wall Street Journal. Additional reporting in this episode by Ben Dumbit, Xavier Martinez, Bradley Olson and Alexander Asapovic. Thanks for listening. See you tomorrow.
