In a previous post, we talked about our core cultural principles. Today, we want to talk through how the combination of having healthy realism and a focus on solving problems combine to create a perspective of wanting to prove that we can create value, rather than wanting to prove that we are right.
It is a fundamental trap for most software startups. We start startups with a fundamental belief that we have a solution to a problem that others do not have, based on our expertise, observations, intuitions, or all of the above. We raise money partially on the basis of these solutions and how much of a technical moat we have over competitors. We make money by convincing potential customers that your solution is the solution.
We may have started with a problem, but problems exist independent of the startup existing, and so it is tempting to start to view our ego-centered reality through the lens of the solution provided or technique used. And that is the trap. At some point, it is common to get tied to the solution that we currently wield, in search of problems that may no longer exist. We want to prove to ourselves, our investors, and our customers that we chose the right solution from the beginning. We want to prove that we are right.
Instead, we would be more successful by realizing that new problems exist (or have gotten bigger) that require us to simply pick up a new solution that might not be that far of a leap. Maintaining a healthy realism forces us to constantly question the nature of the problems and solutions that exist, while focusing on solving problems makes sure that we are always wielding the right solution, even if that is not the solution we told everyone about earlier.
There is no bigger example of this at Codeium than the fact that we did not start as Codeium. Our company started in the summer of 2021, far before ChatGPT or any of the generative AI movement. At the time, we noticed there were large GPU workloads, such as autonomous vehicle simulation workloads, that were inefficiently using their GPU resources. GPUs don’t work the same as CPUs, and they are much harder to virtualize, orchestrate, and optimize. So, we came up with an infrastructure software solution that could take any GPU workload and make it significantly more efficient, reducing GPU bills for our customers by almost 97%. We took a cut of our customer’s savings, and made money. We had a real problem, and a real solution. We had raised a Series A on this solution and had a 1M+ ARR business within a year. It would have been easy to fall in love with our solution. After all, we told the world about it.
But then summer of 2022 came around, specifically the Midjourney’s and DALL-E’s of the world. Large GPU workloads - if we were focused on our solution, we would have been excited! But we maintained realism. Our solution was great because it could work for any type of GPU workload and optimize model inference for any model architecture. In a world where we have companies training powerful LLMs with standard architectures, the generalizability of our software, a byproduct of having the best GPU optimization software on the market, would be unnecessary. It wouldn’t be good business to be an infrastructure solution provider, because either these model companies would take care of optimizing their handful of models themselves or competition in the space would commoditize the optimization technology for these particular architectures. We wouldn’t be solving the problems for this new crop of companies. We wouldn’t be valuable.
So, over a weekend (literally), the company pivoted entirely to building a fully vertical generative AI application that leveraged our head start on GPU software technology. Something with different customers that would get value, but a clear value proposition. That became Codeium. We beat ChatGPT by a couple of months, and if we hadn’t maintained a healthy realism, it is unlikely we would have realized that the largest GPU workloads wouldn’t need our software until something like ChatGPT came out. And by that point, given how fast this space moves, we would have lost precious time.
We told our investors that the thesis of the company was changing. We told our existing customers that we would be sunsetting our existing products. We rolled up our sleeves and started a new journey from scratch, with no real guarantee of it working out at all.
It is not that we were wrong with our original solution. We were right, for that time. But times change, and by maintaining healthy realism and focusing on providing value (rather than building solutions), we hope that we continue to stay right for the times.
This is not the only example. We have not been afraid to walk back features from customers when we have learned that new approaches could bring even better results. We keep ourselves honest with the results of growth and marketing strategies and don’t continue investment in strategies that no longer give the same results. We constantly research the market to understand what value propositions we may be missing and what we may need to reprioritize.