# 🐣│cohort-voice
- Session: 9058b934-6970-4aaa-8895-ba4b3a7d3466
- Channel: Discord #🐣│cohort-voice
- Started: 2026-06-23T17:00:10.802Z
- Ended: 2026-06-23T17:33:30.941Z
- Participants: duckanbro, 0xHunter, Aphilos • Pharo, chickenparm, ECWireless, louchi, samkuhlmann
[00:00:00] [duckanbro]: Do I sound okay on my aircoin or anything? Oh cool, alright. Jake, thanks for joining us. Cool, we just uh had a session with Hunter and um now we're just I've I kind of told you about this, but we're we're doing these fireside chats and kind of drove a whole amount to be invited people on to share kind of how they're using AI with um get their perspective on it and getting people from different industries and uh it's a little bit of market research, but also just kind of my own curiosity and um the team's curiosity about how people are adapting, right? So much. Um industry, not just recorders. And um so uh Jake, thank you for hopping on. Do you we're uh hacker one, right? Um do you mind introducing yourself? Uh oh, and I am we're recording this audio for the uh transcripts? Yeah. Um we'll be kind of like ingesting the audio and and pulling uh things out of it to create some content and so you can create a lot larger research kind of document. Um if there is anything that you want to keep private or um it's just don't you? That's awesome, thanks. Uh yeah, and actually this kind of perspective is really exactly what we want. Um we've talked to a lot of coders too and so we've gotten that perspective, so it's safe to hear other side of the of the business. Um let's not see other guys coming in, um which was really interesting. But i talked a little bit about hacker one. Yeah, I I'm imagining like that we're just getting a flood, so if you be inline support and customer experience and um what are you guys working on internally to kind of manage that stuff? Well I mean that seems like a very um impressive system already. Have you uh did you put that together with Cloud Code or was it is this just kind of bolting on the account system? Things like scores and uh doing like QA and human interview or having automations of work closer and what the f that's interesting. Um probably I'll sort of back on that. The uh the one thing I'm kinda curious about for using channel tools that you guys are able to set up so quickly, do you have so you're usually up in your structure team that was managing too much and and things like that? Is that all kind of are you able to f put directly into that now? Is it all internal? Are you guys using external uh tools?
[00:00:01] [chickenparm]: All clear. Yeah, function. Yeah, for sure. Okay, just out of curiosity, is any is this going to be posted publicly anywhere? I'm just trying to figure out how much I can say if I'm just not gonna share any vulnerability data or anything, but I'll do uh I'll do a quick uh intro. So uh my name's Jake. I live in Saliva, Colorado, not too far from the beacon. Um I run the customer experience department at Hacker One, and I have very recently uh dove head into Cloud Code uh and I've been doing a ton with it and it's kind of changed my life and the way I work. So I think Dick and I were talking about this in person. He was like, hey, this will be a great opportunity for you to jump on, so that's why I'm here. Um and also uh as far as like coding and whatnot goes, not a developer by trade. I'm definitely more on the business side, customer experience side guy. So uh a little bit of technical skills from working at Apple for a little bit. I was a genius there, so I didn't like uh hardware uh repairs. Um but the whole development world is is somewhat new to me. So um that's it's mean in much. Yeah, so Hacker One is a unbound platform. Uh there's a bunch of them at the link where we have this one, uh we've been around for the longest. Uh really uh the way it works is you have uh, you know, any company can can join with us, uh putting together uh whether it's a VDP, a vulnerability disclosure program which doesn't pay out for findings, or you could have a bug bounty program which does pay out for findings. Um we had uh you know we have me and uh a company to likely a customer or habit in one of ours, uh with the US Department of Defense. Um you know, just talking about bounties and whatnot for anyone that's not familiar to give like a perspective of how much you can make. And um, I believe the largest payout on our platform were from Coinbase, uh and they paid out a million dollars for their own vulnerability, and that was like obviously a huge one. Um so yeah, you you kind of set up your own like page like a landing page, uh you have a scope so you make it really clear what domains or whatever, you know, you can do hardware testing as well, but uh typically it's customers who upload domains. Um and the idea is it's not just penetration testing, it's you know, actual hackers that really want to break shit, uh, that also want to make money. Uh and so like the top hackers on our platform, even the mid-level ones, a lot of them this is their career, so they're just hacking on on programs and they are making a lot of money. I always make a joke when I see payouts go out. I wish I was technical, I wish I understood that because there is a huge opportunity to make money. Um and then I'll just say kind of what we're seeing in our space recently with AI is um these kind of a genetic agents come out. So we have our own pencasting agent now. Um, you know, and that kind of is finding these elements. Um
[00:03:00] [duckanbro]: However I'm probably very agents for you and that side. Um so what with your like daily stuff, uh uh I'm sure in customer service you're you know you're keeping a lot of metrics and uh you have like probably photos and and or like um are you touching how long issues take to resolve how these things are like going into this from the old system are being kind of adapted into this new system? Well um I could see even on that analysis side and that you could be building a lot of stuff to to extend that data and and um get some more boundaries from the VP. Do you uh so you you're talking about some tasks and uh automations are sorry, you're running things um on schedule and uh to process some of this stuff. So it just like takes off in the middle of the night and report in the morning. Do you um when you use tickets coming in? Are they are they like primarily hackers or something with uh boundary requests or are there vulnerability reports or um you know what kind of tickets are there? and the new hackers coming on the Slack is that uh the communication point, or is this or is it more internal communications? So you uh issues coming in through email or through webcam? That's cool. So internally you you have other kind of operators that are using the tool you build from the slide bar. Yeah, that's super cool. Um So I guess one of the other things is like it's just too pretty fresh. Um what are some of the things that still feel kind of broken? Like you're maybe too risky or just confusing in general. Yeah, that's uh we hear that one a lot and I think it's for everyone. You know, like managing keys and innovation and other systems. Um I think to the point where a lot of people like actually kind of starting to just like, yeah, we just build the internal and we have control over it. Um there may be some scope issues when you're working in the larger kind of group like you are. Uh oh yeah, that's interesting. Um so I I guess uh some of these are just kind of questions from the script, so um there's other things I'd like to add on too, but I guess let's touch on this interest. Um so what should other people look?
[00:03:01] [chickenparm]: And it's not just all so it's like these hackers are building their own agents now to find things and so at scale they're able to submit a lot more. The challenge that we're having now uh internally is how do we keep up with all of this added volume that I think is probably a problem across the board with with a lot of places for me in support, I'm seeing the same thing. We went from getting, you know, X amount of tickets a month to like three X in overnight at one group from just how much easier it is to submit support tickets now. Um so I think that's kind of in our world uh what's kinda going on is we're we're shifting from this traditional, you know, uh just manual hackers jumping on this and kind of looking for bugs in teams or with them by themselves uh to we have actual agentic uh agents kind of just at all times scanning and searching for things. Yeah, so super exciting. Uh I I've kind of my company is giving us call code. Um we use a bunch, we use them all. Uh we have chat CBT, we have we have everything. Um but uh just kind of unlocking the power of call code and understanding how it works and uh my company being able to run everything through bedrock or uh, you know, we're a security company, so uh security is like super super important. I think that's what everyone was scared about. It was it was kind of funny. For a while it went from like uh, you know, can't use any AI at all, lock in on everything, no vulnerability data, nothing, and then when we kind of figured stuff out and calling code got a lot better, I would say it was like a snap up window overnight. It was like, all right, you know, can you build everything that you can break stuff? We want to see what you can do. Uh and I think that's just because the company's feeling a lot better from a security perspective of of, you know, it not going out of bounds. Of course, we hear stories that can be that and can do all those things. We're hoping that doesn't happen, but uh that's where we're at. So the use case I have right now that I've been using for the past two weeks is um I'm just building uh I just built a Slack app called Disfaction that kinda just happened to do with the volume. We have this increased volume coming in. I always made the joke, you know, we use a ticketing platform, it's like any other company, you submit a ticket and you have a twenty-four hour wait, my agent will reviews it and responds. We have these like custom set automations that would look at like literal keywords and then send you a deflection message, but we would get slammed with CSAP because half the time it's picking up keywords that have nothing to do with it and it just was not accurate at all. So I think that's the the difference we're seeing is I'm able to meet myself as someone who's not super technical, I don't need to work with the dev team as long as I have the API keys to all these things that I'm you know interacting with. We use Fresh Trust at Hack 1, so that's my ticketing platform, but I have an API key that allows me to point to it and look at all the data. Holy crap, it has been like unbelievably powerful where uh we have a channel set up for our team, it sends tickets to those to that channel. For now, I'm still I still have trust issues with it. So we're not just letting the automate responsive out, we have human groups, so what my team is doing to kind of save time.
[00:06:00] [duckanbro]: from your experience so far I mean besides some of those issues um what have you tried some like real winds and yeah yeah I I totally agree with you I didn't uh we see that uh really like a a product group so we have people that are coders, some designers, product managers, some people very technical, some not so much. And you know but everyone very creative. And so I'm just like for the people that are first put in a target on this like just ask the chat. It will tell you and once people put over that first road, then all of a sudden you know, two weeks later they're um writing their own hinds and uh you know with around uh infrastructural and just like you guys just fast forward with my three years of suffering and learn how to become a dog but um but uh no but it's awesome because it's it's just a different way to learn because you're still learning all along the way just like you mentioned. We pick up things along the way and we're able to talk about it. Um things I was kinda curious you mentioned was your the human and the loop concepts and um the confidence scores. Those are actually I think very you know very useful and powerful for what you were doing. I'm I'm wondering how how that works and and how you guys are leaning into it. Like the trust issue is always a big issue. I can see even there's a bit of a learning control of the agent um and the context that you're giving it over time as as it's sitting out sixty percent you're like no this is actually closer to ninety percent, you know, if it's if it's able to kind of mark its human um flag on that and then goes back and looks it's like okay I was at sixty but you didn't say this is pretty good so I'm feeling a little better about and then first age you know are you adding things to like give it a voice or a you know some brand new guidelines to its responses or are you doing like a another in the workout draw maybe a clear pass to make sure not sensitive vulnerabilities or internal information is left out put in that well um we're trying to keep these around thirty minutes for our guests so if you do need a draw it's for I did have one more question and could you get a few more minutes. Oh yeah. Uh well if you gotta draw it was more I okay what's your um how do you see a change in your industry? Um you see it being um doomer or would you be more optimist?
[00:06:01] [chickenparm]: Is going through these states that I've come into the channel. I have a couple buttons set up, you know, approve and resolve, if it's a one-time thing, edit, escalate, and kind of learning based on other sources. But what it does is we suggest your response. I've ingested every GitLab issue, every linear issue, our entire doc site, every ticket we double resolved, and it kind of looks at all that, aggregates that data, and then comes up with a suggested response, as I am with Motes AI. I'm unbelievably impressed with most of the responses that it has. And yeah, that's kind of what we're doing through to save time right now. But the end goal is for high volume areas where it's super hand when there's just literally nothing we can do. A common ticket type we see is you know, hey, this program hasn't responded back to me. I submitted this report two weeks ago, what can I do? It's unfortunate, you know, we we do not work for that program. We can definitely bump it with the program and try and get a response out, but not a ton we can do. Uh so that was like a large, large chunk of our ticket volume. So the end goal is tickets like that. We don't need a human interacting with uh we can get that once the confidence score gets up to like 95%. Uh we're looking at just fully automate that to free up the agents to do more strategic work. Yeah, so yeah, so we started from scratch. I guess I kind of went into uh I was using GPT for a while, so chat knew me very well, uh, what type of manager was how I like to run things, so I kind of went in there. I actually used Chat C BT to come up with a prompt of explaining exactly what I was trying to build, what I wanted to do, and I said, hey, I'm putting this into Clark code, I need you to come up with a killer prompt for me. It did come up with a great one, you know, I had to tweak it along the way. Um, but that's probably the number one thing that I love so much is you know, as I think of these ideas, like the approval resolve button, I just put that in today because I was approving these tickets, so it wasn't resolvable, and then that was hitting our first contact resolution. So it's like, oh, we actually need a way to just approve and resolve in the same thing because some conversations you don't need a backward, it's just here's the answer and we're done. Um so like this morning I was able to go into Cloud Code to say, hey, I need to build this button in here, you know, but it takes about five minutes, it does its thing, and then uh it restarts our polar and I go in and every new ticket that's finally in now has an approve and resolve button. So I think for me to post someone, someone who's not super technical in the past, a year ago, I would have had to submit a dev request with infra. I would have had have a really smart coder kind of jump on and you know, get with me, understand the scope, figure out what to do. Just to add that button would probably be a minimum of a week, uh, just based on how busy everyone is. Uh but today I have it done in five minutes, so I think that's the biggest value is just like me being able to take ownership and actually do stuff. Yeah, no, it's from my understanding, it's all internals, so typically like today, um, I just needed to add uh something to the sort of scope of the app. So there's certain things within uh like the Slack API.
[00:09:01] [chickenparm]: That when I go to change them, most of them I can do myself and I don't need approval. But for some of the bigger ones that could be a security issue. Like, you know, if I'm giving it access to actually write, not just read access or write infosystems, that's almost always going to be your approval. So what happens is I will submit that for approval, and then we have an internal like IT engineering department uh that will review that request, see what what they want to do, and uh yes, it's up to them to approve. It's so funny because the amount of times we use claw to troubleshoot claw is is is really interesting. But yeah, that's exactly it. Exactly. Yeah, that's so we used to primarily work out of Fresh Desk just like within that UI, and now we're all working completely out of the Slack app. So it's it's even less you know, there's a lot of benefits. It's it's less contact switching, so less hooking around um with the info that it has access to. You know, when a ticket came in, we used to need to do something as simple as fill the properties field and that would take time just to track down some of that data. Um and now that's all automated. When the ticket gets submitted, it just comes filled in already. So um yeah, I'd say uh metrics files, I don't have anything fancy or any huge winds to share right now. Uh we're like still in very early development stage, but it's funny you're asking because you know, I want to get credit for this, I want my VP to be pumped out of this. So uh what I added this morning was I have it tagging any of the tickets that it responds to. I just have it adding the dispatch tag to it uh so that when I go back and I want to do reporting on it, I can very easily pull reporting on you know how many tickets we've actioned compared to uh, you know, maybe two months ago uh and the improvements. So uh I'd say in my company, data, data, data, data is the biggest way to be here. People don't want to talk to you unless you come with data. Uh so that's a very tough one for me. Yeah, even having that like that's another quick one, like having nothing to do with even the app, you know, just like admin stuff, right? Where I might have automations or I don't know, something set up and they might be conflicting with one another, but within the fresh SUI, one of my biggest pain points was there was no one to ask. There was no way to go and look and say, okay, if I if I go to production with this, what am I gonna break? What other things is this gonna step over? It couldn't do that in the amount of times we just have fire so you wake up one morning and you're like, holy crap, I just shipped this fix and now it broke seven other things, and our we have very frustrated customers or something like that. Uh you know, just yesterday I could ask it to kind of go, hey, take a look at all my automations. Let me know what looks messy, what I can clean up better. So I think just uh just overall cleanliness and just like the management of of tasks that would have taken me hours to investigate and like track down before I have this ugly old buddy that I could just message and ask to kind of try and figure it out first.
[00:12:01] [chickenparm]: Yeah, yeah, some some things are being scheduled. Yeah, well see, I'm learning too. Again, that's as silly as this sounds as someone who's not super technical, but I wasn't super aware of group limits and kind of that in the developing world of, you know, what do weight limits look like? How much can I do? So I was going nuts in the beginning, because again it was just, hey Jake, here's the keys, fill what you can. Uh and then I was getting every few hours, hey, you hit a weight limit of uh calls to a fresh desk API. Um so you know, just kind of doing stuff like, hey, let's cache all the call all of this data so it just lives locally or on EC2 wherever it is. Uh let's have it there so you don't need to make calls every time. Well I just learned that last week through this. Um and then yeah, kind of uh having saying, Hey, we only need to fetch new tickets. No, you have all of our tickets. Only fetch new ticket resolutions every Saturday night at 10 PM when no one's online. We uh we're actually four back live operations, so we cancel it off. Um so yeah, stuff like that is is what we go. So we don't do a vulnerability records. We have a huge triage team for that. Uh we just expanded recently as well. Um but yeah, that that's triage who's like really reproducing those vulnerabilities. Uh the type of tickets we're getting are so we're a technical support team, we were recently taking a lot off the plates of engineering so we could focus on these other creative tasks. Uh so things that would have gone to engineering before, for example, you know, integration, so if you're uh SAML broke, y you sign in and now you change those cert and no one could get in uh everyone's locked out of the platform, stuff like that. Uh that's like on the customer side the more technical tickets we're getting. Uh and then on the hacker side, you have everything from, you know, a brand new hacker who just joined and just as genuinely curious of how to go about getting bugs and how this this whole platform works, uh, to hackers who are seasoned and are really upset about hey, this you know, this program is whatever is happening, you know, I need you to intervene. Um and then I don't run this team, I I run the customer experience department, but um the the team that would deal with that is we actually have a dedicated mediation team that is specifically working into uh those issues. I'm saying in my space mostly internal. I I I should share that I was uh I dreamed that everyone is a CSM, so I was customer facing before I I'm now new or shifter, but when I was customer facing, I would have shared Slack channels where we can uh see with some of our customers. Um but for me personally right now, this is uh mainly internal and I uh there's no customers that I um have conversations with on the Slack. No, they go, yeah, they go into a form they you have to go into the portal. Uh so tickets are submitted via the portal and then uh with what I felt they're now coming in via uh a Slack channel that we have our own Slack B and you know we can do things like we can literally add this app that I created and then ask you, hey, I have this ticket, you know, can you please find similar tickets to this one? What was the resolution that do you suggest you apply? So
[00:15:01] [chickenparm]: It's not just uh getting tickets and suggesting that if I when it comes in, uh you can actually um like you can communicate with it itself and ask your questions and have a search interaction and whatnot. Yeah, and the the most common thing I'm hearing from my team is they're able to do more fun things right now, like I shouldn't say fun, but they're they're able to actually spend time and determine uh these deeper issues that maybe they couldn't before because the volume was just so much. Um they're now able to kind of answer those, you know, lower effort issues a lot quicker, um, and then kind of just jump in and and really track down and work internally to get some of this stuff. Um I would say uh not fully that's really question, but I I think what I've left out is a lot of these things that I want access to, there's an API key that grabs seven does. So the pain point I'm running into is if I want inf uh data that's from set somewhere that either A, I don't have an API key, B I'm not allowed to grab that key, or C one to seven exists. Um that instead my my biggest pain point right now is actually where to grab that data from because you know, there's a million things I said want to do, or a lot of that information lives locally or wherever it lives within our system, um and our company just isn't comfortable yet getting full access to whatever that database is. So uh uh honestly our team has done a great job of uh really kind of getting us access as it's needed, so over time we're seeing more and more become available, uh as I think we're able to trust it more. Um but that's probably my biggest pain point right now is uh just not being able to grab all the data that I need. I think enabling, seriously. That's what I'm doing with my team right now, is encouraging folks, hey, well so that's something too to do a lot about. I you know, I feel very fortunate that my company is set it up the way it has been, security focused. That's the way uh at least has been sold to us, is kind of you know, hey, it's not gonna do things in or quotes of what it's not supposed to be doing. Um so we kind of just go at it with that attitude of I'm gonna keep going until it tells me I can't. And then when it does get there sometimes, it will actually pump the brakes when you ask it to do a request and just tell you straight up, hey, this is you know, again, it's kind of like the policy, you know, y you might want to go to compliance, whoever it is to get approval. So for my team, I think the biggest thing uh at least what I see, not even just in my company, but just seeking peers in the industry, I see a lot of people who aren't technical that are now crushing it and you know making these cool apps like the one I mentioned. Um and we'll kind of discuss it with others as if they are code and developers, and I think that scares a lot of people. Oh god, this guy's so smart like he got it so quick. Where I have the exact opposite approach with my team because my whole team has uh every single one of my agents have has access and you know, I can't do everything. I I need all the documentation and whatnot, so I it's really just showing them like, hey, when you get stuck somewhere, because this is what I do, if they ask me to input something in the terminal and I don't know what that is, or I don't trust it, or I don't know what what's gonna happen, I mean really
[00:18:01] [chickenparm]: Screenshot that prompt, I will put it back across and I'll say, hey, what do I do here? What you know, just ask it to help me. And almost 100% of the time it leads me in the right direction. It explains what it's doing. Um, you know, I I'm not someone that lives in terminal usually, so when I'm putting you in this out in terminal, you don't always get like a a response back of like, you know, success, so this is right. So that's when I'm screenshoting this and saying, uh, did I do this anything? Can I can I do a bone now? Uh and that will kind of just reassure me and help me along the way. So I think that would be my biggest piece of advice is um, you know, if you're comfortable with it, really failing forward, playing with it, um, and not being afraid to like just keep asking questions, because even just these two weeks that I've really dug in, um I feel like I'm able to speak the language a little bit better and kind of at least understand what's going on in the background. Sure, for sure. Yeah. So I that was part of the prompt, you know. I I actually when I was seeking the chat to initially draw this up, I was saying, you know, think about other uh support departments that are doing similar, you know. I I wanted the confidence score. We have another internal tool that our engineers use. Um that came like that was the first time that I saw confidence scores coming back. We would submit GitLab issues and then would come back and say, Oh, this is the number of that. And that's actually where I got the idea from. Um, so I you know, I I kind of just asked the prompt, fake that in, uh, and then you tweak it right over time. So uh I I've been like tweaking the weight uh and kind of you know, if we hit approved, what does that mean? If you hit no, what does that mean? I'll be honest, the confidence scores are still something that uh like we're working on to get stronger. Again, we're just two weeks in, we've already seen like things jump from I'd say on average 30% to probably around 65, 70%, which is still super low. A lot of the times it's not super confident, but it's still spot on, so it's like kind of making sure it understands that and knows that. Um yeah, that's I think right now that's just uh kind of what we're looking at is teaching it. But the other thing I noticed is as you get it more data and more information to query, the buffering scores way significantly. So we had something like 70,000 historical GitLab issues. And before I uploaded that database, that's when we were it it wasn't doing the technical tickets very well because it didn't have a full context. And then as soon as I got that API key to GitLab and I was able to have it look at all those issues, that's when we saw it really improving on the technical scope. Yeah. Yeah, and something we're still working on is like, you know, it gave 80% of the reply perfect, but this one piece is a it definitely can't be sent, and we want to edit it, you know. How are we gonna tell it that so it knows next time, hey, eighty percent of this was good, just let's let's not put that.
[00:21:01] [chickenparm]: uh that paragraph and then let's replace it with with X, Y, Z, so just like really helping you out over time. Yeah, for sure. So just at a at a bare minimum just in terms of tone, you know, looking at those phones and tweaking it over time, just kind of I don't know, see where that I feel like it it does. And I feel like this with with most AIs is once it gets familiar with it, it does start to to really nail it and get stronger over time. So even this two weeks like I'm seeing it like I'm really impressed with the replies that are coming back and and even the tone of everything. It's like almost perfect for uh customer base or it's a c uh uh hacker uh an actual customer um but in terms of the security piece that part I don't need to worry about as much because so um we have all different so with all these different AI tools we have a data classification policy and then based on which rules we have certain data approved for each one. Uh so there is one tool we have that's approved for critical data and that's the one that you can you can throw in the house out of you can send anything to it and we're okay. Um but it's when you configure that so when you set up claw code internally to hacker one, it's already being built the with like the way that you have to set it up with all this security focus in mind where it's not supposed to provide any vulnerability or anything with that. So uh I have not seen that come through. There's been nothing even close to being like, oof this is if you I can't believe it's suggesting this. Um but I d I do think that's because my data team AI engineering team which we actually just built that team in the last like two or three years they crush it and I think they've done like a really good job of trying to make sure that we're we're protected and knowing that we're giving this tool to people who uh who might not have that full technical ability so let's be careful. Yeah so I I actually do need a jump for another call peak but I'm happy to hear it if you want and I can stick something back to you real quick. Uh I don't think I'm in the middle I don't see either and I'm very real with my agents about this too because you know that's their biggest concern is hey we have all this awesome um this awesome we say this is stuff is answering itself are we training our replacements right now and I I don't think that's the answer. I I I don't think the answer is yes. I think it depends on the space. The states that we're in, the issues that we're getting they really are things that you know uh you could just go in and it's like a simple response or you could look at a resolution send them a lot of these things are like things like payment issues, currency conversions, a a lot of that where you just can't let uh say like an an automated system make these decisions. So that's where at least in my space how I'm building this and what I see with for the next three years until anything changes is I think human loop is going to be very important. Um and then what my team's doing as well is not everyone is technical support. So not everyone I hire is super technical can do those jury integrations.
[00:24:01] [chickenparm]: and sample setups and whatnot. So we actually upskilled any any of those uh uh agents that were feeling worried about like how bad my jobs at risk, you know, I care about these folks genuinely. Um and I don't want them to feel that way for their whole career, you know, for the long run. So we did a whole training program. It was a three-month program where we trained our technical folks up level all of the uh support agents so we can at least ground level start to do some of the sample troubleshooting, start to set up some of these year integrations. So I think uh for me uh the the long-winded answer there is um I think it's like a hybrid of keeping up with your skills, you know, understanding how to use these AI tools uh and optimizing to the best. Um but then yeah just just also kind of making sure you you're staying with the the flow and that you still hold value.