34:38
Every time a developer writes code, they need to run a bunch of checks to make sure everything works correctly. This process is called CI, and Blacksmith creates the infrastructure to execute them very, very quickly. We have grown 20 times over the last year. We recently raised Series C investment. The company's valuation reached $1 billion, the round was led by Sequoia, and we raised about $66 million. We were outsiders. We were not platform engineers. We didn't work in DevOps. In fact, none of us had run CI pipelines before. When I think about why we were one of the few to approach this problem and vision in this way, I think part of the answer is to be naive enough in the right
sense not to kill a good idea before it meets the market. So, today, with agent-based programming tools, developers are writing more code than ever before, but there's no point in writing 100 times more code if you can't release it at the same speed. So now the bottleneck is not writing the software, but testing it, and that usually means waiting for your CI. This means making sure that your tests are passing successfully, and that you are confident that what you are releasing actually works. I always wanted to come to Silicon Valley and start a company. In high school, my favorite activity was flipping through Wikipedia pages; I read about technology companies, how they were founded and what they did, and I
wanted to create something of my own. I went to Waterloo because it had a great startup program, a lot of founders came out of it, and it was a very technical university. After I started working at Fair, I started thinking a lot about ideas and how to move forward. By the way, Blacksmith is my second application to YC. The three of us have been friends since university. We were all in Toronto at the time, all wanting to start a company and discussing ideas about what we should work on. And we were all engineers. At almost every company we've worked for, CI was terrible, and it was because it was too slow, managing that part of the infrastructure was painful for everyone, and we felt like we had a better way.
When we first discussed the idea of what became Blacksmith, we were skeptical about whether it would work, but over time, as we explored deeper, our confidence grew. My co-founders were database engineers. I worked on search infrastructure, but from the outside we saw that the reason for the difficulty in managing CI infrastructure is that the nature of the workload is very different from what hyperscalers like AWS, GCP, and Azure are really good at. And it almost felt like a mismatch, which was what caused the problems.
We felt that the real solution was to build a cloud that could handle CI workloads extremely efficiently. It all comes down to using the best tool for the task. We began to understand that the characteristics of CI workloads are very similar to desktop applications, particularly video games that you run on a typical gaming PC, in that they don't care as much about multithreading, but instead require a single thread of execution to run very, very fast. When we first started, we rented these servers from a German data center services company that was known for hosting servers for Minecraft and Call of Duty. Over time, as we grew in scale, we began
to own more and more of the stack. So now we're working at much higher levels of agreements with data center providers, where we're actually buying megawatts of capacity, not individual machines. So I actually invested in Blacksmith twice. I led their seed round after Y Combinator, which was structured as SAFE, and a few months later I led their Series A round and joined the board of directors. I thought it was extremely interesting that they used gaming CPUs, as I've never seen anyone else try this kind of CI solution. It was even a kind of shock to me that someone would even gather a bunch of such things in one place. That was a bit
strange. They usually sit in one big box at home under the table where you play some games. But the idea is, "No, we need to buy a hundred of these and use them for compilation." That's when I thought, "Oh, this will be something different A big part of speeding up CI is avoiding repeating work that has already been done before. At a general level, developers call this caching. A big part of Blacksmith's "secret ingredient" is moving cached artifacts closer to the compute nodes. The first 200 customers were extremely difficult. At first, we asked our friends who worked at different companies to give us a chance. We asked them to provide logos, put them on our website, and gained some trust. Using this, I personally sent cold emails to
CTOs, VPs of Engineering, and platform engineers, trying to convince them to use the product. Our very first clients were small companies similar to ourselves. They cared about speed, so they needed a very fast CI. In addition, there was a minimum of bureaucracy. The only real way to win the favor of engineers is to actually demonstrate the result. And again, solving the problem in a way that they probably wouldn't be able to do themselves. You know, they are very picky. They like to try out lots of new tools. Sometimes they are not the most loyal customers.
Without a sales force, they were able to land large contracts in the five- and six-figure range simply because Blacksmith makes CI optimization and rapid launch as easy as possible. We make it very simple: in one click you can migrate your entire CI to our platform One of the clients that recently joined Blacksmith is LangChain. Previously, LangChain ran its CI on GCP. On GCP, their developers would typically wait 15-20 minutes just to get the compute power to run a CI job. At Blacksmith, because we provide a virtually infinite pool of compute resources, wait times have dropped from 15-20 minutes to 10-15 seconds, significantly increasing the speed at which their developers can deploy code. The main task for Blacksmith now is to
make sure that any engineering team thinking about CI speed or the trust in the code they create immediately thinks of Blacksmith. To this end, we are launching a large-scale marketing campaign in San Francisco and New York. We cover 15 subway stations in New York City. We are wrapping the subway car in our signature bright yellow color. When I joined, there were about five people on the team. We were running a small cluster of a few hundred servers, and today we have reached a scale that is many times larger. We are now a team of 30 people, 12 of whom are engineers. It comes down to a few key decisions that we have made and continue to make as we work. We have a lot of our own specialized software. One of them is to manage the
growth of our server fleet. We have many customized monitoring solutions at every level of the stack. Second, we ruthlessly limit the product's functionality to keep it compact and manageable, scaling it only to the size of the team, instead of making concessions and fulfilling the unique requests of individual customers. And the big advantage of this approach is that we can scale to a huge number of machines and a very large number of customers, with a very small team. So, one of the biggest challenges is keeping up with demand. We notice that every time a new LLM is released, we see a massive spike in CI usage as people rush to test it and release more code. Since ordering and installing machines takes
time, we had to be technically creative to handle this workload. This means better algorithms for placing our tasks, more inventive approaches to how to group them more efficiently, and more efficient network topologies. Immediately following our Series B round announcement in early summer, a power surge in Phoenix knocked out both the primary and backup cooling systems at one of our largest data centers in that city. Our customers experienced what they call longer startup times for their jobs during this outage period. This highlighted the importance of geographical distribution for Blacksmith.
We want to help developers be confident in their code changes. And as part of that, one of the products we're working on is autonomous quality control, where we collect artifacts like screenshots or video clips that add confidence to changes. Given the amount of code being created, developers won't have time to independently check if everything works. They will look for evidence, and Blacksmith will give them the confidence to merge their code faster. We want to eliminate this "bottleneck" stage and allow developers to bring their software release speed almost to the speed of their thinking.
