All the Cool Things People are Vibe Coding

0:00Today on the AI Daily Brief, all the interesting ways that people are vibe coding now, and appropriately on the headlines, a vibe coding firm is the newest AI unicorn. The AI Daily Brief is a daily podcast and video about the most important news and discussions in AI.

0:21All right, friends, happy Friday, quick announcements. Thank you to the sponsor of today's show, Blitzy, Plum, and Super Intelligent. And to get an ad-free version of the show, go to patreon.com slash ai daily brief. Welcome back to the AI Daily Brief headlines edition, all the daily AI news you need in around five minutes. In a surprise to no one who is regularly vibe coding, Lovable has become the latest AI unicorn. The company raised a$200 million Series A at a$1.8 billion valuation. Now, this is a company that has seen immense growth. 2.3 million active users, 180 ,000 paying customers, and growth from 17 million ARR in February to 75 million ARR today.

0:59Now, Lovable achieved this growth while numbering just 45 full-time employees, demonstrating just how much tiny AI-native teams can punch above their weight class. When it comes to what they're thinking about next, it seems like the big goal is to make the jump from casual projects and prototyping to being able to generate full production-ready code that can underpin a full business. CEO Antoine Osika said, Every day, brilliant founders and operators with game-changing ideas hit the same wall. They don't have a developer to realize their vision quickly and easily. Now, it appears as though Lovable is getting closer to this idea of being able to support fully-fledged businesses.

1:33At the beginning of the month, Antoine tweeted, I'm officially an angel investor in a software startup built with Lovable. The team is cracked and the product is impressive. They're going to go far. He also recently talked about a Lovable built app that made$3 million in 48 hours. Antoine writes, The team behind it was already the largest edtech company in Brazil with 500 ,000 paying users. They used Lovable to build a premium version of their education platform in two weeks. We're going to talk a lot more in the main episode about all the different ways people are using vibe coding tools and the expanded aperture of what they're actually doing with them.

2:06Next up, Hot Deal Summer continues with Perplexity apparently raising an extension of their Series E round at a fresh$18 billion valuation. Sources said the deal would bring in another$100 million to add to the half billion already raised in May. You could kind of chart AI's continued rise based on Perplexity's funding rounds. They entered 2024 valued at around a half billion dollars, but added multiples across three funding rounds that year. This year, they've continued to jump, which is a reflection, I think, of both their relentless push to release new features such as their Comet browser, as well as the scarcity of cap table space at the leading AI companies.

2:41Now, speaking of Comet and web browsers, Perplexity CEO Aravind Srinivas had some interesting comments about competing with Google in a Reddit AMA about Comet that came earlier this week. He lauded them for keeping Chromium open source and also talked about how, yes, they had with their Project Mariner, some features that were similar to Comet. He said he expected them to pay close attention and to copy and adopt stuff. However, he pointed out that they face, in a way that perplexity doesn't, an innovator's dilemma. He writes, Google is a giant bureaucratic organization with too many decision makers and disjointed teams.

3:13And they have business model constraints on letting agents do the clicks and work for you, while continuing to charge advertisers enormous money to keep bidding for clicks and conversions. At some point, they need to embrace one path and suffer in order to come out stronger, rather than hedging and playing both ways. And while, of course, Aravind has an agenda in saying this, it is true that one of the structural advantages that startups have over incumbents in this AI battle is that they are not constrained by legacy business models. Now, so far, frankly, I think Google has done a pretty good job of threading this line.

3:44But the fact that they have to thread the line is a challenge in and of itself. Next up, we move over to another big theme of the summer, which is the AI talent wars. Zuck is apparently not done hollowing out Apple's foundation model team, as Bloomberg's Mark Gurman reports that Mark Lee and Tom Gunter have joined to work under their old boss, Ruming Pang, at Meta's new superintelligence group. Gurman mentioned that Apple is now offering pay increases to around 100 engineers as a retention incentive while they figure out their restructuring and acquisition plans, but he also noted that the retention bonuses fall far short of competing with Meta's offers.

4:17In fact, sources say that Gunter received one of those fabled$100 million packages. Now, as the market gets increasingly convinced that Apple is going to have to acquire their way out of this problem, one of the targets that has recently been reported is Mistral. Now, Mistral is one of the only startups left from the 2023 era that is still working to keep up with OpenAI and Anthropic in consumer-facing chatbots. And in the latest version of their LeChat product, they've added a deep research mode, voice model, native multilingual reasoning and image generation. And while on the one hand, those are just keeping pace and feature parity with tools, Mistral is clearly trying to carve out some separate space.

4:56First of all, it's one of the only companies left that's actively still pushing open models, which could be even more true as Meta is rumored to be pivoting away from open source. But also, they've strategically positioned themselves as a European giant. In fact, with this release, they're saying that they're the only company that's able to offer a high-performance chatbot that actually complies with European laws. I don't know, ultimately, if that's enough to continue to compete, but in a world of giants, they continue to hold their own. Lastly today, some scuttlebutt around Claude code. Users of the tool are complaining of unexpected usage limits.

5:28The platform's GitHub page has been flooded with bug reports as users try to figure out why they've been cut off. They're being told Claude usage limit reached and given a reset time, but no explicit announcement of a change in limits. One user wrote, Your tracking of usage limits has changed and is no longer accurate. There is no way in the 30 minutes of a few requests I've hit the 900 messages. An Anthropic representative said that they were aware of the issue and working to resolve it, but didn't speak to any policy changes. Now, the change demonstrates how critical AI coding has become for many developers.

5:57One user told TechCrunch, It just stopped the ability to make progress. I tried Gemini and Kimi, but there's really nothing else that's competitive with the capability set of Cloud Code right now. Issues seem concentrated on heavy users who have paid up for access to the premium$200 a month max plan. Anthropic's pricing system has tiered limits that are supposed to be 20 times higher than the pro plan, but the company has never guaranteed a certain level of access. Many users were never even aware there were usage limits associated with the max plan until this week. Frankly, Cloud Code might be a victim of its own success.

6:27Anthropic has reported 300 % growth in the product's user base over the past two months. Adam Wolf, who manages the Cloud Code team, told VentureBeat, we've seen five and a half times revenue growth since we launched the Cloud4 models in May. That gives you a sense of the delusion demand we're seeing. And however that gets resolved, I think that this challenge between cost and usage is going to do nothing but continue as AI-powered coding becomes more and more normalized. Speaking of which, that is the subject of our main episode. And now it is time to close the headlines and jump to that. This episode is brought to you by Blitzy.

7:01Now, I talked to a lot of technical and business leaders who are eager to implement cutting-edge AI. But instead of building competitive moats, their best engineers are stuck modernizing ancient codebases or updating frameworks just to keep the lights on. These projects, like migrating Java 17 to Java 21, often mean staffing a team for a year or more. And sure, copilots help, but we all know they hit context limits fast, especially on large legacy systems. Blitzy flips the script. Instead of engineers doing 80 % of the work, Blitzy's autonomous platform handles the heavy lifting, processing millions of lines of code and making 80 % of the required changes automatically.

7:35One major financial firm used Blitzy to modernize a 20 million line Java code base in just three and a half months, cutting 30 ,000 engineering hours and accelerating their entire roadmap. Email jack at blitzy.com with modernize in the subject line for prioritized onboarding. Visit blitzy.com today before your competitors do. Today's episode is brought to you by Plum. You put in the hours, testing the prompts, refining JSON, and wrangling nodes on the canvas. Now it's time to get paid for it. Plum is the only platform designed for technical creators who want to productize their AI workflows. With Plum, you can build, share, and monetize your flows without giving away your prompts or configuration.

8:14When you're ready to make improvements, you can push updates to your subscribers with a single click. Launch your first paid workflow at useplum.com. That's Plum with a B, and start scaling your impact. Today's episode is brought to you by superintelligent, specifically agent readiness audits. Everyone is trying to figure out what agent use cases are going to be most impactful for their business, and the agent readiness audit is the fastest and best way to do that. We use voice agents to interview your leadership and team and process all of that information to provide an agent readiness score, a set of insights around that score, and a set of highly actionable recommendations on both organizational gaps and high-value agent use cases that you should pursue.

8:56Once you've figured out the right use cases, you can use our marketplace to find the right vendors and partners. And what it all adds up to is a faster, better agent strategy. Check it out at bsuper.ai or email agents at bsuper.ai to learn more. Welcome back to the AI Daily Brief. Today, we are once again talking about vibe coding, one of the most important phenomena of 2025, certainly an early contender for AI term of the year 2025. And we're going to talk about a bunch of different aspects of vibe coding today, including what some people are using it for, which is getting way more diverse. But we're going to start with a new study from the information that was trying to dig into not only what tools were people using, but also how effective they were finding them.

9:40Now, this is not some big, comprehensive, professional study. It is the information subscriber survey. Given that, it's going to be a highly enfranchised, tech-oriented audience. So that is, of course, going to color some of these answers. I would not expect the numbers to be the same if you asked a general panel of American adults, for example. Still, the core question of the study, which is very simple but also a really important one, is are you satisfied with your vibe coding experience? Now, this matters because a lot of the discourse around vibe coding is naturally about its limitations. It's quite obviously powerful to have AI that can code, but we're also trying to figure out how that's going to interact with other types of human expertise, what sort of systems you have to build around VibeCoding to actually get things to production.

10:26Startups are trying to figure out what gaps in the user experience there are. And so the simple question I actually think is very valuable. And the short answer is that people are having a great experience with this. 10 % said that they weren't so satisfied, and 2 % said that they weren't satisfied at all. But 53 % said that they were somewhat satisfied and 35 % said that they were extremely satisfied. Now, even in a heavy tech cohort, that means that in total, 88 % of those who were vibe coding said that they were satisfied with the results, at least somewhat. That's remarkable. What's more, a huge preponderance of the respondents were vibe coding.

11:0675 % of subscribers were using these tools. Now, again, yes, this is a tech audience, but still 75 % penetration of a thing that didn't exist at the beginning of the year, at least not really, is enormous. What's more, if the fact that this is a tech audience maybe makes the 75 % penetration number a little bit less impressive, one could argue that it makes the 88 % satisfaction number more impressive in the sense that if you gave an average person who didn't really interact with technology very much, or at least didn't interact with it professionally, these tools, They're going to seem at first glance much more impressive than it would to people who have much more experience with coding specifically and technology in general.

11:46So the fact that you have this incredibly high satisfaction rate is really telling and very bullish for these tools. It also suggests, I think, not surprisingly, that we are in the midst of a permanent shift in how code is written. These are not the numbers that you get around a passing fad. And so with that in mind, I think it's interesting to ask, what are people actually vibe coding? You might remember back at the beginning, there were a couple of high-profile games from people like Peter Levels that got folks really excited. But what are some of the other use cases out there? I think a great way to understand how a technology is evolving is to see how the use cases are evolving.

12:21Now, I've talked about how our team at Super is already soft-banned from making feature requests. Instead, everyone has to vibe-code prototypes, think through the details, and use the actual output to communicate what they're thinking about more fully. But that use case has been firmly established for months, and it doesn't really take advantage of model improvements. What I mean is that it doesn't matter if a prototype is buggy, it's just throwaway code that's meant to communicate an idea. And indeed, what we're seeing now is absolutely a move from that sort of prototyping use case to actually using production applications that are both public-facing, but also just internal-facing and meant to help you improve your work.

12:58John Park Hill, the director of machine learning at Tere Therapeutics, showed off a custom-built frontend for his data. To give you a sense if you are not watching, it looks like a catalog of different drug molecules and a widget for checking how they react. Hill wrote, The craziest thing about vibe coding is that you now sometimes make a UI just to make sure your underlying data abstractions are working right. Presently, I'm writing a fully expressive logical enumeration engine I need to make an ML dataset. So useful. Now again, just to double-click on this point, this is the type of thing that a year ago, if John had coding skills, he would have had to spend a serious amount of time away from his main work doing, or go hire a team of freelance developers for a couple of thousand bucks, and after a bunch of back and forth, he might have had something functional after a week or two.

13:43Now something like this can be done in a couple of afternoons. And what's important about this particular example, I think, is that it demonstrates that a whole new world of custom software is opening up. There are likely no more than a couple of hundred or maybe at most a couple of thousand people that would need this sort of application spun up, meaning there's just not the incentive there for it to become commercial software. But now it could exist for very little money and basically on demand. Joe Butler, a staff member at OpenAI, shared another example. You remember that week off that OpenAI gave to try to get their people all chilled out a little bit?

14:16Well, Joe writes, OpenAI gave me a week off, so naturally I vibe-coded and shipped a new app. The tweet thread continues,

14:28Now he went on to describe a simple tool that enhances photos for Airbnb hosts, a pain point he was experiencing at the time. Now this came out of a real experience. Joe writes,

14:50that fake AI look. Challenge accepted. Now, by way of background, Joe writes that the last production app he built was back in 2018, and the stack back then was, in his words, a janky combo of Angular front-end and PHP back-end. This time around, Joe's stack included the GPT Image 1 API, V0 by Vercel for design, WinSurf, and OpenAI's codex tool. The net of it was this. 10 days, 60 PR merges later, I had a fully functional app built entirely with tech I'd never touched before. Who knew Vibe Coding wasn't just a meme? By the way, if you want to check it out, go to turtledit.com, T-U-R-T-L-E-D-I-T.com.

15:29I think what Joe's story shows is how much Vibe Coding unlocks entrepreneurial creativity. This is a highly speculative product that started with a solution that someone needed for themselves. And not only would the economics of building it for yourself probably not have made sense, certainly if you were thinking about doing something public, the risk is that you have to put down thousands of dollars to hire a dev team or take that time yourself to build an MVP before you know if you have demand. Now you can just push the code. Now, one of the common critiques of Vibe Coding has been that it's only for that use case that I mentioned before of prototyping.

16:01But Marty Markinson points out that prototyping goes a lot farther when it comes to trying out new ideas. Marty writes, Vibe Coding isn't for the final product. It's for validating products in days, not months. Instead of wasting time building production software that never gets used, you can have an idea, build it in hours, and share with friends for feedback. If it takes off, you're going to need to build again from scratch. But increasingly, I'm not sure even that's true. And I certainly don't know that I think it's going to be true for long. Leni Rachitsky of Leni's Podcast and Leni's Newsletter, who I'm sure many of you follow, just about a week ago posted, On LinkedIn X and in my newsletter Slack, I asked you, What's a product or tool you vibe-coded that you actually use regularly in your work and life?

16:44The response was overwhelming. I got over 1 ,000 enthusiastic replies, ranging from a buzzer app that automatically answers apartment deliveries, to a hyper-personalized greeting card generator, to a workplace accomplishments tracker, to a daily newsletter that can help you learn a new language. Now, Lenny shared 50 of his favorite examples, and frankly, these don't feel to me like just throwaway random apps. Morgan Brown built a carb counter, saying, I built CarbScan to help manage my son's diabetes and blood glucose levels with faster carb counting. I used Replit. Has become a daily go-to. Vijay Quadros made an app that was simply about how many layers should I wear today.

17:20Makes me think that he has to be in San Francisco, because that's a question that you really do have to ask every single day. What's interesting about this one is that clearly other people want it as well. He said, I vibe-coded this with Lovable and use it myself every single day. It's grown to 85 ,000 users in total. Now, of course, you could say, couldn't you just use a weather app for this? Vajit's argument was that they had too much clutter when this is all he really cared about when it came to the weather. There were tons of these sort of personal, designed-for-me types of micro-apps. Sue used Bolt to build an app called Flowbound that gives them exercises or games to do when they find themselves procrastinating on something.

17:55It helps them monitor whether they are overwhelmed, confused, hopeless, anxious, distracted, disorganized, et cetera. There was also a pickleball games tracker, a nicotine pouch tracker, tons of parenting and family apps. There's this one very cool one that creates stories by dragging emojis into a pot. It's called Story Pot, and you literally take a bunch of emojis, put them into a little image of a pot, and it cooks up a story based on that. And these really run the gamut from small personal apps to things that people have pushed into full production. Ben Ogren built a chores app for their kids called Chores AI that's now available in the app store.

18:32And of course, there is a ton of work productivity, meeting prep automation using a Zapier agent, a Chrome extension for sharing availability, a personal time trapping app to help with productivity, an inbox focus tool, and a whole bunch more. And the point is that Vibe Coding is really running the gamut from simple, highly personalized, very specific tools that you only think are going to be useful for you to full production apps. In other words, it's not only expanding who gets to participate in mainstream and economically beneficial app creation, it's also radically expanding the surface area of custom personal software as well.

19:07And it feels very much like we are still just getting started with how far you can go with Vibe Coding. A week and a half ago, Greg Eisenberg tweeted, wait, you can actually run your entire business from cursor using AI agents? I had no idea. This cursor expert never leaves his cursor app and gets everything done faster to make more money. So this is a semi-anonymous user known as AmerMXT. His startup is called HumbleLytics and does internet marketing data. Amir wanted to know how far he could scale his company using AI, so he set himself the goal of doing everything without leaving the cursor window.

19:40Now, rather than creating apps for every element of his business, Amir still used common SaaS products for things like accounting and marketing. The shift was that he harnessed MCP connectors to pull data from these SaaS platforms as needed, essentially using Cursor as a general purpose agent across every element of his business software stack. Amir figured out that you can just ask Cursor to pull accounting data to generate a quote, then tell the agent to update your backend once it's finished. Eisenberg commented, What you're talking about is how do you make SaaS programmable? This might look cute, but it's really about saving a lot of time, being more accurate and being more productive.

20:15Amir walked through the same kind of workflow for marketing, sales, and customer service. Now, of course, technically this isn't vibe coding an app, but it's clearly related to the same trend. What Amir did was leverage Cursor's Agenda capabilities and the rapidly growing range of MCP connectors to make custom automations on the fly. What's more, it's not just this guy Amir. Sean Swicks over at Latent Space wrote, By the way, by far the most interesting new pattern we're seeing in Latent Space is that people are using Claude Code and Klein for non-coding tasks, and this is becoming surprisingly effective for things like sales, BI automation, and G Suite, i.e.

20:51read my email, Slack's linear search web, make report, etc., etc. Of course, enabled by MCP, but somehow both ChatGPT and Claude Desktop have not captured this kind of organic integration and white-collar work automation behavior. This has turned into something of a thread where Swix is tracking this. He points to a tutorial from McKay Wrigley on how to use Claude Code for notes and research, and another from Tarek, who's technical staff at Anthropic, who wrote, CloudCode is all you need. When I first joined Anthropic, I was surprised to learn that lots of the team used CloudCode as a general agent, not just for code.

21:21I've since become a convert. I use CloudCode to help with almost all the work I do now. Here's how. And then he goes on to explain it. He says, In CloudCode, everything is a file and it knows how to use your computer like you do. Name your files well and CloudCode will be able to search them like you would. This lets you make custom setups for memory, to-dos, journals, screenshots, and more. He then goes through and talks about journals and to-dos, Apple Notes and iMessage, MCPs, and even gives a bunch of resources to set up your system in a similar way. So really what we have is an explosion of code at the center of everything.

21:53Yes, we are seeing more applications being built, more custom software being built for personal use. We're also starting to see a blurring of the lines between software, agentic automation and features. For example, when Amir is using Cursor to access his accounting software, it's pretty unclear which bucket that falls into. And that seems like we're actually starting to explore what it means to do AI-first computing. Because the AI in agents can code, it's allowing for people to customize their experience around just being able to tell the agent and communicate to the agent what they want to achieve and accomplish.

22:27And while we're not all the way there yet, you're absolutely starting to see the embers of what a completely different computing experience looks like. I can't stress this enough. If you are not vibe coding yet, stop listening to me, get onto one of these platforms, let it rip and see what you can create. For now that that's going to do it for today's AI Daily Brief. Until next time, peace.