As CEO of Pydantic, the company behind one of the most widely used frameworks in AI development, Samuel Colvin is at the heart of the action.
Pydantic works closely with cutting-edge frontier model labs and AI developers, giving you a front-row seat to the rapid evolution of models, agents, and coding tools. Investors are also paying attention. Silicon Valley powerhouse Sequoia Capital led Pidantic’s latest funding round of $12.5 million.
In a recent interview, Colvin talked about where AI development is going next and what it means for us. This Q&A has been edited for length and clarity.
Q: You work with major frontier labs like Anthropic and OpenAI. How do you think their strategy is evolving?
Samuel Colvin: “A year ago, all they cared about was revenue. So anything they could do to take advantage of our reasoning is great. Now both companies are IPO And if you want profit margins, what you don’t want is to compete only on the quality of the model, because at that point you have to spend a lot of money to train the best model, and then you have to provide inferences about that model as cheaply as possible, so they’re doing their best to lock people in there and it has nothing to do with the quality of the model. I think of it as the source of codes and codexes and all work.”
Q: How will OpenAI and Anthropic go beyond model performance as a path to success?
Samuel Colvin: “It’s obvious why there’s a discount on Codex and Claude Code ($200 a month subscription when you’re actually spending probably thousands of dollars on inference for that subscription). They’re trying to increase their market share. They’re trying to get as much usage as possible. But maybe there’s something deeper than what they’re trying to do. They’re trying to get their customers to use these huge code bases, which are basically written in AI. You get to the point where it’s impossible for you to maintain them as a human. You used AI to generate 20,000 lines of code overnight, so you can use the model to fix it, but you can’t maintain that code as a human.”
With such a huge code base, enterprise customers should stick with Anthropic and OpenAI’s AI coding services. And once that usage is fixed, those companies will likely raise prices, Colvin said.
Q: How is the coding provided by OpenAI and Anthropic changing?
Samuel Colvin: “My personal guess is that pretty soon they’ll say, ‘When you use us through an enterprise subscription, we’re not only doing the coding generation and the coding agent work, but we’re also storing a trace or trajectory of the complete exchange between you and the model as you write the code.’ And for any line of code in your code base, you’ll have a database that you can search for the intent of when that code was created. The argument will be that the agent will be even better, and they’ll probably say, “We’ll give it to you for free, but you can’t export it.” That means your entire business is tied to who you use to do that. ”
Q: Please explain this a little more. How does this work?
Samuel Colvin: “Imagine there’s a software bug. Some line of code exhibits strange behavior. Maybe the developer, human, or AI who wrote that line of code has left some helpful explanatory comments about what’s going on there. There might be too many comments. Five lines of comments for each line of code makes it unreadable. But if you click on that line of code, your colleague AI What if you could see the complete interaction you had with your model, along with all your code? You’d have a complete explanation, including inferences from the model and input from humans, giving you a deeper understanding of the intent behind your code base, knowing what someone was trying to do when they wrote the code, and lowering your risk by determining whether it’s really a bug or the intended behavior.
“I think this idea of basically ‘we store your trajectory and provide a database of your trajectory’ is appealing and valuable. Those two aren’t necessarily the same, but on this occasion, it’s not only appealing but actually valuable.”
Sign up for BI’s Tech Memo newsletter here. Please contact us by email. abarr@businessinsider.com.