Codesphere: Shaping the Future of Collaborative Cloud Development

Most of the applications we interact with daily already run in the cloud – someone else’s computer in a data center that provides computing power based on an application’s demand and allows the application to scale up and down quickly.

Yet, application development still mostly happens locally on private computers, which creates silos. This leads to additional overhead when the application needs to be deployed, that is, sent to the cloud to run there.

Codesphere bridges this gap by leveraging the cloud to develop applications easily, efficiently, and collaboratively. Founded in 2020 by Elias Groll, Jonas Zipprick, and Roman Frolov, Codesphere raised a $5M seed round in early 2022 led by LEA Partners and with the participation of 42CAP, New Forge, 468 Capital, and angel investors from Shopify, Google Cloud, and Digital Ocean.

Learn more about the future of collaborative cloud development from our interview with the co-founder and CEO, Elias Groll:

Why Did You Start Codesphere?

The idea for Codesphere was conceived when I was working at Google as a software engineer. We had an internal system to deploy applications to the cloud, replacing Kubernetes, which made it really easy to develop cloud applications collaboratively. I thought everyone should be able not only to deploy but also to develop applications collaboratively in the cloud – and thus ship them much faster than if they were developed locally.

Also, after joining a large corporation, I felt like my life had slowed down and that founding a startup would be more exciting. So, I quit my job at Google and started Codesphere.

How Does Collaborative Cloud Development Work?

I believe that eventually, everything will run in the cloud, just like applications run on Android or iOS today. But for this to come true, we need to be able to build much faster for the cloud and move the whole development process there: Build it where you run it.

Let me give you an analogy. People know Salesforce as a configurable CRM to manage meetings and contacts, but it’s a complex enterprise product and requires a lot of maintenance and setup. In contrast, HubSpot is a full-fledged solution that not only does CRM but also marketing, email automation, and much more, thus providing a better experience if you want to get started quickly. HubSpot is analogous to what we’re building with Codesphere: a full-fledged solution so you can get started easily with cloud development and not need to hire many DevOps engineers upfront to deploy your application.

This leverages all the benefits that come with the cloud: scalability, faster development pace (as everyone is on the same platform), and better compliance since there’s just one platform to audit. Also, we make it really easy to preview features so you can make sure everything runs as expected.

Currently, deploying to the cloud requires Datadog, configuring multiple Kubernetes clusters, buying from different cloud providers, and managing all of that. This complexity is unnecessary and hinders productivity. We make it effortless to deploy your code to the cloud.

You can instantly see how your application is doing, scale it horizontally by adding more services, or vertically by scaling the server size up and down. You can see performance peaks, perform A/B testing, and manage load balancing in real time. All of this is available at your fingertips, making it easy to make data-driven decisions.

In ten years, software will be omnipresent in the cloud and not dependent on any particular device. It will be a great world where a lot of computing power will be available to everyone. But we have to build the tools to get there first. Currently, we’re building the infrastructure to accelerate the development and deployment of applications, but a lot is yet to come, especially through the inclusion of AI.

With large language models like ChatGPT and GitHub Copilot, we just got the first glimpse at what will be possible in the future. Fundamentally, however, it’s the wrong approach to generate code as it just looks for statistical patterns. Code has a very clear structure: for example, the five public methods of a class in Java are known to us, so we don’t need a statistical prediction based on other classes in the codebase. Also, training a large language model on the entire internet and GitHub involves training on a lot of badly written code, which dilutes the quality of the code suggestions.

But despite these challenges, large language models are still pretty good not only at completing a sentence but also at generating the next few lines of code. ChatGPT won’t produce new and better algorithms, large-scale code bases, or replace algorithm developers. Yet, it is already very useful for answering a developer’s questions, for instance, by building on top of StackOverflow, thereby boosting productivity.

As AI evolves, it will become much better at generating useful code, converting code from one programming language to another, or answering questions that are specific to a code snippet. We’re actively monitoring the progress in AI and looking into ways AI can help developers within Codesphere with code development.

How Did You Evaluate Your Startup Idea?

We started out building something technically interesting, and when we launched after half a year and got to our first 30,000 users, we didn’t charge them nearly enough to cover our expenses. That’s when we started looking more seriously into product management and business development.

When you’re trying to build the new AWS, you cannot rely on AWS but have to build your own infrastructure first. So our approach involved everything from Linux to Kubernetes to the applications running on top of it, and that took us some time and detailed product management.

In our process, we first define our objectives. Here, it is really handy that we have very experienced people who previously worked on Google Maps or Kubernetes on our team, and so they have a really good intuition for things a typical user might not think of. Then, we combine this experience with our objectives and come up with a hypothesis about what might work, and create some mockups.

We also ask potential customers about their requirements and budget, and then let them test the mockups of new features themselves. As we observe them play around, we test for two things: whether the features work in principle and whether they understand how the features work and can use them.

We then write a product brief and start the development. Once developed, we market the feature and dedicate further research to what needs to be changed going forward. This process is super detailed because every feature is complicated, involving multiple microservices, and one has to avoid state conflicts between them.

What Advice Would You Give Fellow Deep Tech Founders?

Before you worry about finding a co-founder or hiring, first do the work and test different solutions for the problem you’re solving. Focus on speed of iteration, quick prototyping, and maximizing your learning on what works instead of falling in love with one particular idea.

Only start looking for co-founders and employees when you’re confident that you’ve found a viable solution, and hire them only if they can help you address a particular risk that your startup is facing. Then implement a clear organisational structure and hierarchy; otherwise, diffuse roles and responsibilities will slow you down.

Finally, talk to investors only when you’re really confident that your solution works and you need to hasten its development or scale it to reach more customers. Don’t get investors on board if you still have lots of risks that you could have derisked yourself. Research the problem and solution yourself first and get your prototype out there as fast and as simple as possible.

Codesphere: Shaping the Future of Collaborative Cloud Development

Why Did You Start Codesphere?

How Does Collaborative Cloud Development Work?

How Did You Evaluate Your Startup Idea?

What Advice Would You Give Fellow Deep Tech Founders?

Further Reading