Rita Kozlov
speaker
102 appearances
1 recordings
1 series
first heard Oct 2024
last heard Oct 2024
Rita Kozlov’s voice in public audio — every appearance, attributed to the second.
Trend
recordings per month · last 12 monthsNo recordings in the last 12 months.Older appearances are listed below; set an alert to hear about the next one.
Appearances
And you're like, oh, you know, in retrospect, that was a mistake. Or I'm so glad we did that versus, you know, because it really does take a long time for these things to play out. Like in one or two years, you only see the beginnings, if any, of the ramifications of the different choices. And yeah, I think it's always there's a quote for this or something like that.
But anyway, yeah, you get less done, I think, in a year than you set out to. But at the same time, you get more done, I think, in five years than you would have thought you would. So when I look at how much is available on the platform now in terms of, you know, like primitives, that blows my mind. I don't think I would have predicted that we would at this point have
object storage, queues, workflows, logs, like there are so many AI, there are so many products. But then there are other things that, you know, one of the first PRDs that I worked on at some point was gradual rollouts. I think that was one of the first things that Kenton and I talked about.
I would talk to a customer and they'd be like, you know, it's super cool that I can push out a change and it's live to your network of 200 data centers in less than 30 seconds. But that also terrifies me. Can I please have a lover by which it happens over time? And there's this weird thing that happens where you're like, yes, that makes a ton of sense. Let's start building it.
And then a ton of stuff gets in the way where you're like, OK, but we really need to do this first. priority and this is another priority and all of these things kind of come up and there's unexpected complexity because this is another thing about, you know, building something for a long time is, you know, as you grow, you get more teams, there's more interdependencies.
So all of a sudden you realize like, okay, we built, you know, the like that part, but we need this other part and we need to connect these dots. And so, yeah, certain things definitely take longer than you expect them to. And then other things, A year ago, you were asking about how these innovation weeks come together. And, you know, probably like a month and a half, two months before that.
Last year's birthday week, we're like, we have to get AI running on Cloudflare. And we'd announced Constellation previously, which was the original iteration of it. And that was where we got a lot of feedback that was like, this is cool, but I need GPUs. And we went from idea to...
ai running on gpus all over our edge in less than two months so that's yeah kind of thing where you would expect it to take longer and you know that's the magic sometimes of having a deadline or forcing function is we're just like we just don't have the time for it to take a long time so it's just going to happen real fast
Yeah, I think to your point about having primitives, right, that's where, I mean, what allows us to sometimes get the stuff running so quickly is we build on workers. A lot of our new announcements are built on workers. So at the very least, getting to a POC can happen really quickly. Workers, AI...
is actually largely runs on something that we talked about during birthday week, runs on containers. This is a big conversation that we've been having internally about, is that something that we expose to customers or not? The fact that we've had this platform for some time and we originally built it for Cloudflare runs a browser isolation product as a part of our zero trust suite.
And you need a container for that. So we've had that platform for a while. And the fact that, you know, that was something that we had and we had a team that was ready to go and build on it made it that much, you know, the cold start was much faster on that.
I'm curious to actually hear more from you about what your experience has been And, you know, to your point of on one hand, it unlocks so many things and I think really changes people's perception of what's possible. On the other hand, yeah, it was really interesting. You know, I think we were all wondering how that announcement was going to go and what people's takeaways were.
And the takeaway surprised me. A little bit, or maybe surprise is the wrong word. But yeah, I'm very curious to see what happens when we actually put this in people's hands. And so we'd love to hear about your experience of making that shift from serverless to containers. And did you see things that surprised you? Are people using it in the way that you expected?
Are people doing things that you're like, you know, this is... Not how I would advise you to do it, but knock yourself out.
I mean, people mixing all that stuff together, I think, is the really interesting part. And that's what we hope people will take away from this is that, yeah, you now have another tool in your arsenal that allows you to complete the picture or, you know, yeah, you don't have to... force everything into the same model.
But at the same time, you know, the thing that I thought was kind of interesting is I really think people see the word containers and stop reading. Which is not too surprising, I suppose, in some ways. There are literally some things like... I saw questions when we announced this that are like, does this run on Firecracker? And it, there's literally a subset that's like that.
Having been doing this for seven years, one thing that you realize is distributed computing is really hard. I mean, computing actually period is really hard, right? That's the challenge that most people have with applications. And if containers were perfect, then what would we even be doing in the first place?
Because, you know, people have Kubernetes, like, why wouldn't they just happily be running that? And I think the reality is, There are a lot of challenges that come with that, both from if you have a distributed audience standpoint, which most people do, from a scaling standpoint and how many resources it takes.
And, you know, I think that that's kind of been the longtime argument about serverless versus not is... Yes, you can run your $5 VPS, but most things scale beyond that. And then do you want to spend your team's time on that or do you want to offload that to someone else? And a lot of what I think historically has been done with serverless and where we saw the opportunity with workers was that
If you try to build serverless on top of containers, you end up with something that's not really all that serverless because it's an abstraction and it gets leaked really quickly. You're still dealing with regions. You're still dealing with concurrency. You're still paying for idle time. And workers flipped all of that on its head. And so...
Showing 41–60 of 102 · page 3 of 6
← Previous
Next →