Marblism (YC W24) generates software from a single prompt 💡

CV Deep Dive

Today, we’re chatting with Ulric Musset, Co-Founder and CEO of Marblism. 

Marblism is a ‘prompt-to-app’ platform that uses AI to build full-stack software. Co-founded by Ulric and Cyril Pluche in 2024, the startup’s mission is to make developers 10 times more productive by automating 90% of front-end and back–end code and allowing builders to iterate on their ideas faster. 

Today, Marblism has thousands of developers using its platform for web development. Over 8,000 individual AI apps and SaaS tools have been built on its platform since launching just a few months ago. The young startup recently graduated from YCombinator’s W24 batch, where they spent 3 months refining the core product and onboarding users onto their platform. 

In this discussion, Ulric shares his thoughts on AI-powered code generation, Marblism’s approach to full-stack, and how YC helped them find their first customers. 

Let’s dive in ⚡️

Read time: 6 mins

Our Chat with Ulric 💬

Ulric - welcome to Cerebral Valley. First off, share a bit about your background and what led you to start Marblism. 

Hey - I’m Ulric, co-founder of Marblism. Our platform enables developers to generate software from a single prompt. Before Marblism, I was running Vauban.io, a platform I co-founded that simplified launching VC funds, which Carta later acquired. After our exit, Cyril and I were going through ideas for our next project and we found ourselves with a pretty long list. 

So long, in fact, that we thought to ourselves, "How amazing would it be to build an AI engine that we could use to execute all of these ideas?”. That's how we got started. We are building the product we would have dreamed of having when we started all of our previous software projects.

How would you describe Marblism to a new developer? 

You go to marblism.com and type in a prompt for what you want to build - whether it’s a SaaS tool, an AI app, a marketplace, or a social network. The AI then builds it for you - the frontend, backend, the database - everything, you get a fully functional software. After that, you can preview your app and prompt it to make changes or add new features. One important thing is that you own the codebase, you can add your Github account and work directly on it. 

Tell us a bit about your current users. What can you share about how they’re using Marblism? 

Today, our users are mostly engineers and are using the platform to save themselves months of development time. For example, we have two users building a SaaS tool to help doctors record their patient calls and summarize them using AI; they built it entirely on Marblism, which is fantastic! So far, there are over 8,000 applications that have been built this way. The common denominator is that our users are building AI-enabled products because I guess that’s the area everyone is concentrated on right now. 

The overall user feedback has been amazing. We received a lot of feedback as well to support more frameworks or more types of software and so we are working on that at the moment. 

Walk us through Marblism’s tech stack - how are you pulling together all of the pieces required for the AI to generate software accurately? 

Essentially, our core engine is an orchestration layer where various agents work on different aspects of the code base - for example, one agent working on the product requirement, multiple agents for the back-end and front-end, and an agent that's going to fix the code and make sure the application runs and compiles correctly. 

We ended up recreating our LangChain-like orchestration layer because we had specific use cases that we wanted to get right. We don't use AI to generate absolutely everything in the code base. For example, you don't need an AI to generate your authentication because you want your authentication to be secure and stable, so the AI is just going to customize some modules and integrate them into an application. That's how we can generate customized codebases that are scalable, maintainable, and secure.

How do you prioritize what to build internally, given the pace of AI advancements is so fast relative to the size of your team? 

We spend a lot of time talking to our users about what they want from Marblism. Our main focus is to increase the complexity of the apps you can generate. Eventually, we want to be able to generate mobile apps or VR apps for example. The idea is really to be able to generate any kind of complex software from a prompt.

In terms of keeping up with AI more broadly, we track a list of AI publications on Twitter so that we can keep an eye on the new research. That being said, we try not to get too distracted from our main focus, we don’t want to spend years in R&D and not productize anything valuable as a startup.

What’s the biggest challenge around building Marblism - cost, or accuracy, or speed? And how are you planning to monetize in the long-run? 

Our biggest challenge is speed. When you make a change to an app, if you wait 1 minute to see the change, it’s frustrating. It’s interesting to see how people deal with that. They prompt to make big changes or add a new feature, and for tiny changes like changing the position of a button, they will directly edit the code because it’s quicker. With AI getting better, I'm sure we'll cut down that wait time soon.

On monetization, users get a free pack of “marbles”, our in-app currency, which allows them to start and iterate on a project. If you want to do additional iterations you need to get more marbles. We’re also going to launch our hosting solution soon, where you can host your application on Marblism, and that will be another revenue driver for us. 

There are a number of startups, including Vercel and Replit, who are chasing the code-gen opportunity for full-stack developers. How are you planning to differentiate Marblism from others in the space? 

A lot of startups have a very generalistic approach to code generation. They’re approaching the problem via a broad range of languages or frameworks, whereas we’re super focused on one specific framework, React + Node.js, and one specific language, TypeScript.  

At my previous startup, we made the mistake of serving all sorts of related customers - real estate funds, hedge funds, crypto funds, and VC funds. At some point, you lose your focus and it becomes unmanageable. So, we pivoted to only focusing on VC funds, and that gave us much greater depth. Taking this approach with Marblism, we can go really deep and create complex code bases with 500 files due to this narrow focus. We think this approach is better - mastering one vertical before moving to the next. 

Tell us about your team - what do you look for in prospective new team members? 

We have one unique value at Marblism, which is that we expect new members to bring a quantum leap or a 10x to the product. Essentially, we want people to bring true innovation to what we’re building - for example, if you can somehow reduce the time a prompt takes to execute from 30s to 5s, that’s what we want. We’re looking for groundbreaking changes and major leaps forward. That’s the most important thing for us at this stage. 

Lastly, how has YC impacted the trajectory of Marblism in the past few months? 

YC is legendary, of course. We knew it would be really helpful, but we’ve been amazed at how much it’s helped us in terms of investors. At my previous startup, we were preaching in the desert to convince investors - this is definitely a different experience. Another thing people don't fully recognize about YC is that it's really helpful for finding users. When you are on the YC directory list, both investors and customers reach out to learn more about you, which has been insanely helpful.

Anything else you want readers to know?

We’d love for people to give marblism a try and share their thoughts with us on Discord. 

It’s still Day 1 and we’re gonna release a lot of very cool stuff in the coming months. So stay tuned!

Conclusion

To stay up to date on the latest with Marblism, follow them on X and learn more at them at Marblism.

Read our past few Deep Dives below:

• 3/15: Baseten is pushing the boundaries of AI Inference ⚡️

• 3/11: Martian's interpretable alternative to the Transformer 🔌

• 3/8: Our chat with Groq's Chief Evangelist, Mark Heaps

• 2/26: MultiOn is building software with a brain 🧠

• 2/23: Galileo AI's groundbreaking prompt-to-UI tool ✨

• 2/19: Our chat with OpenAI’s Logan Kilpatrick

If you would like us to ‘Deep Dive’ a founder, team or product launch, please reply to this email ([email protected]) or DM us on Twitter or LinkedIn.

Join Slack | All Events | Jobs | Home