What Vibe Coding is Turning Into

Today on the AI Daily Brief: What vibe coding is turning into. Before that, in the headlines: Agents get their own credit cards. The AI Daily Brief is a daily podcast and video about the most important news and discussions in AI. Alright friends, quick announcements before we dive in. First of all, thank you to today's sponsors KPMG, Robots in Pencils, AIUC, and BlitzE. To get an ad-free version of the show, go to patreon.com/aidailybrief, or you can subscribe on Apple Podcasts. If you are interested in sponsoring the show, send us a note at sponsors@aidailybrief.ai. And as I mentioned yesterday, the latest fun new little project from the AI Daily Brief, you can find it agentmadness.ai. We obviously have got a lot of builders in this community. Many of you have done AIDB New Year. Now some of you are doing Claw Camp or Enterprise Claw, and I wanted a place to show off what you've built. That is Agent Madness. We're basically taking the NCAA's March Madness, i.e. 64 teams competing, moving down through the bracket to the championships, and we'll do a human voting on AI agent system. So go to agentmadness.ai, share the coolest agent you've built, and next week we will build this bracket and get this thing live. Again, that's agentmadness.ai. They grow up so fast, don't they? It seems like just months ago we were watching agents fumble around, not really successfully doing the things that we asked them to, and now here they are all grown up getting their first credit cards. Yes indeed, both Ramp and Stripe are introducing virtual payment cards designed specifically for AI agents. The Ramp version allows agents to tap into payment options via API, MCP, and CLI. If those initials don't mean anything to you, go check out my episode yesterday about Google Workspace CLI where I go into all of that. In any case, credit card numbers are not exposed through the new workflow, which should improve security, and users can also set spend limits and have real-time visibility into transactions—just in case your Open Claw has expensive tastes. Ahmad Asino tweets, "My ramp agent has already spent $5,435 on Yu-Gi-Oh cards. No regrets." The Stripe version is just getting started with a limited beta available to agent builders. Once again, the virtual cards have programmable spending limits, controls around merchant categories, and real-time risk scores to make sure your agent spending is under control. Stripe product manager Jeff Weinstein explained the logic behind introducing the cards, commenting that we're about to see a lot more agents doing a lot more for us, and that's often going to involve spending money. Now, this is one of those obvious-in-retrospect kind of things. You're already seeing people experimenting with handing a credit card to their Open Claw. Entrepreneur, supermom, and friend of the show Jessie Gennett has recently been going ham on Open Claw experiments, and at the end of February got her Open Claw wired up to using your Amazon account, but overall there is still clearly a need for better ways to both enable and control agent spending. One other interesting little detail about Ramp and Stripe doing this is that they're opting for normal cards rather than promoting some agent-focused payment protocol like X402 or AP2. In short, advances in agent capability mean they are more able to use payment rails that were designed for humans than when those protocols were designed 6 months ago. They also have the huge structural advantage of plugging into existing merchant infrastructure rather than having to bootstrap a new system. Separately, but also from RAMP, the company published a new edition of their AI Index, and economist Era Karazian is ready to call it. "I've seen enough," he writes. "Anthropic is the new default for businesses." The new update for March shows that AI adoption is growing quickly for RAMP customers across the board, but Anthropic's growth is in a league of its own. Overall, AI adoption is now at 47.6%, meaning almost half of the businesses on RAMP now have at least one AI subscription on the books. Anthropic is the only company seeing accelerating growth in adoption, now at 24.4% of ramp customers, compared to 34.4% and falling for OpenAI. The big signal, however, that Karazian is paying attention to is Anthropic's share of the first-time AI spending. Compared head-to-head against OpenAI, Anthropic is now winning 70% of first-time business. The two companies were neck and neck at the beginning of the year. Writes Karazian, it's a complete reversal of the trend we observed in 2025, when OpenAI adoption accelerated faster than any other model company. Now keep in mind this is of course just one source of data from one provider, and there are even limits to this methodology. As Adam from OpenAI points out, this might not be fully inclusive of enterprise customers who purchase via purchase order, but certainly this is part of an overall trendline that we have seen numerous examples of. Speaking of OpenAI, the company is planning a bit of a shift around Sora, planning to integrate the model into ChatGPT. The launch of the standalone Sora app in September did garner a lot of downloads, shooting in fact to the top of the App Store charts. Perception, however, was that usage quickly waned, and last fall Sam Altman told staff that only a small percentage of Sora users were sharing their videos publicly, making it not all that effective as a social network. Now one little bit of a narrative violation, given that many people had written off Sora, is that recent data from SensorTower showed that over the last month Sora jumped over 3 million daily active users, and was continuously growing. And yet still, the information reports that by bringing Sora back into ChatGPT, it signals a strategic shift, refocusing on ChatGPT as the core OpenAI experience. The information points out that OpenAI has put more resources into ChatGPT recently in hopes of hitting 1 billion weekly active users. OpenAI's leadership is also cognizant of increased competition from Google's Gemini, which now does offer video generation as part of the core app experience. Over in Elonland, the unification of the Musk empire continues as Tesla plans a joint AI initiative with xAI. Musk announced on Wednesday that his companies will work together on a project called MacroHard or Digital Optimus. He said the project was part of Tesla's recent deal to invest in xAI. In a post on X, he wrote, Grok is the master conductor and navigator with deep understanding of the world to direct Digital Optimus, which is processing and actioning the past 5 seconds of real-time computer screen video and keyboard mouse actions. GROK is like a much more advanced and sophisticated version of turn-by-turn navigation software. You can think of it as Digital Optimus AI being System 1, instinctive part of the mind, and GROK being System 2, thinking part of the mind. Musk noted the system will run on Tesla's low-cost AI4 silicon, rather than requiring expensive Nvidia hardware. So stripping out the sales pitch, it sounds like this will be xAI's computer use play, perhaps joining the rest of the industry in the open-qualification of everything. However, Elon of course positioned it as a more grandiose vision, adding, "It will be the only real-time smart AI system. In principle, it is capable of emulating the function of entire companies. This is why the program is called MacroHard, a funny reference to Microsoft. No other company can yet do this." According to Business Insider, the MacroHard project has run into multiple roadblocks at xAI since it was first mentioned in August. Sources said that two project leads left the company last month in quick succession, after Musk expressed disappointment in a lack of progress. They also noted that a massive data annotation project attached to MacroHard was paused last month to allow for architectural changes to the underlying model. So despite some negative behind-the-scenes reporting, we'll see if the axiom holds that you should never bet against Elon Musk. Lots of chatter about Ben Affleck's big payday. Netflix could pay up to $600 million for his AI startup Interpositive, which would make it the largest AI media acquisition to date. Now last week's news that Ben Affleck had sold his AI startup called Interpositive was already a big story in Hollywood. It came as a shock to many that Affleck was secretly working on AI for filmmakers given his comments on AI where he basically thought that it couldn't create genuine works of cinematic art. But of course there is both more nuance to what Affleck has always said and more nuance to what Interpositive does. Affleck was never a throw the baby out with the bathwater kind of AI critic. And in fact, in all of the comments that I've seen of his, it was less about what AI couldn't do and more about what humans were uniquely capable of. Interpositive's work is similarly nuanced. The company is working on custom-trained models for individual productions. The technology reportedly was designed to allow filmmakers to quickly edit shots by automatically adjusting lighting, reframing the camera, and replacing backgrounds. It was not, however, designed to generate entirely new scenes, keeping creative control firmly in the hands of the human filmmaker. Essentially, it's an AI that is fully embraced as a cost-cutting tool for post-production rather than as a replacement for human filmmaking. Bloomberg reports that the technology is already being used by David Fincher for an upcoming film with Brad Pitt. Now, it was already an interesting acquisition, but the sky-high price tag puts the deal in a very different context. This would now be among the largest deals that Netflix has ever done. Affleck is also joining Netflix as an advisor as part of the deal, so my guess is that this is not the last we've heard of this. As I've said before on this show, I actually think that the entertainment industry has a unique opportunity to carve a path showing that AI doesn't have to be a choice of humans on the one side or robots on the other. Call me naive, but I'm going to choose to be optimistic that this type of deal can be part of a future where we get more amazing entertainment content enabled by AI that isn't just cost-cutting slop. Lastly today, a massive jump in revenue for Lovable. The company has added a staggering $100 million in annualized revenue in a single month. Lovable told Business Insider that ARR jumped by a third in February, rising from $300 million to $400 million. Coupled with Cursor doubling ARR to $2 billion over the past 3 months, it's clear that we are very much in a rising tide lifts all boats kind of moment. Indeed, Lovable's Chief Revenue Officer Ryan Meadows said, it's a rising tide. We've been super happy with what we're seeing. Lovable also launched their debut brand campaign this week., which features a normal woman going through her day with a song rattling around in her brain. She eventually arrives home, fires up Lovable, and prompts it to build an app for the songs in my head. At no point in the ad anywhere does it mention AI or vibe coding. It is instead a story of the way that these new tools collapse the space between idea and something real. Now, speaking of vibe coding, that is going to be the topic of our main episode, so that is where we will close the headlines. Agentic AI is powering a $3 trillion productivity revolution, and leaders are hitting a real decision point: do you build your own AI agents, buy off the shelf, or borrow by partnering to scale faster? KPMG's latest thought leadership paper, "Agentic AI Untangled: Navigating the Build, Buy, or Borrow Decision," does a great job cutting through the noise with a practical framework to help you choose based on value, risk, and readiness, and how to scale agents with the right trust, governance, and orchestration foundation. Don't lock in the wrong model. You can download the paper right now at www.kpmg.us/navigate. Again, that's www.kpmg.us/navigate. Quick update on something I've been following. AIUC-1 is the first real standard for AI agents, developed with Fortune 500 security leaders to basically define what safe, enterprise-ready AI agents should look like. A little while back, I mentioned that ElevenLabs became certified against AIUC-1. This week, two more big players joined: Finn from Intercom and UiPath. What that certification means in practice is real-time guardrails that block unsafe responses, protection against manipulation, and a full safety stack designed for enterprise environments. And that's why this matters. You've now got leaders across three major AI agent categories—enterprise automation, customer support, and voice—all certifying against the same standard. That starts to look less like a one-off and more like the beginning of a real industry trend. To learn more about the world's first AI agent standard, go to aiuc-1.com. That's aiuc-1.com. Most companies don't struggle with ideas. They struggle with turning them into real AI systems that deliver value. Robots and Pencils is a company built to close that gap. They design and deliver intelligent cloud-native systems powered by generative and agentic AI with focus, speed, and clear outcomes. Robots and Pencils works in small, high-impact pods. Engineers, strategists, designers, and applied AI specialists working together to move from idea to production without unnecessary friction. Powered by RoboWorks, their agentic acceleration platform, teams deliver meaningful results including initial launches in as little as 45 days depending on scope. If your organization is ready to move faster, reduce complexity, and turn AI ambition into real results, Robots and Pencils is built for that moment. Start the conversation at robotsandpencils.com/AIDailyBrief. That's robotsandpencils.com/aidailybrief. Robots and Pencils: Impact at velocity. Want to accelerate enterprise software development velocity by 5x? You need Blitzy, the only autonomous software development platform built for enterprise codebases. Your engineers define the project, a new feature, refactor, or greenfield build. Blitzy agents first ingest and map your entire codebase, then the platform generates a bespoke agent action plan for your team to review and approve. Once approved, Blitzy gets to work autonomously generating hundreds of thousands of lines of validated, end-to-end tested code. More than 80% of the work completed in a single run. Blitzy is not generating code, it's developing software at the speed of compute. Your engineers review, refine, and ship. This is how Fortune 500 companies are compressing multi-month projects into a single sprint, accelerating engineering velocity by 5x. Experience Blitzy firsthand at blitzy.com. That's B-L-I-T-Z-Y.com. Welcome back to the AI Daily Brief. Yesterday we got a set of interesting product announcements that are both worthy of coverage on their own terms, but also I think reflective of some bigger shifts that we are experiencing in this very fast-moving field of vibe coding and AI and agentic building. We of course recently just had the 1-year anniversary of the term vibe coding, so it makes sense that it has completely and entirely changed in just the last 13 months. The first announcement was Perplexity's Computer for Enterprise. This was actually an update on a product that they had announced at the end of February called Perplexity Computer. Perplexity Computer is effectively an attempt to build an AI everything machine. They believe that we are moving beyond the paradigm of chat interfaces or even agentic task interactions to complex systems that in their words, create and execute entire workflows and are capable of running for hours or even months. Like other advanced agentic systems, the human interacting with Perplexity Computer doesn't tell the system what to do. Instead, it describes a goal or an outcome and allows Perplexity Computer to figure out how to do it. It does so by making a plan that breaks what it needs to do into tasks and subtasks, from which it can then spin up agents and subagents for execution against those tasks. Subagents can do anything from research to document generation to data processing to, importantly, interacting with connected services. When it runs into a problem, the promise is that it can just figure it out. Now, when Perplexity announced this, they tried to argue that it was a natural evolution for the product, and certainly to the extent that they had always been a very experimental company, that might be true. However, it certainly felt to me and to many like it was Perplexity being willing to change and evolve and try to skate to where the puck is headed. And as compared to tools from the model labs, one of Perplexity's big pitches is its inherent multi-modelness, The fact that it can interact with Opus, Nano Banana, Gemini, Grok, and ChatGPT all at once. So that came at the end of February, and then on Wednesday of this week, we got Computer for Enterprise. It is pretty much exactly what it sounds like, a platform that can run multi-step workflows across all sorts of different use cases like research, coding, and design, taking advantage of a wide variety of different models from different labs and connecting to a wide array of the core applications that run the enterprise. Perplexity promising that Enterprise has 400+ applications already integrated. One of the big featured integrations was with Slack, so now enterprise users can interact with Perplexity Computer from directly inside of their Slack workspace. Now interestingly, in an accompanying VentureBeat article, the Perplexity team actually revealed that Computer was initially developed as an internal tool and was initially launched as a Slack bot for Perplexity's own employees before they ever considered launching it more broadly. Said Dmitry Shevalenko, Perplexity's head of business, with no hyperbole, the introduction of Computer inside Perplexity was the single biggest productivity unlock in our entire history as a company. There's no other feature we've ever built that has changed how much we work as this one. Now, this week alone, we've talked about the expanded capabilities of Gemini in Google Workspace, as well as Microsoft going deeper with Anthropic to launch a version of Cloud Cowork called Copilot Cowork for M365. And these are the types of productivity suites that Perplexity is now up against. One little subnote from a business model perspective is that Perplexity is charging enterprises on a usage-based model rather than a seat-based model. Administrators can control spending at the per-employee, per-team, or per-company level, But the rationale from the Perplexity team is that the underlying cost structure is so different depending on what type of task you're doing. Generating a video, for example, is going to cost a lot less than generating a text memo. So in that context of variable workloads, where one type of user in the enterprise might look very different from another type of user in the enterprise, the usage-based model starts to make more sense. And that wasn't the only thing Perplexity announced. Perplexity also announced Personal Computer. In the announcement tweet, they write, Personal Computer is an always-on, local merge with Perplexity computer that works for you 24/7. It's personal, secure, and works across your files, apps, and sessions through a continuously running Mac mini. Personal Computer runs in a secure environment and is controllable from any device anywhere. You can run Personal Computer on a Mac desktop computer connected to your local apps and Perplexity's secure servers. In other words, you guessed it, as Matthew Berman put it, they're building OpenClaw. The integration of the Mac Mini even really puts a fine point on the clarity of what they're trying to do here. Now, this is of course something that we've talked about extensively on this show, the OpenClawification of everything, by which we really mean companies racing to launch highly performant agents that have access to your actual files and systems and can control those systems. Unsurprisingly, given that one of the major concerns around OpenClaw is security issues, there are a lot of attempts to pitch these alternatives as the more secure version. But from there, we flip over to Reddit, who just announced Agent 4. A couple of weeks ago, Y Combinator founder Paul Graham tweeted that Replit founder Amjad Massad has showed him something that he thought represented a real paradigm shift. Graham confirmed that Agent 4 is that, adding, it generalizes the idea of vibe coding beyond what people usually think of as coding. So what is Agent 4? Effectively, you can kind of think about it as a complete collaborative surface for individuals, teams, and agents to work together on building a wide variety of things. In some ways, it takes the core spark or genesis of VibeCoding, i.e., explain what you want to build in plain English and have the tool code it up for you, and just expands it to all of the different types of things that you could explain in plain English that you want built. In addition to expanding the set of things you can build, however, it also is changing the form factor of how they get built. First of all, the canvas through which you interact has been updated. Specifically, instead of it being one artifact field where you see what's produced, and one chat bar on the side, you can interact a little bit more directly with that artifact field, both through design tools of the type that you might see in Photoshop, as well as by new types of annotation tools and the ability to precisely edit in natural language, but on some specific part of what's been generated rather than the whole thing at the same time. Another change is that you don't have to do just one thing at a time. You can in fact add and layer on a whole bunch of different things that it's working on at the same time, and rather than queuing them to be done in sequential order, it can do them all at the same time. So imagine that you want to update the fonts, build a new feature, and change the copy. Whereas before, vibe coding tools would have forced you to pick and sequence those things, now you can just do them all at once. With Agent 4, you can also have multiple people working on the same file at once, which of course, as awesome as the superpowers for solopreneurs have been with these tools, a significant portion of the work out there still happens inside of teams. The sum total is a vastly expanded creative canvas. That's not only building websites and web applications, but effectively anything digital that you might want, from sites to slides and beyond. As they are wont to do, Latentspace did a great job of summing up the significance. They write, "This Replit is unrecognizable from the coding with some AI tacked on platform that Replit was just 2 years ago, with a bunch of now veritably antiquated conventional wisdoms of the time. Now that software engineering is approximately solved, where does a coding platform go?" Well, for Replit, it means going up the stack to be a fully integrated productivity suite with a canvas, apps, sites, slides, videos, and others. This is a smart pivot, they continue, that is in line with one of the most dominant themes of 2026. Now that coding agents have solved coding, it is the same coding agent builders that are expanding their scope to more and more knowledge work tasks. So let's take a step back now and try to capture some of the themes here and why these things combined represent, I think, where vibe coding is heading. One theme is blended user experiences where it's no longer chat or traditional input methods. It's extensible canvases that have all of those things all at once. This is both an enabler and a byproduct of that expansion of coding infiltrating other knowledge work outputs. Basically, as you move not only towards new complexity, but just a bigger variety of types of work that the coding agent can complete, you need different types of control surrounding it. Another theme captured by Perplexity Computer, particularly the personal version, is the idea of persistent context. One of the reasons that we are in the open qualification of everything is that people are discovering that giving agents persistent access to your systems is a good way to solve some of the memory problems that have held back what agents can do. Another theme, of course, is multi-agent systems. Replit 4 and Perplexity Computer aren't an agent. They are an entire team of agents that get spun up in a purpose-built way to solve whatever it is the challenge or achieve whatever it is the goal that the prompter has put to them. Another theme is multiplayer mode. Between computer for enterprise integrated into Slack and Replit Agent 4's multiplayer canvas, we're moving away from siloed agent use into integrated team interactive agent use. One small theme, which is admittedly less important than the other ones, is the fact that the Mac mini all on its own has given Apple an entire AI strategy, which I just think is hilarious, especially as this now gets embraced not only by the Open Claws, but by the Open Claw competitors. Going a little more meta, The way that we think about pivoting and changing what a company does is, I think, itself going to change pretty dramatically. It used to be in startup land that pivots were what you did on your way to product-market fit. You tried one idea and if it didn't work, you pivoted into something else. Increasingly, it seems that running a successful company and having the nimbleness to pivot and evolve are going to be one and the same. And while one might argue that companies have always had to evolve to changing times, The speed at which things are changing is dramatically different now. Perplexity is a company that people had counted out about 2 months ago, and now people are super excited again. And guess what? It's not because they made an incrementally better search product. It's because they were comfortable saying, it doesn't really matter ultimately what we built before. This is what we need to be building now. Now, alongside all of this, there still remains so much that is unknown. One other way to look at Perplexity Computer and Replit Agent 4 is that we simply don't know yet what the right form factors and full set of form factors for agentic interaction and agentic orchestration are going to look like. In a different context, Andrej Karpathy was talking about this this week as well. He tweeted, expectation, the age of the IDE, i.e., integrated development environment like Cursor, is over. Reality, we're going to need a bigger IDE. It just looks very different because humans now move upwards and program at a higher level. The basic unit of interest is not one file, but one agent. It's still programming. Now he goes way deeper than that in that conversation, but the point for us is that as much as Perplexity Computer and Replit Agent 4 are giving us glimpses of the future, glimpses is all they are, and we have a lot more to discover. For now, if nothing else, it's two very cool new products for you to get your hands on and try to build more interesting things with. And that is going to do it for today's AI Daily Brief. Appreciate you listening or watching as always, and until next time, Peace.