Varun Mohan, Co-Founder & CEO of Codeium – Interview Series

Varun Mohan, is the Co-Founder & CEO of Codeium, a code acceleration toolkit built on cutting edge AI technology. Currently, Codeium has two main capabilities: Autocomplete, which suggests the code you want to type, saving you time on everything from boilerplate to unit tests, and Search, which helps you search through your repository using natural language questions.

Could you share the genesis story launching Codeium as a platform to make deep learning workloads more efficient?

When Douglas Chen and I founded the company in 2021, we named it Exafunction, a company focused on GPU virtualization software. We were managing ten thousand GPUs across the public cloud at the time. In a short time, Exafunction gained customer traction very quickly and became a leader in deploying deep learning models at scale.

But our extensive experience with AR, VR, and autonomous vehicle technologies—domains that represented significant, real-world applications of deep learning—eventually brought us to a bigger realization. With the advent of generative models, particularly transformer-like models, we predicted a pivotal shift in the industry.

We observed that, in a world increasingly dominated by generative AI models, the value for end users would not merely lie in providing infrastructure. Instead, it became clear that the true potential to impact was in leveraging this infrastructure to create applications that directly enhanced productivity and efficiency for developers. This is why we decided to focus on building a code acceleration tool on top of our own infrastructure. That was the birth of Codeium.

This strategic pivot was not just a response to the evolving technological landscape but also a proactive move to place Codeium at the forefront of AI's application in software development. By focusing on code acceleration, we aimed to empower developers, enabling them to harness the power of AI to streamline their workflows and innovate at an unprecedented pace.

By mid 2022 we had no users, but in 2023 Codeium shot up to 300,000 active users and 100 enterprise customers. We took a risk but it quickly paid off.

You initially made the Codeium platform entirely free for developers, what were some of the logistical and technical challenges behind this choice?

The decision to offer Codeium for free to developers stemmed from a vision of widespread accessibility and adoption, yet it introduced a suite of logistical and technical hurdles.

Firstly, the financial implications of maintaining a deep-learning-based platform presented a significant challenge, as infrastructure costs could spiral without meticulous optimization—similar challenges faced by competitors. So, refining the infrastructure for maximal cost-efficiency was critical, while ensuring the platform could deliver high-quality services without necessitating prohibitively expensive hardware for both the company and its users.

This led to a “GPU light” approach, allowing for effective operation with minimized hardware investment. Ultimately, sustaining a free service that was scalable and performant demanded infrastructure capable of supporting a growing user base without degrading service quality.

There were also security and privacy considerations we had to solve for, which are essential to safeguard user data against unauthorized access.

It’s all a complex balancing act of innovation, strategic resource management, and steadfast commitment to user value.

Codeium offers capabilities to be self-hosted, why is this beneficial for enterprises?

The security of IP is critical for enterprise. Our self-hosting capabilities address their needs for security, customization, and control of data—especially for those wary of external managed services' potential vulnerabilities.

Beyond security, Codeium stands out for its adaptability; it can be personalized according to an organization's private data and workflows, ensuring that the platform not only integrates seamlessly into existing processes but also optimizes them.

This level of customization extends to its compatibility with various source code management tools, providing flexibility for companies not using mainstream platforms like GitHub and thus broadening its applicability.

The self-hosted option, despite potentially higher initial costs, emerges as a cost-effective choice in the long run. It minimizes the need for expensive hardware through a GPU-light experience and avoids the recurring fees typical of cloud services, making it an economically sensible solution for large-scale enterprises.

Another important thing to note: Codeium's extensive feature set, which supports a wide array of programming languages, development environments, and specialized tools like Jupyter Notebooks, ensures it meets the diverse needs of different departments within an organization. This level of support not only streamlines various development tasks but also enhances productivity across the board, making Codeium an indispensable tool for enterprises seeking to maintain control over their development infrastructure while leveraging advanced code acceleration capabilities.

Code generation tools is a very competitive industry, how does your platform differentiate itself from competitors such as GitHub?

There are many differentiators I listed above but, to sum it up, one of the big ones is we are not tethered to GitHub or any source code management tool. We provide a personalized experience to cater to however the developer wants to work from wherever they want to work.

You’ve spoken before about improving the personalization of the product, could you go into some details as to the current personalization that is offered, and what your end goal is?

Personalization for us is adapting Codeium to each user's or organization's specific coding practices, preferences, and the technical stack they operate within. This means we can tailor suggestions, code completions, and other features based on the unique data and patterns of the organization or individual.

For instance, if a company predominantly uses a certain programming language or framework, Codeium would prioritize support and optimizations for those technologies. This approach extends to integrating with various source code management tools and environments, ensuring seamless operation regardless of the users' existing infrastructure. Personalization also manifests through supporting specific tools, like Jupyter Notebooks for data scientists, as an example.

This all means companies can enable their best engineers to do high-value work. They can write more software and provide more value to the company by spending more time doing the things that they actually liked doing, which is building products, not writing boilerplate.

How do you see Codeium impacting software development practices?

We’ve all seen movies where an individual suddenly gains a superpower and must learn to control it, to save the galaxy. Codeium is a superpower engineers learn to wield very effectively, but the superpower never replaces them.

Engineers will generate a larger portion of their code using Codeium, faster and more easily, with fewer corrections needed, but their code still requires testing. Codeium dramatically increases the engineers’ abilities and efficiency. It makes their day more enjoyable and satisfying. Developers will write more and better code to fill a never-ending need, and it will just keep getting better.

What advice do you have for developers integrating Codeium into their workflow?

I think the biggest advice is to understand these programming tools are purely meant to augment the developers, not replace them.

So they should think about these tools as helping automate routine tasks to save time, like an assistant. You shouldn't be using these things yet to completely generate code without reviewing software.

At the same time, developers will be pleasantly surprised at how well a tool like Codeium can predict your thoughts and be an additional muscle. For example, right now Codeium generates over 45% of all software developers. So it's definitely a massive help and accelerating development right now.

Can you share a success story of how Codeium has benefited a particular project or organization?

Codeium is already actively used by over 300,000 developers. But the success doesn’t just lie in the numbers. Large companies like Dell, Anduril and Clearwater Analytics swear by us.

These have emphasized Codeium’s value within their software development lifecycle, and have also expressed the enthusiasm among their developers.

Whether you're a single indie developer, or someone writing hardcore enterprise software, Codieum provides value across the board.

Thank you for the great interview, readers who wish to learn more should visit Codeium.