90% cheaper tokens sounds like a pricing story. I think it’s actually an infrastructure story. Anthropic just launched Haiku 5.5 at 90% lower per-token pricing for prompts under 100K tokens. But Anthropic also says the average cost reduction is closer to 75%, partly because the new tokenizer uses more tokens to complete the same work. That distinction matters. We’re entering an era where the economics of AI are increasingly defined by tokenomics. The question isn’t simply which model is smartest or which API is cheapest. It’s how many tokens your application needs to accomplish a task, how often it needs to call a model, how much context it carries forward, and how much work it can now afford to delegate to an agent. And when tokens get cheaper, usage tends to go up. That means more agents. More generated code. More autonomous workflows. More software dependencies. More opportunities for vulnerable code to make its way into production. For security teams, this creates an interesting second-order effect. The cost of AI is falling faster than the cost of managing the complexity AI creates. We’ve spent a lot of time asking whether AI can write code well enough to replace parts of the development process. I think the more important question is becoming whether our security infrastructure can keep up with the amount of code AI makes economically possible to produce. Tokenomics is going to be a security conversation, whether security teams are ready for it or not.