The Rise of Vibecoding
15:15 Share

Today on the AI Daily Brief, it's a vibe-coding world and we are all just living in it. The AI Daily Brief is a daily podcast and video about the most important news and discussions in AI. To join the conversation, follow the Discord link in our show notes.

This is the weekend, which means, of course, we are doing a long reads episode. And today we are talking about vibe coding, vibe everything, really. And to give you a little bit of background and context, since this is a term that you're hearing more on this show and probably other places as well, let's go back to the origins of this term, which somehow, unbelievably, is actually only about 40 days ago.

On February 2nd, Andrej Karpathy, who was of course on the founding team of OpenAI, posted on X, there's a new kind of coding I call vibe coding, where you fully give in to the vibes, embrace exponentials, and forget that the code even exists. It's possible because the LLMs, e.g. Cursor Composer with Sonnet, are getting too good. Also, I just talked to Composer with Super Whisper so I barely even touched the keyboard. I

I ask for the dumbest things like decrease the padding on the sidebar by half because I'm too lazy to find it. I accept all always, I don't read the diffs anymore. When I get error messages, I just copy-paste them in with no comment, usually that fixes it. The code grows beyond my usual comprehension. I'd have to really read through it for a while. Sometimes the LLMs can't fix a bug, so I just work around it or ask for random changes until it goes away. It's not too bad for throwaway weekend projects, but still quite amusing. I'm building a project or web app.

But it's not really coding. I just see stuff, say stuff, run stuff, and copy-paste stuff, and it mostly works. And just to be clear, there are a lot of people vibe coding now. It is not just Andre. As he calls out, part of the reason for this is the availability of new tools like Bolt, which grew to 20 million users in just a couple of months. And then there's Lovable, whose growth chart is so steep, I have to scroll down right now to show you what it looks like.

Founder Antoine Osico writes earlier this week, lovable growth increased by almost 50 percent last week, now adding 1500 customers per day.

On February 18th, AI creator Riley Brown showed a Google search comparison between the term prompt engineering and the term vibe coding, saying that he would check in on this in one year. Now, clearly what Riley was arguing is that there was going to be more vibe coding one year out than there was prompt engineering. Pretty bold claim, given how much prompt engineering has entered the lexicon. On March 11th, he came back to point out that it had taken only three weeks and that vibe coding was exploding and actually about to surpass prompt engineering when it came to Google search interest.

So this is the setup for this week's long read, which comes once again from Professor Ethan Mollick's One Useful Thing blog. The piece came out earlier this week and is called Speaking Things Into Existence, Expertise in a Vibe-Filled World of Work. I'm going to read a number of excerpts from it, and today because I'm on video it will actually be me reading them, and then we'll do a little bit of conversation. Professor Mollick starts by reflecting on Karpathy's coining of the term vibe coding, but then says, I think the implications of this approach are much wider than coding, but I want to start by doing some vibe coding myself.

From there, he points out that depending on what interface you're using, there can still be some trickiness. He writes, I decided to give it a try using Anthropic's new Cloud Code agent, which gives the Cloud Sonnet 3.7 LLM the ability to manipulate files on your computer and use the internet. Actually, I needed AI help before I could even use Cloud Code. I can only code in a few very specific programming languages and have no experience at all with Linux machines. Yet Cloud Code only runs in Linux. Fortunately, Cloud told me how to handle my problems, so after some vibe troubleshooting, I was set up to Cloud Code.

The very first thing I typed into Cloud Code was make a 3D game where I can place buildings of various designs and then drive through the town I create. I got a working application about four minutes later, with no further input from me. And Ethan shared a video of what came out. He continues, it was pretty neat but a little boring, so I wrote, hmm, it's all a little boring. Also, sometimes the larger buildings don't place properly. Maybe I control a fire truck and I need to put out fires in buildings? We could add traffic and stuff.

A couple minutes later, he writes, it made my car into a fire truck, added traffic, and made it so houses burst into flames. Now we're getting somewhere. But there were still things to fix. I gave Claude feedback. Looking better, but the fire truck changes appearance when moving. Wheels suddenly appear, and there is no issue with traffic or any challenge. Also, fires don't spread and everything looks very 1980s. Make it all so much better. After seeing the results, I gave it a fourth and final command as a series of three questions.

Can I reset the board? Can you make the buildings look more real? Can you add in a rival helicopter that's trying to extinguish fires before me? Andy says, pointing to the video, it's a working if blocky game, but one that includes all day and night cycles, light reflections, missions, and a computer controlled rival. All created using the hottest of all programming languages, English. Now, Professor Malik says the one thing that he left out in the story is that at some point between the third and the fourth prompts, something went wrong and he had no idea how to fix it.

It took a bunch of back and forth with the AI itself to figure out how to fix it. That did end up driving up his API fees. Ethan writes, He continues,

Expertise clearly still matters in a world of creating things with words. After all, you have to know what you want to create, be able to judge whether the results are good or bad, and give appropriate feedback. But applying expertise need not involve a lot of work. Take, for example, he says, my recent experience with Manus, the new AI agent out of China. It basically uses Claude, but gives the AI access to a wide range of tools, including the ability to do web research, code, create documents and websites, and more.

It's the most capable general purpose agent I've seen so far, but like other general agents, it still makes errors and mistakes. Despite that, it can accomplish some pretty impressive things. He then shares an example of what Manus produced when he asked it to, quote, create an interactive course on elevator pitching using the best academic advice. He writes, you can see the system set up a checklist of tasks and then go through them, doing web research before building the pages.

As someone who teaches entrepreneurship, I would say that the output it created was surface-level impressive. It was an entire course that covered much of the basics of pitching and without obvious errors. Yet, I could also instantly see that it was too text-heavy and did not include opportunities for knowledge checks or interactive exercises. I gave the AI a second prompt, add interactive experiences directly into course material and links to high-quality videos. Even though this was the bare minimum feedback, it was enough to improve the course considerably.

If I were going to deploy the course, I would push the AI further and curate the results much more. But it's impressive to see how far you can get with just a little guidance. But there are other modes of vibe work as well. While course creation demonstrates AI's ability to handle casual, structured creative work with minimal guidance, research represents a more complex challenge requiring deeper expertise integration. This all leads to Ethan's next section, vibe working. Today's episode is brought to you by Vanta. Trust isn't just earned, it's demanded.

Whether you're a startup founder navigating your first audit or a seasoned security professional scaling your GRC program, proving your commitment to security has never been more critical or more complex. That's where Vanta comes in. Businesses use Vanta to establish trust by automating compliance needs across over 35 frameworks like SOC 2 and ISO 27001. Centralized security workflows complete questionnaires up to 5x faster and proactively manage vendor risk.

Vanta can help you start or scale up your security program by connecting you with auditors and experts to conduct your audit and set up your security program quickly. Plus, with automation and AI throughout the platform, Vanta gives you time back so you can focus on building your company. Join over 9,000 global companies like Atlassian, Quora, and Factory who use Vanta to manage risk and prove security in real time.

There is a massive shift taking place right now, from using AI to help you do your work,

to deploying AI agents to just do your work for you. Of course, in that shift, there is a ton of complication. First of all, of these seemingly thousands of agents out there, which are actually ready for primetime? Which can do what they promise? And beyond even that, which of these agents will actually fit in my workflows? What can integrate with the way that we do business right now? These are the questions at the heart of the super intelligent agent readiness audit.

We've built a voice agent that can scale across your entire team, mapping your processes, better understanding your business, figuring out where you are with AI and agents right now in order to provide recommendations that actually fit you and your company.

Our proprietary agent consulting engine and agent capabilities knowledge base will leave you with action plans, recommendations, and specific follow-ups that will help you make your next steps into the world of a new agentic workforce. To learn more about Super's agent readiness audit, email agent at bsuper.ai or just email me directly, nlw at bsuper.ai, and let's get you set up with the most disruptive technology of our lifetimes.

It is at the cutting edge of expertise, he writes, where AI gets to be most interesting to use. Unfortunately for everyone writing about this sort of work, they are also the use cases that are hardest to explain.

I have a large, anonymized set of data about crowdfunding efforts that I collected nearly a decade ago, but never got a chance to use for any research purposes. The data is very complex, a huge Excel file, a codebook that explains what the various parts of the Excel file mean, and a data dictionary that details each entry in the Excel file. Working on the data involved frequent cross-referencing through these files, and is especially tedious if you haven't been working with the data in a long time.

I was curious how far I could get in writing a new research paper using this old data with the help of AI. I started by getting an OpenAI deep research report on the latest literature on how organizations could impact crowdfunding. I was able to check the report over based on my knowledge. I knew that it would not include all the latest articles, because deep research cannot access paid academic content, but its conclusions were solid and would be useful to the AI when considering which topics might be worth exploring.

So, I pasted in the report and the three files into the secure version of ChatGPT provided by my university and worked with multiple models to generate hypotheses. The AI suggested multiple potential directions, but I needed to filter them based on what would actually contribute meaningfully to the field. A judgment call requiring years of experience with the relevant research. Then, I worked back and forth with the models to test the hypothesis and confirm that our findings were correct. The AI handled the complexity of the data analysis and made a lot of suggestions, but

while I offered overall guidance and direction about what to do next. At several points, the AI proposed statistically valid approaches that I, with my knowledge of the data, knew would not be appropriate. Together, we worked through the hypothesis to generate fairly robust findings. Then I gave all of the previous output to O1Pro and asked it to write a paper, offering a few suggestions along the way. It's far from a blockbuster, but it would make a solid contribution to the state of knowledge. More interestingly, it took less than an hour to create, as compared to weeks of thinking, planning, writing, coding, and iteration.

Even if I had to spend an hour checking the work, it would still result in massive time savings. I never had to write a line of code, but only because I knew enough to check the results and confirm that everything made sense. I worked in plain English, shaving dozens of hours of work that I could not have done anywhere near as quickly without the AI. But there were many places where the AI did not yet have the instincts to solve problems properly. The AI is far from being able to work alone. Humans still provide both vibe and work in the world of vibe work. This leads to his conclusion, work is changing.

Work is changing, he writes, and we're only beginning to understand how. What's clear from these experiments is that the relationship between human expertise and AI capabilities isn't fixed. Sometimes I found myself acting as a creative director, other times as a troubleshooter, and yet other times as a domain expert validating results. It was my complex expertise or lack thereof that determined the quality of the output.

The current moment feels transitional. These tools aren't yet reliable enough to work completely autonomously, but they're capable enough to dramatically amplify what we can accomplish. The $8 debugging session for my game reminds me that the gaps in AI capabilities still matter, and knowing where those gaps are becomes its own form of expertise. Perhaps more intriguing is how quickly this landscape is changing. The research paper that took me an hour with AI assistants would have been impossible at this speed just 18 months ago.

Rather than reaching definitive conclusions about how AI will transform work, I find myself collecting observations about a moving target. What seems consistent is that, for now, the greatest value comes not from surrendering control entirely to AI or clinging to entirely human workflows, but from finding the right points of collaboration for each specific task, a skill we're all still learning. All right, so another great piece from Ethan here. Appreciate you writing it. Where I want to jump in and just add a little bit is actually at this conclusion. On the one hand, I think...

Professor Malik here is completely correct. It is my experience, and I think most people's experience, that the key thing to figure out is how to collaborate with AI. But that at this point, giving it all the control or giving it none of the control, neither extreme is correct. And yet, at the

At the same time, I think that this point is leading people to incorrect assumptions about the future, or rather, is making it hard for them to imagine a future that is almost certainly coming down the pipeline, where in many cases it will simply make more sense to surrender control entirely to AI. It is so hard for us to imagine trajectories and patterns. There are so many contingent factors that will shape exactly what the role of agents will be,

But I sometimes feel that we're clinging to these ideas, that AI won't replace us, but a human using AI will replace us, which by extension means if we just figure out how to use AI to do our work better or more efficiently, we'll be set. I'm working on a piece now that I haven't published yet that I'm tentatively calling Yes, AI is Going to Take Your Job. And the point is not that no one will have any jobs anymore.

It's that when it comes to knowledge work, at least, I think that functionally, all of the tasks that we do, at least ones that aren't about taste, choice, consideration, and planning, although even that last part is up for debate, are, yes, likely to be done by AI in the future.

What our jobs mean will shift so radically that if a person was in a coma from the period before Chachapiti to the period when this is instantiated in five years or whenever it is, would not recognize the roles of people who have the exact same titles in each of those two periods and

as doing the quote unquote same thing. Then of course, the question becomes what to do about it. The only thing that makes sense to me at this point is to lean all the way in, to try to stretch the frontiers of absolutely everything that you can do as far as is possible with the help of or the replacement of AI or agents.

It is only on the margins and on the edges, even the edges that don't quite work yet, that we're going to be able to glimpse what's coming down the pipeline. And to the extent that you want to be first in line for helping shape whatever it is that your role actually does in the future, by being the first person to do it like that, you have to lean out all the way. Anyways, this is a subject that we will continue to come back to. But for now, big thanks once again to Professor Ethan Malek for this piece. And to the rest of you, appreciate you listening as always. Until next time, peace.