In addition to bolt.new and Cursor, Lovable from Sweden is a recent AI programming rookie.

After the official launch in November 2024, ARR grew by $1 million almost every week, from 0 to $17 million in 3 months, with more than 30,000 paying users, making it the fastest growing start-up in European history. According to the CEO, Lovable’s current first-month retention rate of paying users is as high as 85%, which has surpassed ChatGPT and is still rising.

Formerly known as the open-source project GPT Engineer, it has more than 50,000 stars on Github and was once shut down by Github because users used it to create 15,000 projects a day.

Lovable’s founder Anton Osika recently sat down with Lenny’s Podcast, a former product team leader at Airbnb and a well-known angel investor, about how Lovable’s team works, how to hire and how to scale quickly with a very small team.

---

01 

The way software is built in the future

It’s talking directly to the AI

Moderator: How do you understand Lovable in the simplest way?

Anton Osika: Lovable is a personal AI software engineer. You describe an idea, and then AI gives you a fully functional product. This means that today’s entrepreneurs can turn their ideas into real business. We have a lot of designers and product managers who use it to create first versions of their products to show to their teams, and some even become founders because it’s so powerful.

Of course, developers themselves can use it to write code or create products faster. Almost all of my friends have asked me for help, “Anton, I want to make a product, how do I find a good software engineer?” We built Lovable for the 99% of people who can’t code. At the moment, if you have a technical background, you will go further. But over time, it’s clear that the way to build software in the future is to talk directly to AI.

Moderator: Can you share some data on the size of your business? Because that’s just an exaggeration.

Anton Osika: We launched Lovable three months ago, and now we have 500,000 monthly active users, of which 30,000 are paid users. And the rate of growth is accelerating, almost entirely by natural transmission.

Moderator: Okay, let me add some more revenue figures, you guys got $4 million in ARR within four weeks of product launch, and $10 million in two months, with a team of just 15 people. You are the fastest-growing startup in Europe. You’ve recently rewritten your entire codebase and haven’t been able to release new features for a while, right?

Anton Osika: That’s right. Some people say, “You’re releasing so fast,” but we were frustrated because we originally wrote the service in a scripting language. As we scaled, we realized we had to rewrite everything and rewrite it in a more efficient way.

Moderator: You mentioned that some companies started their businesses based on Lovable, can you give a few examples?

Anton Osika: An early adopter, Harry, who started delivering real web applications to customers, not just design drafts. Later he said, “Why don’t I start an AI startup.” His company has already launched on Product Hunt and is starting to make money, where users can upload photo galleries and AI will automatically parse and categorize them. If you look at the “Launched on Lovable” app – which is also built with Lovable – it’s like a version of Product Hunt that shows a lot of small SaaS businesses.

---

02 

Just two words,

Generate a clone of Airbnb

Anton Osika: Lenny, have you ever thought about using Lovable to quickly replicate a product?

Moderator: I haven’t thought about it yet.

Anton Osika: How about you try it?

Moderator: Okay, we’re going to build a clone of Airbnb.

Anton Osika: Okay, I just typed in the first prompt, create an Airbnb clone.

Moderator: What is the prompt word?

Anton Osika: Just two words: “Airbnb clone”. The AI gets to work, and it thinks about what a nice Airbnb clone should look like and makes some design decisions. I zoom in and you’ll see a UI with all the elements you’d expect from an Airbnb clone: different categories, two examples of listings, a login button, and more. At the moment it only has a UI and does not have the full functionality of Airbnb. I can now ask for improvements, such as showing different listings when switching categories. Do you have any ideas for adding to that?

Moderator: Under normal circumstances, how long does it take to go from prompt words to generate code?

Anton Osika: The first prompt takes 30 seconds.

Moderator: 30 seconds, okay. I’ve always wanted to explore if I can buy the property I fancie, so how about we try adding a buy feature?

Anton Osika: Okay, let’s add a button. “Add a button to your listing that says ‘Buy this Airbnb listing,’ and click it to pop up a modal box to buy your listing.” Is that okay?

Moderator: Perfect. The website you asked the AI engineer to build is an actual functional website that can be browsed, not just the design draft. Of course, there’s no real listing data right now. If you really want to build an Airbnb clone with real listing data, what’s next?

Anton Osika: Yes, it’s just a mock UI right now, but it’s interactive. If you want to add login functionality and listing management, you’ll need to connect to the backend, where data and user logins are stored.

Let’s take a look at the execution results of the AI after we prompted us to add the purchase function. It says “Book Now”. The AI may be surprised by the “buy a home” because it’s Airbnb, so it’s still a booking.

Moderator: Let me interject that this is a good example of why a good product manager is important. Miscommunication can waste a lot of time, and with people, you have to clearly explain the problem to be solved and why. With AI, you have to be good at describing what you want, but you don’t have to explain the “why”, the AI doesn’t care about motivation, it just needs clear instructions. This is a product manager’s forte.

Anton Osika: For AI, it’s important to accurately describe expectations and point out what’s wrong. I’ll start by showing you the quickest way to fix the error. It generates a “Book Now” button, which I want to change to “Buy Now”. The feature we just released is that you can edit visually just like you would in Square space or Wix. I changed the text to “Buy Now”, and it changed immediately, and the underlying code was updated at the same time, very quickly.

Moderator: This represents the cutting edge of tooling. While other tools only let you request AI modifications again, the direct editing feature you show is a big breakthrough.

Anton Osika: Yes, it’s now 「Buy Now」.

Moderator: This feature was just launched a few days ago?

Anton Osika: It was launched a few days ago. I want to show a fully functional build. We use Supabase, an open-source backend service.

Moderator: Where is the backend actually hosted?

Anton Osika: Everything can be deployed with one click, hosted on Cloudflare and Supabase on the backend.

Moderator: Thank you for the presentation. For a lot of people, it’s too shocking. Building a prototype website in a matter of minutes and cents used to cost tens of thousands of dollars and weeks or even months.

Anton Osika: These tools are already powerful, and they’re progressing very quickly. The bottleneck is that it is not fully integrated into the existing product workflow. But because it’s progressing so fast, people who want to be part of the future economy would do well to try these tools for themselves, and being in the top 10% of users will set you apart for months to years to come.

Moderator: If you could sit next to everyone who is using Lovable for the first time and give them one piece of advice, what would it be?

Anton Osika: Mastering a tool like Lovable requires patience and curiosity. We have a “chat mode” where you can ask, “How does this work?” I didn’t get what I wanted, what was missing? How can it be more efficient?” It’s one of the best ways to learn software engineering, you don’t have to write code, but it’s useful to understand how products are built. Patience and curiosity are super important.

Second, I would suggest that your prompt isn’t clear enough not to say to the AI, “You’re not doing a good job,” and be specific about what you expected, what’s right and what’s wrong.

Moderator: When working with engineers, miscommunication can be costly and can miss features or requirements. And if Lovable isn’t good enough, it can try again in 30 seconds.

Anton Osika: Yes, it can be more expensive to work with people.

Moderator: What do you call it in chat mode?

Anton Osika: Just called Lovable.

Moderator: Why did you choose the name “Lovable”?

Anton Osika: I think the goal is to build a great product, and the best word for it is “Lovable”. I like to use that term as a term: minimum lovable product, cute product, absolutely lovely product. That’s why the company name is called this.

---

03 

Because the amount of use is too large,

Once shut down by GitHub

Moderator: Now you have well over 10 million ARR, right? Do you make this data public?

Anton Osika: We’re not particularly numerical, but we could tweet a “2x increase” soon.

Moderator: I want to go back to the beginning, what is the origin of Lovable? How did this journey begin?

Anton Osika: After the launch of ChatGPT, large language models became adept at generating code based on human instructions. I was the CTO of a YC investment startup at the time, and my team felt that I was exaggerating the impact of AI on programming, saying it wasn’t going to change anything in a year.

I wanted to prove my point, so I made an open-source tool, GPT Engineer, and you typed it with a prompt that said, “Make a snake game,” and it generated the code and ran it. GPT Engineer is now the most popular open-source tool for showcasing applications for creating large language models, with more than 50,000 stars and dozens of academic citations.

Moderator: I’d like to add that GitHub shut you down for a while because it thought it was some kind of attack, and your star growth and usage were exaggerated.

Anton Osika: That’s Lovable. In the early days, Lovable created new Github projects every time someone used them. We asked if there were any restrictions, and Github said no. But when we were creating 15,000 projects a day, their server was so loaded that an engineer on duty might wake up in the middle of the night and shut us down with an email saying we had broken the rules, and we didn’t know what was going on.

Moderator: This reminds me of a story where when ChatGPT was first trained, the Microsoft server thought it was a crawler attack and blocked it.

Anton Osika: I did GPT Engineer and I think it’s a major change in human history. It was manual labor that was replaced by machines, and now cognitive labor is surpassed by machines. How can it make a positive impact? It’s not about making engineers more productive (with Microsoft Copilot), it’s about empowering people who can’t find good developers to bring their ideas to life. I pulled up my former colleague Fabian, who was also the founder, and said we were going to make something like GPT Engineer for people who couldn’t code. That’s the origin.

Moderator: And then it became Lovable, moving from open source to a product that everyone could use and pay for. And then I saw the data, you guys got 1 million ARR per week after you launched Lovable, is that true?

Anton Osika: Yes, the first version was called GPT Engineer App, and we launched it with a waiting list to collect feedback and iterations. When the product is good enough, it is named Lovable, which mainly optimizes AI. We’ve been adding 1 million ARR per week since three months ago and we’ve been growing even faster since then.

Moderator: Faster than 1 million a week! It’s definitely a product-market fit (PMF). You mentioned unlocking the extensibility secret on the backend, the new Scaling Law, can you talk about it?

Anton Osika: There are a lot of rules for scaling AI systems. We’ve found that the more effort we put in, the better the product will continue to be. Usually when AI generates software, it starts out well and gets stuck later. We find this stuck point, quantitatively adjust the system in different ways, and quickly feedback on key areas for improvement.

Moderator: Do you mean that the AI doesn’t know what to do next, or introduces a bug?

Anton Osika: yes, it’s not smart enough to fix that bug.

Moderator: This is a common problem with this kind of tool, it gets stuck at a certain point, and the user doesn’t know what to do if he’s not an engineer.

Anton Osika: It’s definitely going to be this problem right now, but it’s being resolved quickly, and we’re targeting key areas like plus logins, data persistence, Stripe payments, to make sure you don’t get stuck. Now the stuck place needs users to know how to solve it, but in the future it will not be so important, and the problem of AI getting stuck will become less and less.

Moderator: I know you don’t go into technical details, but that’s your advantage, so I won’t force you. Going back to the growth rate, a team of 15 people has 10 million ARR in two months, which is too exaggerated. How do you do that?

Anton Osika: Based on the basic model, we find the right way to present it to the user, design the best human-machine interface, and seamlessly integrate functions such as authentication. People love this product, and that’s a growth driver. We mainly post our progress on social media to let people know about us.

Moderator: This is “public building”. Your presentation was impressive, and the growth numbers were fascinating. But most companies have interesting things to share, what do you think you’ve done uniquely to make the product so “cute”?

Anton Osika: The team is at the core. You need people who can deliver quickly, have taste, and know how to simplify and abstract. Together, we make products better and better.

Moderator: How many Lovables has your team members written with AI?

Anton Osika: We set Lovable to be self-modifying, and we did a lot of things with it; But a lot of the ultra-specific, like booting up a dedicated PC for each user, Lovable can’t do. We use developer tools, and almost everyone writes code with AI anytime, anywhere.

Moderator: Tools like Cursor?

Anton Osika: Yes, almost all of the team uses Cursor.

Moderator: I recently did a survey and 17% of my podcast listeners code in Cursor, which is an exaggeration. Speaking of competitors, people are curious about the difference between you and Bolt and Replit. How do you understand the difference between Lovable in the simplest way?

Anton Osika: We package products for the non-technical crowd. In the demo, you can see that you can change the text and color on the fly, without having to wait in the code editor for 30 seconds. That’s the big difference. And team efficiency, we sync with Github, Cursor for technical staff, and non-technical people don’t have to worry about the local file system and commits. This makes it more reliable, and although we were late to the AI code space, users have reported that we are the most stable.

Moderator: Lovable does most of the work of building the software, and then uses Cursor to tweak, which you say other companies don’t do?

Anton Osika: I don’t know anyone who did.

---

04 

A product built by a team of 18 people

1 million new ARR per week

Moderator: What is Lovable’s ultimate vision? What will it look like in 5 to 10 years?

Anton Osika: We build the “last piece of the software puzzle”. It’s hard to predict what the world will be like in five years, but I think it’s almost instantaneous to go from wanting to change a product or building a new product to a complete implementation, and it will be possible to seamlessly integrate with existing systems or third-party services. After that, AI can also analyze user behavior, suggest improvements, and automatically run A/B tests, which will soon come.

Moderator: It makes me think that people are curious about which jobs are more important and which skills are not important in the future. I think discovery and ideation (knowing what to build) and taste and craftsmanship (judging whether it’s right) is more valuable, because the specific work of building software has been replaced by AI. Software engineering used to be the hardest and most valuable thing to implement, but now it’s about figuring out the requirements. What do you think?

Anton Osika: If you’re the founder, I totally agree with that. It’s important to find the pain points, then figure out how to solve them, and make the product 10 times better. For existing products, taste is even more crucial. Engineer skills are still important because of the ability to understand technical limitations. However, they should not just focus on the technology stack, but think abstractly and be able to translate human problems into technical solutions.

Moderator: Will the future of software product building be more like a product manager supervising AI engineers, or will programmers still need to code deeply?

Anton Osika: Generalists are more important than they used to be. When I build a product team, I will look for people with multiple skills, and it is best to know a little bit about architecture, design, product taste, and user communication.

Moderator: How many of you are there now?

Anton Osika: Now 18 people.

Moderator: I thought you were going to say 100 people. The team has only expanded from 15 to 18 people, what exactly do you look for when hiring?

Anton Osika: The most important thing is the attitude of extreme care, not just what you do as work, but also about the product, the users, the teamwork. They have to have superpowers and can learn any skill quickly, but they’re especially good at extracting value from AI and large language models and understanding how to optimize our products.

Moderator: Team of 18 with 1 million ARR added every week. Your people are clearly world-class. How do you evaluate during the interview?

Anton Osika: I asked them what they had done in the past, and these people tend to be very committed to what was before. I’ll dig into the technical details and give them a new problem they haven’t seen before to see how they think about it. We give them a trial work for at least one day, preferably a week.

Moderator: You mentioned looking at whether they were obsessed with something in the past. 18 How many of them are engineers?

Anton Osika: 12 of them can code.

Moderator: You said earlier that you have engineers, and now your job is to make content, which is multifaceted. I have a job posting from you, and I read a few sentences: “Long hours and high intensity, candidates need to thrive under the high pressure of the AGI timeline, difficult tasks ahead, success is honored and recognized, those who seek comfort do not disturb.” There are also “working with great minds”, “a mission that goes beyond ordinary engineering roles”, and “success sharing the rewards”. What do you think of these job openings?

Anton Osika: I have AI to help with formatting, but the content is mostly written by me.

Moderator: For some people, they will be intimidated by this job advertisement, but for the people you want, they will feel that they are in the right heart, great screening. You are the fastest-growing startup in Sweden and Europe. How is starting a business in Europe/Sweden different from the US/San Francisco?

Anton Osika: I think the AI era is the most influential moment in human history, and you have to be super ambitious to get into it. This kind of ambition is not common in Sweden, and motivating them in places with low average ambition but a lot of potential talent is a good formula, somewhat double-edged, but also advantageous.

Moderator: Are you saying that there are great people in Europe, but they are hard to find, so you have to attract them?

Anton Osika: Yes, most Europeans don’t want to pursue careers that require extreme ambition. Finding these people is a big challenge.

Moderator: I’d like to talk about priorities. You have a myriad of things you want to do and are asked to do, how do you decide exactly what to do?

Anton Osika: In short, find the biggest bottlenecks and product issues and solve them quickly and iteratively without making too long a roadmap. Understanding the biggest issues is not easy, we take the time to talk to users and look at feedback and feature request boards. Once the problem is chosen, it is more engineering-oriented, and a product like ours is difficult for a non-engineer product manager to decide, because the solution may involve technical details or large technical initiatives.

Moderator: What is the general cadence and the transition from idea to release?

Anton Osika: Before March, we mainly decided what to do, we planned as we made the product, we used Figma Jamboard to list the main issues, prioritize, set the priorities every week, synchronized on the weekends, and unified understanding. Now there’s a roadmap, such as support for custom domain names in the next release, plus collaboration features. But the biggest move is to make the system more “agential”, and there is a longer roadmap for this.

Moderator: How far is the roadmap usually set?

Anton Osika: The next few months are the clearest, the furthest March, but it may change in a month.

Moderator: What tools do you use?

Anton Osika: Linear, we use it for talent tracking, and Figma Jamboard.

Moderator: When will AI engineers join your company?

Anton Osika: That’s an interesting question. The definition of AI is always changing, Lovable is a human interaction interface, and it doesn’t matter how it is implemented internally, it may be considered a “proxy”, but it is not critical.

Moderator: You said that office work and having lunch together helped you speed things up, are there any other ways to speed things up that you haven’t mentioned?

Anton Osika: Most of our members are in the office, which is great. Feel free to communicate, “We’re thinking wrong,” or “This is the right thing to do.” An hour of lunch together at noon is good, and everyone is subconsciously thinking about problems, office work, both focused and high-bandwidth.

Moderator: The CEO of a cutting-edge AI company said that the secret to a fast pace is to “have lunch together”, which is too humane.

Anton Osika: Right.

Interviewer: What would you do if you set up a new product team, and how would it be different from the past?

Anton Osika: People have to be excited about AI, it’s good to work as a team and solve problems together. Now the bottleneck is not engineering, but taste and user intuition, and engineers and others would do well to listen to users and understand what they care about.

Interviewer: Do you have any commonalities in the backgrounds of the people you hire? Super awesome? Winner of a coding competition?

Anton Osika: Primitive cognitive ability is the strongest correlation, as well as a hacker mindset, a love of rapid iteration, and a concern for the overall business rather than just focusing on personal expertise.

Moderator: You said that empowering the 99% of people who can’t code excites you, is there anything you’d like to share?

Anton Osika: Engineers or founders often fail to find technical talent. Now tools can solve everything, and it will lead to an explosion of entrepreneurship and better software. There is no need to put up with rotten technology, people with ideas will directly build a software and display it on social platforms, which is an empowerment for all industries and all regions.

---

05 

Add AI capabilities to software from the perspective of user experience

Moderator: What’s next for Lovable? What’s coming soon?

Anton Osika: I said “agentization”, which gives the system more freedom to decide the next step, such as writing tests, running tests, fixing bugs, which is a big breakthrough. There is also support for custom domain names and team collaboration. Further, after helping the founder build the first product version, help him move forward, such as customer acquisition, feedback, and promotion.

Moderator: I was about to say this. Many people are good at building products, but few people know how to promote them. Growth is another skill, and your ideas are so cool.

Anton Osika: We’ve put a brochure on our blog.

Moderator: You said that you can work with the existing codebase, and a lot of people care about this, can you tell us about it?

Anton Osika: You can’t use any codebase yet, there are research previews that you can import. But when engineers use Lovable, they can edit the code with any tool.

Moderator: Most of the audience is working on existing products, and you say they will be able to use Lovable when they do that work in the future?

Anton Osika: Right.

Moderator: Last question, we have a “Corner of Failure” session. What have you learned from a complete failure in your career?

Anton Osika: I can’t think of a complete failure, but there’s a lesson in the product. I was the first employee of Sana Labs, an AI startup, and we wanted to help users be twice as efficient with personalized learning. We made an API for educational software to make it more personalized, but adding an AI API with advanced features to an existing product is like replacing an internal engine, which is a bit harder and not very successful.

The lesson is that you should add AI capabilities to your software from the full picture of the user experience, rather than shoehorning technology into your product.

Moderator: A lot of people who only focus on technological advancement will take this for granted and then fail. Lovable, on the other hand, helps users focus on problems, requirements, and validation.

Anton Osika: Maybe we should add a “learning mode” and say “Wait, why is this product doing this?” next to you like a product coach.

Moderator: Finally, Anton, is there anything else you’d like to share?

Anton Osika: The world is changing at an accelerated pace, it’s interesting, enjoy the change. The best way to advance your career or change jobs is to be in the top 1% of AI tool usage. Try Lovable and other tools to try to understand them and use them effectively. Here’s my advice.

Moderator: Specifically, how do you know that you are in the top 1%, how do you do it?

Anton Osika: Spend a week using AI to help you achieve your goals, find a pain point, solve it thoroughly, and make it something that someone can use.

---

Author：Founder Park
Source：https://mp.weixin.qq.com/s/atAFAacpH2_-vh3GH3br5w
The copyright belongs to the author. For commercial reprints, please contact the author for authorization. For non-commercial reprints, please indicate the source.