Is this what you saw too, um, Ari?

Um, yeah, I would agree with what John said.

I mean, if we look at the way that Akamai's always thought about computing, um, from the very beginning, you know, creating the CDN gave us an opportunity to start playing around with ways that we could deliver code and not just content at the edge.

Security gave us a way to do that, to keep applications and endpoints safe, and we extended that when we started to really evolve Cloudlets into serverless, this lightweight language that really any developer could code against and use the big breadth that Akamai had established to very easily change the way their application worked, customize the results that a user would get, or otherwise optimize the experience.

And as we started to realize how much our customers wanted to do and the art of the possible in working with them, we said we've got to add infrastructure as a service, but true infrastructure that you could address, that you could build on, that you could really think about with the, the breadth and the scale that Akamai has as a, a distributed network that had all of the fittings of a cloud except for the infrastructure.

So we added the infrastructure to the scale and the breadth and the reliability and the security.

And then as we started to see ChatGPT giving way to changing the way that people were working and, and companies were defining, um, productivity, we started to realize that we could use our edge for more than just delivering an application, but it's really how you would build it, what powers the application that you're gonna be delivering, and then how you would protect it on the other side.

How are you thinking about that?

Now, when we look at, let's say, whether it's agentic AI systems or how RAG models itself, right, how does that change the economics of inference when you're talking about agentic AI systems and RAGs?

Yeah, it's really interesting.

I mean, if we think about what's in the news cycle right now, we keep hearing about these trillion weight models that are coming out and how agents can spawn other agents and solve problems a lot faster.

But when it comes time for an enterprise or a business user to think about building an application that needs to scale to thousands or millions of users.

you start to really get sensitive to the economics of how many tokens you're going to be consuming.

And it just seems like every time we get a new model, it could be trained maybe on less GPU infrastructure and less overall capital, but it's consuming more tokens all the time.

And we hear that from our customers where they engage with some of these frontier LLM providers and they're told about the efficiency gains that are possible, but all they see is more and more tokens consumed for whatever they do.

Now, one of the things that you talked about is design considerations, right? When we look at why data movement, egress, and latencies are becoming a major design consideration, what was your rationale behind writing that?