"If you're an AI startup not using graphs in some form, you're building in technical debt" Alison Cossette's hot take on graph databases. In a recent conversation with Eli Chen, the two discuss how…

51,325 followers

1mo

"If you're an AI startup not using graphs in some form, you're building in technical debt" Alison Cossette's hot take on graph databases. In a recent conversation with Eli Chen, the two discuss how AI systems are naturally networks. The behaviors of agents, tools, and software interactions are inherently graph-like structures. But as the conversation develops, this isn't about using graphs everywhere - it's about having the right tool in the right part of your architecture. Eli highlights the real challenge for technical leaders: avoiding common failure patterns: - Choosing something because it's quick and easy (then living with the consequences) - Copying what others are doing without understanding if it fits your use case - Sticking with the familiar approach too long when a different tool would be simpler This applies beyond databases - whether you're deciding between fine-tuning versus prompt engineering, or any architectural choice. The key is thoughtful decision-making rather than following trends or taking shortcuts. As engineers, we need to think like architects: understand the problem first, then choose the tool that best solves it. Sometimes that means going against the grain if it leads to a simpler, more maintainable solution.

Transcript

Though we were chitchatting the other day and I and I said to you, I said, if you're a startup and you're in AI and you're not using a graph in some form, you're innately building in technical debt. And it's definitely true. Because when we think about our AI systems or our genetic system, I mean it, it is naturally a network, right? The behaviors of the agents, the behaviors of the tools and the software and how it's working is innately graph. And so to be able to even understand our systems, we need that. Graph. So to your point, not everything is a graph. Graph isn't the right answer for everything, but it's a matter of the right tool and the right database in the right portion of your architecture. I love that. And I would say that's actually probably the primary role of an architect, which is to use the right tool for the right job. And and some of failure patterns I've seen is to do something because it's quick and easy or because. Uh, it's just other people say they're using it, so you kind of copy their pattern. It applies to when you're fine tuning a model versus, you know, prompt engineering. It goes both ways, right? Like sometimes you're doing something because it's easy for too long or you're not. Doing something? Even that that goes against the grain, but it could be much simpler. So I think that is a good kind of description of graph thinking, which is it's a, it's another tool in the toolbox and.

To view or add a comment, sign in

LinkedIn respects your privacy

Explore content categories