Nick and Zack from WorkOS @ AIE: building real-world AI tools & running conference workshops

Hi, Ron. We're AI Engineers Europe in London, and I'm joined by Nick and Zach from WorkOS. Do you guys want to tell us what you've been, I know you've been doing a lot of workshops, and you've been very much in the weeds of building actual AI tools that get used in the real world. What are you what are you excited about and sharing with people at the conference?

Yeah. So we we had a great workshop recently. I think the things that are kinda top of mind for me and thing that I did talk about yesterday was, you know, these tools are sort of nuclear now, if you will. Right?

And it's incredible how fast you can go, especially if you have certain kind of dev patterns in your brain, and now you're no longer typing, the agents are doing it for you. And there's like a there's almost a temptation to just continue stacking more and more loops endlessly and go faster and faster and faster. Mhmm. And so what I've been thinking about is like how do we find developer balance where you're still getting all of the key work that you need to get done done, but you're not burning yourself out faster than ever before, essentially.

Is that kind of, like, related to I don't know if you saw Matt Pocock's talk was talking about having, like, bigger chunks of things that have, like, defined kind of inputs, outputs for the airline to use. Is that I I didn't catch that. I was I was thinking of it in terms of just kind of, like, in keeping the human in the loop, but almost like monetizing the self in the sense of the one trick that I use is is to feed my biometric data in through Oura and other things into my LLMs, and so I can actually have the same interface that I use to develop all week long and that is aware of my projects kinda push back on me and say, yeah, that sounds great, but, you know, you also slept three hours last night.

No way. Yeah. So maybe we put a pause on that until tomorrow after you've had a nap. Is it actually is it actually useful?

Will yeah. Claude will push back, and it's enough of a kind of, you know, pushback for me to think about it. I I don't always listen to Claude, but I feel like If you had three hours sleep, you're just like, shut the hell up. Like Yeah.

Know what I've been doing. I know what I'm doing, and I gotta go go faster than ever before, and sometimes I do do that. But so it's an evolving thing, but that's kind of been fun for me to kind of experiment with. Okay.

Well And then also, of course, like, can we get the same amount of quality work done but spend less time at the desk? And that's something I was kinda focusing on in my talk is that, you know, especially with voice first mode, you know, I can hit a 184 words per minute as opposed to 90. So does that afford a two hour walk in the woods or at lunch kind of thing? And I think it does some days for sure, especially because you can still poke especially with remote control now and Claude, you can kinda still review code as it's coming in and still direct your agents to say, that's not quite what I meant and do this piece differently.

So That's amazing. Yeah. So that's something I've thinking about. But we've been having a blast doing workshops.

We put on a really good workshop for the first day too. Yeah. I I'm excited about all the things that he's excited about, but I don't have my biometric data hooked into that. But I'm just excited about the, like, the redefining of this profession.

And there's a lot of pieces to that and a lot of, like, self realizing that's going on, like, with everyone, I think. And there's a lot of fear and uncertainty and doubt and all of that too, but it's really exciting. And it's kinda taking us back to the first principles on what we're actually doing and just allowing us to redefine that, whether it's taking long walks in the woods and and pushing code or just figuring out how to have the symbiotic relationship with the machine now that can kind of reason on its own a little bit more.

Yeah. Maybe this is just a I mean, it's an in somewhat in some ways, it's an arbitrary point in time, but I remember coming to this amazing master workshop that he did at the AI engineers. I think that was That was so that was probably that was more or less a year ago. Right?

June. Yeah. How do you think, like, you're working differently since then? If you could even I don't know.

I impossible questions to remember what you were doing at that time. That was really the the start of this. And in a lot of ways, that workshop like, building that workshop was like, wow. These tools can actually do that because Claude was very instrumental in getting that workshop done on time.

May have may have may have may not have been done in the lobby of the hoop car. Got an hour's. Those are Right. Yeah.

So I could And and that was, like, a real proving point for us because we were like, we did this in such a short amount of time, and the quality didn't didn't reflect that. The body was there. Yeah. And it just helped us to move so much faster and get things laid out in the way the exact way that we wanted Mhmm.

Without having to have all this minutiae going back and forth. Like, the main thing that we did was, like, you know, we did the thing where you have separate branches and get with, like, the different steps that we were on. So if you didn't get caught up on, you know, the the second step Yeah. Skip to the third step by just checking out that branch.

And doing all of that is so tedious, like, when you're building a workshop. And I just, like, described that to Claude, and it did it. And this this isn't Claude four. Right?

This is, like, 3 seven or whatever it was back then. And it did it. And then, like, you know, we changed a bunch of stuff, and it was like, we have to fix all of this. And it just fixed it.

And it just worked every time. And I really left that workshop realizing that I'm not gonna be writing code in Numero. And I don't think that I've written a line of code by myself since then. Yeah.

I changed directory the other day and piped something to p b copy and felt that was, like, the most muscle memory I've used in a really long time. I think to some degree, like, you've seen certain, like, signs of maturation, like the workshop that we did a couple days ago now with skills, like, being able to you know, since that first workshop, being able to talk about skills as being like a defined unit of work that we can develop ourselves, and it's useful even if I'm just a solo dev.

It's especially useful when Nick and I are sharing skills back and forth as we do, and it's now amplifying it across the team when we put it in a in a way, you know, maybe in a repo and then behind a plug in marketplace, as Nick has done, so that the whole team can benefit and share, you know, and that's kind of what our our workshop was about this time. So I I think there's there's, like, signs of, like, the ecosystem maturing and people finding out which primitives are making sense and the fact that Skills kinda started as a a Cloud thing, but just now, you know, was, like, applicable to Codecs and and other systems as well.

Yeah. Yeah. We we both gave talks today separate from the workshop, but I think they tie back to the workshop in that in those talks, we both said, like, that we are the bottleneck in this whole process of software development now. And the the focus on skills in that workshop was really to to just say, like, how do we, you know, remain human in the loop, but not be so much of a bottleneck, and have these reproducible ways of either extracting the information out of us in a way that doesn't feel like it's pulling teeth, and also just keeping it on track with, like, this is exactly how I like to work.

This is what you need to do. This is the the steps that you need to take, and just, like, having a way to codify that so that we can repeat that over and over without having to type it out every single time. Yep. Yeah.

What have been the skills that you have, like, found most useful? Image generation's huge. So that that for us is, you know, not just for mock ups and diagrams and stuff, but for just for fun, you know, like doing pixel art and adding some some color to some writing or a blog post. But Nick's got a terrific ideation skill that we all use where if you're working on something large, let's say some architectural refactor or you're starting a new project, it'll kind of internalize this sort of, like, variable of of certainty and keep on asking you questions to discover and kind of, like, clear up the fog of war until you have a really clear picture, and then it says, okay, now I'm clear enough to begin implementation.

Yeah. Finding that to be super useful. In a lot of ways, it's like superpowers, if you've ever used that one, the brainstorm skill in that. But it has its own internal rubric about what it means to understand what I'm asking for, and it grades that.

So it really takes advantage of having a confidence score on five different points. And if the confidence isn't greater than 95% on what what I'm asking it to do, and then, like, it understanding what I'm actually asking for, and it understanding, is it actually possible? Do I actually understand how to do what you're asking for? Yeah.

If either of those are less than 95%, then it just repeatedly asks you questions over and over to get to that point, and it won't stop. And you find that that's very helpful, I'm guessing. To me, it's like, I can start with a sentence. I wanna do this, and I pass it to the ideation skill.

And then it's like, well, what do you mean by that? And instead of me having to, like, have the foresight of, like, coming to it and describing, you know, a big monologue about the thing I want to do. We would both use Whisper Flow too to just talk to the machine all the time. Instead of me having to come to that, I can just start with a sentence, and then it, like, extracts the information out of me through those questions.

And I just find that such a more pleasurable way to work for me Yeah. Just because, like, in my in my head when I'm starting to do that, I'm just like, I forget, like, everything in that moment. So it's like a reinforcement to just make sure that I am actually giving it the information that makes sense. Mhmm.

And I guess it's just making less assumptions at that point than Yeah. Like removing those. Once once it comes out, it creates a contract, like a human readable contract. Yeah.

And I go through and read that. And if there's any issue, anything that I don't agree with, we just keep iterating over and over. But then once I approve that, then it starts making, like, the specs that it's actually gonna work from, and then it goes from there. And it's all set up to do it in phases so that you can reset the the context and start from scratch each time.

So you're not, you know, starting with 300,000 tokens already consumed and and going from there. You're not gonna face that context drop. Yeah. I'm I'm finding even and there's a skill builder skill built into Claude, and that infrastructure, if you will, is is also incredibly useful.

We talked about scanning your last developer's week worth of messages in Claude Yeah. And looking for patterns and saying, you know, where were the inefficiencies? What what did we struggle to do? What did I have to re describe to you over and over again?

And then kinda plucking those out as discrete units of work so that the system or your harness, if you will, like Mhmm. Gets even better for next week, and you don't have to re explain yourself. It's token saving. It's more efficient, and it can be shared.

Those those units of work can then be published and shared with other team members too. That's amazing. Yeah. It makes makes sense.

And is it just like when you're when you're sharing it with people, is it like how are you doing that? Is it Right now, it's it's a variety of way. I mean, it tends to go in a GitHub repo. There's still awkwardness around secrets and fairing secret, you know, like how do you it's still you sort of roughly follow the 12 factor app pattern, like keeping your secrets on the source code, etcetera.

It could be really helpful to put a plugin marketplace in front as like an interface where everyone can run uniform commands, plugin, add. You know, first to install the marketplace, then you plug in, add the specific skill, and that would lend itself to us, you know, a script that everyone could run, for example. Yeah. To get on the same page and have the same capabilities as well.

Yeah. So Okay. But I expect that's gonna continue. Yeah.

That's really the easiest way right now that we found. It's a little bit harder for, like, our, like, nontechnical colleagues who don't maybe don't have a GitHub account, like, to understand how to do that and, like, pull all of it in, especially, like, in the quad desktop app. It is like, they have a GUI for it. It's pretty simple.

But we've also you know, for them, they're doing a lot of just, like, create skill, and it creates a dot skill file in quad desktop, and they drag and drop that around. So we need a better way for that for sure. Yeah. Okay.

I have a question, which is probably a bit of a tee up question, but it's actually, like, something that I'm genuinely interested in because I'm in my job, I'm working on a lot of the, like, MTP stuff, and, you know, part of that is, like, authentication for the users, you know, say, be it Gmail or whatever. How are you seeing people, you know, like, using how is, like, off evolving with agents and with, you know, you granting your users needing to, like, manage their own auth with your own with their their apps with within your app and stuff?

And There's a really wide gamut right now everywhere from Yeah. People that are insanely experimenting like OpenClaw, such as myself, you know, granting certain access to your Gmail account and CLI tools and saying, you know, go because I don't really care, all the way up to really, you know, mature, like, machine to machine systems that are kinda coming on the scene. That's something that we're working on at WorkOS. Probably have more to share eventually.

Yeah. It's a problem we're really looking to solve, like agent identity and how how that goes. And we have all of the pieces there. It's just how do we assemble them in a way that works best for agents and users Yeah.

And puts it together. But in a practical sense, something I've been experimenting with PIE Yeah. Kinda switching over to that. And I realized that there's something that I greatly miss in PIE compared to Claude, which Claude, just got switched over to, like, the enterprise Claude plan through work.

And through that, like, it's now connected to the desktop app. And I I think somebody in our IT department, like, set up all these connectors, like, within that. So they're like Slack, and Linear, and Notion, and all of these different connectors. And when they're authenticated in the desktop app, I can just, like, paste a Slack thread into Cloud Code, and now it can read that thread, and it knows where to go.

Or it can search Notion, or it can do any of that, and it just has all of that full context. And honestly, once I realized that, it made it feel like magic, and it makes me desperately miss it when I'm in PIE. Yeah. I I will puss on that.

That was something I I mentioned in my talk is that I I was kind of trying to look backwards retrospectively to see what are the things that caused me the worst context switch penalty. Yeah. And for me, it was Slack because there's always something interesting or different conversation going on, and but that's that's also where asks come in or feedback and bug reports. Right?

And then I already had Claude able to read and write to my linear, which I find very useful because you can kinda put subtasks for bugs and also feature development that you want. But once I got Slack connected, the loop kinda became complete, and I could remain in focus mode in Claude and do feature development, and then say, you know, kinda you can loop and look for, you know, incoming asks on Slack every ten, fifteen minutes, and then if that's not already a linear ticket, you just make it for me, but we're not being pulled off task every time, and I don't have to alt tab, you know, 30 times to get one ticket fixed.

Yeah. Not to mention that the thing I was working on had a a Slack bot component to it, and so when I said this is a bug I want you to fix and verify yourself, Claude now had the complete tool loop that it needed to go all the way through, post its own test message that had triggered that bug, and then verify by reading the Oh, because I did. Yeah. Because did.

Then mark the linear to get done, and then I'm like, Now we can move on to the next deal. Right. That's that's wild. Yeah.

That was a big, big unlock, and it's it could be as little as two connectors that people need to add, to get to that loop. Mhmm. Yeah. Are you seeing anything with, like, permissions and, like, g like, say, Gmail, for instance, like, how people are giving, you know, permissions because, you know, sending an email could be really destructive.

And then, obviously, like, when using, like, these MTPs, it could be, like, 800 different Gmail tools and, like, you might have, like, multiple apps. Like, are you seeing any, like, ways that people are, like, managing all that stuff? I mean, yeah, I I do it myself. I do it the dangerous way to just to experience it and live it.

Oh, yeah. Yeah. But to to kinda get that scar tissue, but at the same time, it's possible now that, you know, to install AuthKit and MCP server and get really granular about this tool needs permission, this tool doesn't, like the way that you code it. I think that the the DevX there has, you know, miles to go still, but it's something that we're actively working on.

Yeah. And we do have this this Pipes MCP tool. Yeah. I saw that.

That really, like, helps with that, and it gives you, like, these short lived sessions short lived tokens that you can use to get third party data into your apps and stuff and do it in a more secure way. And that's something that we're that's out now and that we're actively working on to improve and Yep. And adding new connectors too. So I think that'll be part of it as well.

And I think when that kinda comes online with another few pieces, then eventually, we'll have a kinda coherent experience where you develop from an agent first perspective. Yeah. Super cool. That's I I need that.

Okay. And then we're getting on to, like, the more, like, we're all, you know, trying to reach developers. Right? And workshops, I think, is something that you guys do a great job at.

If someone's thinking of running a a well, I guess, like, maybe someone doesn't even know what, like, a workshop really means here, but, like, could you tell us, like, how you think about it and how you Sure. What makes it good? Yeah. Absolutely.

Yeah. The main thing is, like, focus on the content. What are you trying to teach? Tell a complete story in that, and, like, the narrative is driven through those stories.

And that's how people are gonna remember it and how they're gonna have fun with it. And you can have a lot of fun in that experience. One thing that we learned from the the workshop that you went to with on Mastra last year versus the one that we gave this year is, like, focusing on, like, the actual output of it. So, like, that was a big benefit for 80 people in that room Yeah.

Last year, but it was also recorded and put on YouTube. And we had several times where we just had, like, let's go work for twenty minutes. Dead air. Just dead air.

But also, like, we had we had a the very fortunate, like, experience to have Sam Bagua in the audience Yeah. The creator of Astro. And then giving out books as well. Which we did they would probably arrange for it.

Yeah. It of happened that way. Yeah. And it it was so wonderful, but he wasn't mic'd.

And so All his founder answered. He was not Yeah. Accurate. And so we really focused on that this time.

Like, not just like for the people in the room, but for the video experience as well. And so we designed it so that we still have those working things because we think that it's important for you to experience things and, like, actually get hands on keyboard and try things out. But at the same time, like on that slide where it told you what to do, on the other side, it had a list of topics that Zach and I can just ramble about. But they were not necessarily like specific to that section of the workshop, but it's something that we're passionate about.

We can talk about this, or we can answer your questions. And the first time we did that, like the first break we got to, we did nothing but answer questions. And it was honestly great because it was just all audience led. Wow.

We passed mics around, we got it all recorded, and it was fantastic. So I'm really excited to see the video when it comes out and see how just how much better it is as an experience to watch after the fact. Yep. Also recommend, you know, both times we've provided GitHub resources in advance, like we make a Remote that guys with it.

It's free. It's open. It's open source. And that way, if you're, you know, really champing at the bit, you can kinda download that first and and get started, but also something super cool that Nick built into both of them using Claude is the concept of like checkpoints.

So if you Yeah. Got here late, you can checkpoint to where we are now at three. If you've gotten behind and got stuck on something and now where you want to see the end experience, you jump ahead to the 75. But it's also yet another resource that you get, and it shows like this is an actual application Yeah.

That you can run, and it's like we're not just kind of hand waving like this actually works. You can run it yourself if you don't believe us. So that that's been useful too. And then do stuff that you're actually passionate about that you like to build Yeah.

Work with. I always come in now I I think now I'm gonna take it as like a positive sign going forward that I come in thinking like, oh, no one's gonna find this interesting. This is art. This is like a year ago stuff.

And people are just eat it eat it up and say this is amazing, and it's like super helpful. So it's because we've spent every day playing with that stuff that Yeah. That's just Yeah. It's like a genuine intro.

Yeah. And and people are at different levels, and you just have to realize that. You're always gonna know more than some people and less than others. For sure.

So there's always like value in adding content at all layers of that spectrum. And I think that another thing that, like, is a really, like, good final point to that too is to make it fun and interactive. And so for this one, we had it was on skills. And I thought, oh, how fun would it be if we could, like, show the skills that people in the workshop are actually building?

Oh, yeah. And so, just like, very quickly with Claude, I built, like, a little worker that and a Bash script, and anyone in the workshop could run it, and it would just immediately package like, zip up their their script and put it into a Cloudflare KV, and then I could run another script on mine and just pull it down with their name and everything Oh, cool. And then immediately run it against my repo and test it out. Yeah.

And A roast your Repo tool. Yeah. Yeah. It was a Roaster Repo.

We did that live in the middle of the the workshop. Someone had put put one up, and then Nick pulled it down, and then used his Mhmm. To roast our own repo. Yeah.

Oh, yeah. Add all of his personality and and all of his, like, everything he injected in it. We got a 60%. Oh, yeah.

That's tight. Okay. Well, I think, yeah. But I think also one of the validating things that goes against that 60% is that one person in the audience didn't do the workshop at all.

They actually took a picture of the slide that we had with the QR code on it and sent it to their Hermes bot. And the Hermes bot did the whole workshop and submitted suggestion like, submitted examples to that tool, and we were pulling them down and running them, and he came up after the fact and told us and showed us the whole transcript with Hermes where it did everything. Really? Yeah.

It was good. It was good. Yeah. So I highly recommend the the QR code and repo pattern Yeah.

If you're doing a workshop, and and do something that you've been kinda hacking on for a while that you've kind of have actual hands on experience with. Yeah. I remember even the master one. I can't like, my brain's failing me, but there I remember, like, people everyone was posting in Slack.

Yeah. It was a meme generator. Meme generator. It was actually it actually had the concept of slots.

Yeah. It the memes that you you put out were it was not just the image, but it was multiple slots, and so you could actually get pretty complex memes, and you could describe in one word what you wanted. Yeah. But the cool thing was that it was using the JenticTools calls to to flow through multiple steps to get that end result.

Yeah. So that's the thing where it's like, you you think you're having fun, we're just joking around and roasting, but it's like, but you spent the afternoon actually pounding and sharing skills. It it was a it was a workplace venter Right. Is the idea.

Anyway, you just Yeah. Vent your day and it would make the perfect meme for Place. Yep. Yes.

I think I remember making one about the Wi Fi. Yeah. Yeah. Indeed.

That's all yeah. The bane of all conferences. Yeah. Think one day we're all gonna walk around with just hard hard line Ethernet.

Yeah. Yeah. Yeah. We'll go back.

It's gonna be yeah. Yeah. Right? That would be great.

We'll just have, like, backpacks. Yep. Starlinks on them. Yep.

Yeah. Okay. So that workshop, I don't know. Like and maybe this one too.

Like, you know, you guys work for WorkOS. Like, the workshop was about Maestra. There was like yeah. It was like it was a very, like, you know, strong like, because as an attendee, I thought it was amazing because it was like they genuinely wanna, like, teach you something, and it's great.

But I love to hear, like, the thought around that because that's quite an unconventional thing to do. Most people, you go to the workshop, and it's like, alright. I work for x y z. Here's how to use x y z with, your b o two hour app.

Nice. End of the yeah. Yeah. So allergic to that.

He's my Same here. I feel like do it. Yeah. Can't do it.

And, you know, we we're doing a lot of really cool things. Like, the the skills one that we did this time is just built out of, like, the day to day work that Zach and I do and, like, everyone at WorkOS does. Yeah. And so, like, it's really just us sharing our knowledge and our excitement and our passion about it.

And the tools and our products speak for themselves. And, you know, we have we had one slide in there that just showed the WorkOS logo, and we've we mentioned we're hiring and everything like that Yeah. Yeah. Yeah.

Of course. But, you know, then we get to just move on and have fun doing the stuff that we love as developers and teaching. And Zach and I are really fond of of doing that and teaching to people. And so we focus on that, and we wanna have get that value out of it without it feeling like we're just selling to you the whole time.

Yeah. 100%. Yeah. Yeah.

We get a lot of good leeway for that at WorkOS. Like, we're trusted to kind of just know and design the workshop and often on the on the plane ride over. But, you know, it comes out well because it's coming from a place of pure developer joy. Yeah.

Yeah. I always like when with guests, I always say to people, you know, I think if if you're saying useful things, people are gonna be interested in you and then wonder, like, oh, where does this? What are they building? Where do they work?

Like And then the same way. And then but if you just start start with, like, we're the greatest way to do, it's like, no one's gonna really care anyway at that point. Yeah. And it's like they might be like, oh, that they're a bit salesy.

Like, I don't Oh, yeah. And, hey, it worked. A year later, you're asking us to do Well, it's like, well, say that developers are allergic to marketing, like Yeah. Developers our whole lives, you know?

And so that's like I I almost think that we can't make that kind of workshop because it's like, wouldn't attend that Yeah. Yeah. Shop. So Yeah.

My brain would turn off. Yeah. Yeah. I think it's great.

I think more people should do that. Just be useful, like, regardless, and then I think people will care about what you're doing. For sure. Yeah.

That's amazing. Okay. So I guess you haven't been able to go to as many talks as possible there. Is there anything that you're feeling, like, really excited about at the moment in the kind of, like, AI engineering space?

Harness engineering has been a fascination for me for the last couple of months. I've been working on my own, and just, like, the idea of I'm gonna build this thing that is trying to be me and do the work as I would do it. And specifically, like, it focuses on two things that I think are so underutilized and so important to focus on. And that's, like, prove to me without me having to look at the code that you did the work that I expected to.

So really, like, focus around evidence, but then also do retrospectives on yourself. Have memory. Understand what you did right, what you did wrong. Did I make too many tool calls, like, that are the same without changing anything?

Why did I do that? Let's fix that. And write that down. Like, have the tool.

Do that. We like, my tool does a retrospective after everything that it does, and then creates memory files for that. And then it prunes it as well, so it'll clean it up as things become less relevant. But it stays focused, and that means that every single time I use it, it's getting better.

And I can also guide it. Yeah. Similar for me. I mean, it's kind of the stuff I'm hacking on is OpenCLaw.

I'm gonna experiment with Hermes next. Most recent upgrade I gave to my OpenCLaw was a HackRF one radio that I software defined radio that I plugged into the same system. And then within maybe a thirty minute chat, OpenCLaw had built me, like, OpenRTX so that I could log in over a tail scale service on my phone and see and tune the radio and, like, pick up, like, ATC and pick up, like, pilots talking. And I wanna I wanna I because I find radio waves fascinating, I know nothing about them, and so I wanted to have, like, a sort of substrate to go and work on that.

And then I wanna figure out, are there algorithms that we could either use or design to, like, do scanning and pick up, like, these little ones that are active and, like, this is what people are saying kind of thing. But that's but the same way, I I I find the value in having that that system is just being able to talk into you know, I'm a traffic light. I've got two kids in the back of the car, and I'm like, that's the idea for how I'm gonna solve that, or that's the idea for how I tie up blog posts together, and I just fire off a thirty second voice message to OpenClaw.

And then by the time I can actually sit down and work work on it again at 9PM that night, the context has been preserved, it's already in one place. So I could just get that artifact out. So that's the value that I'm finding there, and I think that that's gonna be I mean, probably next year, our agents will be able to have some secure side channel where they can pass information back and forth or something. You know?

Next year, we're gonna come on, and Zach's gonna tell you that Yeah. The perfect time to to plan your work for the day is when your heart rate's at 78 beats per minute, and there's an AC one thirty, light Yeah. That's because the the frequency is yeah. It's not interfering with your Wi Fi at all.

It is now it's only time to push to get up. Yeah. That would be weird if you imagine that, like, when a plane was above you, you were, like, hyper productive or something. That would be weird.

You realize that Yeah. That Things you don't wanna know. Weird stuff. Yeah.

Maybe there's there's there's such a thing as too much knowledge. Yeah. Yeah. I did think it would be really interesting to make like a almost like a home brewed citizen app where you're just like picking up enough frequency to tell you, here's what's actually going on around you.

Right? Yeah. Yeah. The the real question is, do I want that information?

Yeah. Are or not? That's true. What's better for mental health?

But, yeah, I think it's gonna be agents, and I think it's gonna be more insane tool calls. Yeah. Okay. Amazing.

Well, thank you so much, guys. Thought it was really fun. Yeah. This is a blast.

Yeah. Thanks everyone for listening.