Create - your AI product engineer 🎛
Plus: Co-Founder Dhruv on how the team plans to make software creation accessible to everyone...
Published 03 Jan 2025CV Deep Dive
Today, we’re talking with Dhruv Amin, Co-Founder of Create.
Create is an AI “text to app” builder that lets entrepreneurs build apps—covering frontend backend, databases, and authentication—with English. Create’s making software creation as simple as describing what you want and powering a new generation of entrepreneurs who use it as their AI product engineer.
Launched a year ago, Create’s growing fast and is being used by internet builders, entrepreneurs, and PMs to build everything from AI micro-SaaS products to internal tools and marketing experiences. The platform combines the simplicity of no code tools with the power of code generation.
In this conversation, Dhruv shares the story behind Create, the challenges of building at the frontier of applied AI, and how the team plans to make software creation accessible to everyone.
Let’s dive in ⚡️
Read time: 8 mins
Our Chat with Dhruv 💬
Dhruv - welcome to Cerebral Valley! First off, give us a bit about your background and what led you to co-found Create?
Hey there! I’m Dhruv, co-founder of Create. Create is an AI text-to-app platform for entrepreneurs that helps anyone build real working products using natural language. We started it a year ago, and it's growing quickly. The original insight for it came from the experiences of my co-founder Marcus and I. We started our careers in product at Google—I was the first PM for YouTube TV, and Marcus a PM on the Google Maps team, growing daily usage 30% there. We both have always loved making new products and think it’s one of the greatest ways to help others with your own unique perspective.
After moving on to start our own companies, we were struck by how difficult it is to build software and how few people in the world can actually do it. That inspired us to get started on Create. Initially, we were building proprietary code generation systems to help larger companies build applications, but as LLMs got better (and cheaper) at code generation, we realized there was an opportunity to take the company to new heights.
We’re now building an AI product engineer to let anyone, regardless of technical experience, to build products. Many of the product choices we’re making are to make it easy for non technical users to build real software for the first time.
How would you describe Create to an AI engineer or user who isn’t as familiar?
Create is a text-to-app platform for entrepreneurs—your AI product engineer. You chat with it and it turns your requests into apps. It lets you build everything you need for a new product: the front end, back end, database, authentication, payments, but also the marketing and internal tools. Many compare us to traditional no-code or low-code tools, but with one big difference: Create is a “yes-code” tool. It’s as easy and intuitive as no-code platforms but significantly more powerful in what you can make because we’re generating code. It means the UI and business logic can be exactly how you want with way less time and effort.
We’ve been building the product for about a year now. We’ve been iterating quietly with our initial set of users, but it’s now growing fast as folks find out and spread it to their friends or teammates. It resonates most with people who want to launch new products online, particularly AI micro-SaaS projects. For example, one user in the safety industry built an entire suite of AI tools with Create—everything from the marketing pages to the core application and backend. He’s now monetizing and selling the product online to his niche.
In addition to entrepreneurs, we see strong pickup from PMs, marketers, sales, students —those who’ve never had the ability to code applications but now can with a few lines of English.
Talk to us about your users today - who’s finding the most value in what you’re building with Create? Any use-cases that are the most prominent to date?
We think of it as building Shopify for SaaS. Just as Shopify made it easy for anyone to set up an e-commerce store, Create is opening up SaaS entrepreneurship to people without a developer team or coding background. For a decade, SaaS has been the world’s greatest business model, but historically, this kind of entrepreneurship was only accessible to those with technical expertise. The most impressive projects I’ve seen are where someone uses Create to launch a product for their niche that generates real revenue, making everything from the core application to the marketing to their backend admin tools. Create lets them get off the ground, get their first customers, and iterate fast.
You can prompt Create for whatever you want, so we do see a long tail of use cases beyond product building. While many build products, sites, or internal tools others use it in completely creative ways—as a game engine, creating presentations, a design tool, or an educational tool. The flexibility of code gen makes it fun and opens it up to audiences even larger than our core target users.
How are you measuring the impact you're having on your key users so far? Any specific metrics or analytics you’re most focussed on?
We’re in growth mode now, so we’re focusing on metrics like signups, engagement, and retention—the classic product loop. Beyond those top line metrics, we do many evals around quality and speed.
On the quality front, a major focus for us is the rate at which users encounter code errors in the builder. We’ve done a lot of work using a mix of deterministic methods and AI to improve reliability. Currently, 98% of the time, the generated code compiles successfully in Create, and we can even auto-fix runtime issues on the fly to deliver a higher-quality experience. We also use new AI analytics tools to measure satisfaction with the assistant as users build with us - and ship updates to improve categories of requests.
Speed is another big area of focus—how quickly users can see the first thing they’ve asked for come to life. We’ve put significant effort into how we generate code, intelligently load the context for a request, do diffing, make targeted updates to specific parts of the codebase, and how we render the app to make sure everything feels fast and immediate for users.
Ultimately, though, the most important downstream metric is whether the apps people are building are gaining traction. We track how often users publish and host apps on Create. We also look at the views those apps are getting and monitor runtime errors, so we can continue to help fix and improve the experience. We want to help entrepreneurs launch software businesses on top of Create so our long term north start metric is the success of those businesses.
Take us under the hood of Create’s stack - how are you thinking about merging multiple models and fine-tuning to get the results you’re looking for?
We use all the major models and combine them into our assistant experience, while also fine-tuning our own models on top. Our focus is really around helping users who can't necessarily read or write code to still build with code. We have to come up with clever UX paired with AI advancements to unlock that.
We’ve built a ton of infrastructure around routing to different models for different requests and different parts of the experience. For areas where we need more speed, like error correction or design controls, we rely on our fine-tuned models. Building a real product is a surprisingly deep workflow - whether it’s getting the design right, adding an integration or an external API, getting auth and the schema correct, deploying it, and making it all work together. For each part of that flow, we’re thinking about how to do it in the LLM-first or AI native way that abstracts complexity for our users and “just works”.
It’s been fascinating building the product over the past year as the models have improved. Each model release pushes the frontier further in terms of the sophistication of applications users can build on Create. I clearly remember every major release—like GPT-4o, Claude 3.5—as moments where the types of applications you can build became much more powerful. We’ve also done major UX and tech changes to how we do code generation with each model release so users can harness that power and better steer Create to build what they want.
We’re now starting to work on workflows with o1 and other reasoning models, where it can take more work off the users plate, like generating multiple parts of the project in parallel or designing the optimal schema for the database or accomplishing more tasks in the background. These advancements raise the abstraction level, letting users communicate at a higher level with Create about what they want their applications to do, while Create handles more of the details.
We've seen a lot of excitement around AI agents and the idea that autonomous systems will be able to complete a lot of different tasks. How are you thinking about integrating AI agents into Create itself?
Our earliest version of Create was actually an agent around the time of GPT-4. We found back then that it was too slow and too bad at planning to build aligned with what users wanted. So we shifted to a faster assistant model that guides you through building the product bottom up. We found that the immediate feedback encouraged users to iterate more and give us more context, which then meant we could do more work for them to get it right.
It’s exciting to me that the models are now getting faster, cheaper, and good enough at reasoning to make agent flows viable from a user experience perspective. We’re soon introducing some new agentic workflows where you can tell Create what you want and it will kick off building it for you. I think this time around they’ll be a hit just because they’re much faster and can take more of the context into account. Create does a lot for you today but there’s even more leverage we can give the user. I’m especially excited about introducing agents on top of the English first app builder we’ve built. It means after an agent finishes its work, you can update or tweak it to get it exact in a few sentences.
We also have an increasing opportunity to let users build their own agents. Today, our most successful users are building AI SaaS where its AI functions, custom UI, databases, auth, and payments. I think we'll see more requests where builders want to turn the AI into their own autonomous agents that can do work for their users. We have an opportunity to make creating those super simple for our users.
What has been the hardest technical challenge around building Create into the product it is today?
In the early days, and I’m almost forgetting some of the work we did here, we focused heavily on incremental JavaScript compilation. This involved compiling JavaScript on the fly so users could get very fast visual feedback on what was happening on the canvas. It was a breakthrough in usability. We also put a lot of effort into error correction, including parsing the syntax tree to understand and fix a wide range of errors deterministically. Additionally, we implemented on-the-fly error correction to ensure the code both compiles and runs. This foundation is what makes a product like this possible. It’s a pretty challenging task, especially when streaming from models and running code simultaneously.
As models become more capable and projects grow larger, the interesting code gen work now revolves around how to accurately update the right parts of the code for users who aren’t actively looking at the code. We're using both algorithmic and AI-driven methods to ensure the right context from the chat and code versions is always fed into the model to get the best results for each request. We’re also developing our own algorithms to do accurate, instant application of code diffs. It’s a lot of iterative improvement focused on reliability, speed, and quality.
One of our strongest value props is that you can set a Create project live on the web in 1 tap. It might be one of the fastest “time to live”’s on the internet. So we’ve had to solve instant deployment and how to run apps at scale with untrusted code with strong uptime, error handling, and performance.
App building itself is a deeply intricate workflow. It involves everything from design and ideation to implementation to deployment to production changes. We're developing AI native techniques for each stage. There’s still so much interesting work to be done.
One of the technical challenges we’re starting to tackle now is that we have hundreds of thousands of projects built on Create. Chances are, if you’re building something on Create, someone has already built something similar. The question is, how do we now do great retrieval-augmented generation and discovery of everything built on Create, and then use that to help improve the next user’s experience? I’m excited to nail this so that building on Create feels like building with a rich community where most things are solved for you.
How do you plan on Create progressing over the next 6-12 months? Anything specific on your roadmap that new or existing customers should be excited for?
We're doubling down on what our users love. We want to make things that are hard even when writing the code super easy in Create. So whether it’s setting up Stripe or changing how auth works, we’re baking these into the platform to make them as simple as using a Notion like slash command or chatting with an assistant. We'll continue expanding the roadmap in this direction.
We’re also excited to enhance all the core primitives we’ve built, like components, databases, and authentication. There's a lot of surface area to cover, and we’re making these primitives AI-native — rethinking workflows like setting up a database or making your own component libraries to match your brand in a way that makes sense when building alongside a powerful AI assistant.
Another major area is discovery—exposing what people are creating in Create and making that accessible and useful to everyone. This is a key focus for us.
Finally, we’re constantly pushing the boundaries of code generation and what these models can do to accomplish jobs for users. There’s a lot of potential, from larger projects, to agents that can do work for you, to better code generation for design, and we have a rich roadmap ahead. Right now, there are five of us, so in the next few months, we’re growing the team as we turn our initial momentum into the next few million users.
Lastly, tell us a little bit about the team and culture at Create. How big is the company now, and what do you look for in prospective team members that are joining?
It’s five of us working together in downtown SF—a small, tight-knit, talent-dense team of people who care about building great products. It’s myself, Marcus, James, and Zobeir on the product & eng side, and Zaria our designer.
Over the next few months, we’re hiring three more engineers and a first marketer. We look for folks who are intense and real owners - you go from ambiguous problems to solutions relentlessly and autonomously. We’re all fired up when a new entrepreneur launches something with Create - and want to work with people similarly fired up about that mission. We’re hiring for full stack builders who want to ship fast and have strong product taste themselves - so we pay special attention if you’ve built products people love to use or who have built interesting things with LLMs.
This is an ambitious problem—building everyone’s AI product engineer—and we’re constantly adapting to changes in models and what people want to build. You have to be fearless and a little crazy. It requires innovation at both the AI and engineering layers, turning those advancements into delightful user experiences. We’re looking for people who can contribute across all these areas.
Conclusion
To stay up to date on the latest with Create, learn more about them here.
Read our past few Deep Dives below:
If you would like us to ‘Deep Dive’ a founder, team or product launch, please reply to this email (newsletter@cerebralvalley.ai) or DM us on Twitter or LinkedIn.