AI agents are knocking. Is your API ready to answer? | GraphQL’s Matt DeBergalis

**Andrew Zigler** (0:06)
Welcome to Dev Interrupted. I'm your host, Andrew Zigler. And joining me for this week's news is past Dev Interrupted guest and my friend at so many conferences on the ground, Andrew Boyagi, customer CTO at Atlassian. And congratulations, by the way, Andrew on your recent promotion. It's really well deserved, and I appreciate you taking time out of that to fill in for Ben this week.

**Andrew Boyagi** (0:29)
Thanks, Andrew, you're really good to be talking to you.

**Andrew Zigler** (0:32)
Amazing. Well, I know you've had a lot of exciting stuff going on in your world, and we're going to touch on that a little bit today in our new segment. But for those listening, as you know, we kick off every episode by going through the weekly news and the tech world, which gets weirder and weirder every week that we talk about it, it seems. So we picked out some fun stories for y'all. This week, we're covering the recent Cloud Code usage cap and what it really tells us about developer and user consumption of these tools. We also get a glimpse at OpenAI's new study mode, which could be a really interesting new utility for those using ChatGPT to learn, as well as a new wave of age verification requirements hitting the web. And personally, how I see a lot of parallels to that and things happening in our own engineering stacks. And last but not least, we're going to touch again on the recent developer survey from Atlassian. I know we've mentioned it here on Dev Interrupted, but we have Andrew here today, and he's actually one of the voices behind this survey. So we're going to get in a little deeper and learn some more. But Andrew, I'm going to make you work for it. Let's start on the opposite end of the track and work our way towards Atlassian. I'd love to get your thoughts on some of the things that we're seeing, because it's been an interesting week. So I'm going to start us off with Anthropic. Does that sound good with you?

**Andrew Boyagi** (1:48)
Sounds good.

**Andrew Zigler** (1:49)
Okay, perfect.
So, you may have heard in the last week that Anthropic is capping the usage of Cloud Code. And to anyone else tracking like metered usage and subscription services of these tools, this is just another pricing change dropping in the bucket. We've been seeing all of these providers and tools experiment with new pricing models, put things behind very high paywalls. But this one's a little different, because as part of this revelation, many took to the internet to express their frustration with those who were perhaps using Cloud Code in extreme ways, querying hundreds or thousands of agents to do all sorts of things and perhaps ruining the fun for everyone else. But you know, it's not really quite ruining the fun as experimenting, because my friend Joffrey Huntley of Sourcegraph's AMP and a thought leader in the agentic coding space whose work we've shared on the podcast, you know, he's one of these power users, in fact, that got ranked on reddit in the past week. And you got to see it to believe it. So we're going to include the link to this post in the newsletter. But he tops the chart on the quad code consumption leaderboards with a purported $450,000 in inference. And that's spent on Anthrophic's Dime, way beyond his subscription cost, of course. And you know, you might be wondering, like many scratching your head on reddit, what is Joffrey even building? And well, it's a programming language, in fact. And if you haven't checked it out, it's pretty cool. It's C-compatible. It's called Curses. It uses Gen Z slang and logo for all of the namespaces. And he livestreams for agents building it in loop while he's totally AFK. So this is a really interesting usage of coding tools and an expensive glimpse at the future maybe. So, you know, Andrew, what do you think of this wild usage of coding tools? And what do you think it means for how developers are experimenting?

**Andrew Boyagi** (3:46)
Yeah, I mean, like, that's probably not a standard use case of using Claude. So I wouldn't be expecting the average person that is spending half a million dollars on building something with Claude. But you're right. I mean, like we're in an experimental phase. Like if you think about where we are in AI and its adoption with developers, we're probably still at the beginning. It's only been a few years. So I'd expect pricing changes in different models in different ways. Obviously, a lot of companies are still adjusting to usage patterns and they're learning as much as the users are learning on how to use it. So hopefully, it's not going to become this really expensive thing that puts it out of reach for most people. But it shouldn't be unexpected that we're seeing the pricing changes and different ways of consuming AI.