Eric Siu weighs the trade-offs between expensive frontier AI models and open-source alternatives like Kime. He argues most companies should skip the all-in bet on frontier models and focus instead on return on investment, local infrastructure costs, and the gap between tool adoption and actual production usage.
These frontier models, they cost a lot of money and they have a lot of guardrails on them. And now people are talking about using Kime because it's open source and it's extremely powerful. People are thinking about how do we think about buying or renting local infrastructure where it's more cost effective and, you know, 80% of your tokens are running on something that is cheaper. These companies are buying a lot of tooling their employees. So you might have Copilot, you might have a bunch of different things that you're paying for. But the reality is adoption rates are really low. And not only that, even if they're adopting it, That's not enough, right? If people are adopting these tools, that's one thing to say that they're using it. But also, what do they actually have in production right now? And more importantly, what do they have in production that's actually driving a return on investment right now? So return on investment, meaning save time, save money, whatever it is exactly. And then maybe you're paying some for subscription and then maybe some you're paying for frontier models, but you don't want to go all frontier right now.
