Welcome to Energy Bytes with John Kalfayan and Bobby Neelon. Your essential guide to the intersection of data and energy. This podcast dives deep into the world of energy, shedding light on how data, AI, and technology are revolutionizing this sector. Each episode equips listeners with insights into the most efficient tools and resources, paving the way for a data-driven future in energy. From technical nuances to broader industry trends, Energy Bytes offers an unparalleled perspective on the evolution of the energy industry. Join us as we decode the algorithms of energy, one byte at a time.
We're back as always. I hate that we say that every time. We got to come up with a new one still a year [laughter] later. But it's not exhilarating anymore, so It isn't exhilarating. It might be exciting or another e x superlative, but we're back here today. My trusty counterpart Bobby Neelam. What's good? Thanks for making the trek up up to the big city. Oh, yeah. A lot better than at 7:30 in the morning. Yeah. [laughter] The afternoon ones are much easier, I assume. And Mr. William Fox, the GM at of Drilling at Corva. Thanks for having me, guys. Thanks for coming on, man. We appreciate it. [snorts] So today's I'll let you introduce the the topic cuz you know a little bit more about it, the story of Yeah, again just kind of another AI gone wrong or just AI used improperly. Well, maybe not improperly cuz the guy made guy made $10 million. Um [laughter] But then he got busted. But yeah, guy of the music producer was uh created thousands of bots to uh basically listen to his [snorts] songs on all the different streaming platforms and apparently made $10 million off of it and the FBI eventually cracked down and caught him, but uh I mean, don't hate the player, hate the game. Yeah, I know. I mean, it's a
But yeah, just automating fraud still fraud? [laughter] Believe it or not. Um but um yeah, I I think I think it's just kind of highlights just in general these days. I mean, like this is such a new space. AI, you know, I think it's been a concept forever, but I mean as far as actually seeing it in practice, you know, you know, commercialized, you know, for good or for bad, you know, is it's we're on the forefront of it and just, you know, there's a lot of price, you know, FUD, fear, uncertainty, and doubt around AI and you know, probably nowhere greater than that. We've talked about a lot, but then oil and gas where um [snorts] we've talked about the risk and everything associated with it. We have FUD everywhere. Yeah. [laughter] It's not unique But to one specific thing. I think it's interesting because it it opens up the, you know, the whole AI safety stuff and all that, but it's to me it's one of those things, especially with the whole open-source community, once it's out, it's out and you can't really control The cat's out of the bag. You cannot put it back in and to certain things. So that means by default, kind of like we were talking about on a another episode, but it's like by default you have to kind of architect things knowing that people will at some point try and use them improperly. Yeah. But it's then that becomes this whole like philosophical debate about, well, should you create it if you know that there will be the potential that it could be used improperly. But if not us, then Right. Someone Russia, China, or whoever else, but Someone else does or it's like how many technologies have been invented that are that like nuclear, right? Like that's a big counter thing for It could either save the world, you know, if you're you know, worried about the climate or it could destroy the world World War III, nuclear war. Yeah, it's [snorts] one of those things where it's like I don't think you have to be terrified of it. You just have to be aware of the potential consequences of of these things and stuff like that. But Sure. But yeah, I mean, with William's experience at Corva and uh [snorts] you know, I think probably their foray into some AI stuff. Love to get your your kind of pulse on it some of that and Well, it's it's funny because as somebody who kind of deals with an operational AI system every day, I have uh much less fear about uh it taking over the world uh than maybe You know it's You definitely know its limitations. Yeah, exactly. And uh you know, our our flagship AI product uh uh is called predictive drilling and uh there there really is a an AI piece to this, which is that uh the the R&D team at Corva spent at [snorts] this point going on 7 years um figuring out how to build a predictive model for rate penetration based on, you know, about 20 inputs. Um and then what we're doing is training models either for one basin or for one particular customer in a basin or across and weird situations sometimes they work across basins you kind of like counter intuitively and the AI portion of what we are doing is never going to be pressing the nuclear button because what it is doing is creating set points for an auto driller and we're integrating with a rig auto driller to then based on the simulation to yield the highest rate of penetration within the limits that were given the safety limits of the rig the limits from the road map from the drilling engineer pick the best set points make the most use of weight on bit stay under your differential pressure limit and so on and what's different from your best driller doing that basically cuz it won't beat your best driller is it's like your 85th percentile driller all the time day or night rain or shine and follows the plan and then every 15 seconds completely does the parameter sweep and sends those set points back to the rig to execute so it's it is actually making use of a of a multivariate objective function using an AI model to predict ROP and then selecting set points but it is very constrained you know this is not a general purpose AI tool Yeah. Yeah, well you've got I mean we were we're kind of relating this back a little bit on the development side we actually implemented I've been talking about this development tool called cursor for a while but we just signed up for the enterprise plan and are rolling it out to all of our developers to start using in order to get that 10x out of a junior developer enabling him to do stuff with AI literally amplifies or magnifies the amount of work he can do it's the same kind of concept as what you guys are doing right you don't have to have a 10 year driller out there you can bring someone in that's a little bit more junior maybe a little bit more cheap maybe in a more established basin to try out so they can learn in an accelerated fashion without the risk that we always talk about in oil and gas which is you know catastrophic failure and all of these really scary things and that's but that's like a perfect use case in my from my perspective but coming from the field the only way to learn in the field is by doing and so that's terrifying when you go out as a on your first day week month as an engineer in the field like well [ __ ] this could go wrong really quickly and I don't and you're there with your mentor or whatever but it's like at some point you know the mentor leaves and you're there by yourself and Right. you know it might be 6 months a year whatever you still don't have a ton of experience so to speak. Yeah and then a downturn comes and you get fired exactly. And [laughter] And so that's I think the big the big macro trend I'm seeing across pretty much every platform at least in oil and gas on the software side is really at least the ones that are being successful is a really around how we do more with less from a personnel perspective right because we have had such attrition Yeah attrition downturns etc. that just kind of compound all these things and so I don't see you know operators running to hire more people back anytime soon unless prices just go crazy. So these are the types of things that we have to implement in order to keep maintaining because no one has hired back after negative 30 and then Right. COVID Right. But stuff has slowly crept up right? Well and but you bring up a good point which is sort of twofold one um you know we did a study with neighbors they can go pull the number of HMI interactions human machine interface interactions and a driller without something like this is doing about 250 interactions per thousand feet Wow. and with this we've gotten as low as one and a half per thousand feet. That's insane yeah. But the flip side of that is we do our weekly or end of pad or end of well sort of call in with the super and the the company man and they go my driller's getting lazy. [laughter] And I'm like well that's an interesting point because your point about how do I learn so one of the things that Brian our CEO has challenged us to do is to make the app tell you why it's doing what it's doing. And they may not want to look. Maybe you don't care. Maybe you trust it enough. But on the other hand, to your point about learning, okay, why is it lowering weight on bit right now? Well, now the banner says, "Well, you're about to hit We're projecting that you're going to hit your differential pressure limit. Therefore, we're going to bring weight on bit down 8% for the next I don't know, 50 ft. Yeah. So, at least it is it's not a black box that we're just asking you to trust. It is a system that is giving you enough information back if you're in that chair and you can understand why it's doing what it's doing and then go, "Oh, okay, I believe that. That makes sense." loop element to it. Exactly. I think that's brilliant. my You know, my favorite things that I heard it was like augmented intelligence Yes. versus artificial intelligence. I mean, like you still have people out there. I mean, obviously there's so many I'm sure can override it if they if they needed to. But like this is enhancing what the people are doing. It's not replacing them. Like in my mind when we get to full artificial intelligence, it's just going to be like some machine that drives itself out on location and sticks holes in the ground and it can just do it all itself. I mean, like we still have humans there. We're just making them better at what they do. Yeah. And or more consistent, which you know, there's a lot of to your point, there's a lot of value there. That That's That's exactly right. Yeah, we're not at the I press a button and that says find me money [laughter] and it takes care of the rest like permits, the whole thing. Uh well, you know, the other thing that I we found is that um the getting the model working is great. But then 90% of the remaining work is in the training, getting earning the trust of the crew, but also how does that model then translate into real world action? And I'll give you give you an example. You know, depending on which drilling contractor you're dealing with, which auto driller, which control system, each one of those control systems has their control loops set up differently. And so, uh rig A may allow you to overshoot a particular set point by or a limit, I should say, by X percentage. But then rig B might be configured that if you touch that limit, ROP's going to zero, slack off immediately. And so, we realized kind of early on over the last 6 months that we had at the software level try to abstract that out, try to mitigate that by having the software do more. And so an example recently we rolled out a weight on bit and RPM ramping so that when it determines it needs to make a change or maybe the road map has changed because we've changed formations. Instead of just going, "Well, I was at 37,000 weight on bit, let me go to 23." [snorts] And it's like, you know, which is causing dysfunction. We've got a stepping function in because we we can't necessarily from the cloud know exactly how every individual rig is going to handle that change. So we have to sort of take on smoothing that out from the perspective of the end user. Just as an example. So is that AI? No. It's It's scripting logic, but it's it's what's necessary to write on top of the AI to actually make it useful. Yeah. Yeah, it's intelligent and augmented, right? It's [laughter] basically the same thing. I think that's uh I mean, there's again, everyone talks about the AI is going to replace my job thing, but I still think we're a ways off from that, and I think the iterative steps are AI takes away all the [ __ ] that you don't like doing that's boring, monotonous, tedious, whatever, and enables you to do the stuff that computers can't do, and the computer does the other stuff that you don't like doing and are probably bad at anyway. Yeah. And again, I we've talked about it and you brought it up too. Like now you're consistently getting this 85th percentile, you know, driller, you know, and like the quality of your wells should always be, you know, roughly at that, whereas, [clears throat] you know, it doesn't matter how good the person is, even if they're a 100th percentile driller, you know, for 4 hours, you start getting hour 5, 6, 7, 8, you know, whatever. Like I mean people just That person is going to go on days off, yeah, at some point. Yeah, there's fatigue, stress, there's all kinds of others. I mean, that's it's one of those things that's very hard to describe until you are sitting in a in a dog house or on a frac van, and you know, there are moments where there's it seemingly nothing going on, and you're just sitting there, but then there's also moments where it's just like pure ass chaos, and you don't want to be entering a bunch of [ __ ] that's already been recorded into a spreadsheet just so that it can be put into a, you know, PowerPoint. Yeah. [snorts] You know, and and people ask us all the time, "Well, what about what about LLMs? What are you doing with LLMs, chatbots?" And uh I think, you know, what we found is that kind of two concerns. One, uh we are not going to nor would would we even ask permission to ask an end user, an operator to put their data into Microsoft servers to run OpenAI stuff. Right? It's just a non-starter. Mhm. But there are things, to your point a minute ago about what are the things I don't want to do. One thing I hate doing is building dashboards. I hate trying to figure out, "Hey, go pull me uh all of the lateral sections for the month of August and graph uh gross ROP on bottom ROP, utilization of software X and spit it out." Like I I could literally waste an hour just trying to do that. And one thing that uh we found LLMs can actually be trained to do is you can train it to use our query language, so then you can chat to the bot and have it build a chart. So, is that earth-shaking? No, but is it, to your point, removing something that I don't want to do and doing it in a much more efficient way? Yeah, so there's that. Um you know, we had another experiment where we do, right now, across the street over there, [snorts] we've got a 24/7 service center, we've got people who are experts in drilling completions who are answering questions. We thought, "What, you know, they can't be cross-trained on everything." Right. Uh maybe we can train a bot to know everything about predictive drilling. And so, uh shout out to Curtis uh Cheatham, our VP of R&D, fed everything that we've ever produced internally, all the SPE papers, all the PowerPoints about predictive drilling into a bot. Said, "Okay, now give me a summary." And it was all right, except it added in "And we control the pumps." It's [laughter] like, "But we don't." So, you know, it was like 95% of the way there and then it hallucinated "We control the pumps," which is not true uh and is explicitly not true in any of the underlying documentation. So, I feel like we're on the cusp of that, but we're also not quite there yet. Yeah. For really trusting it to [clears throat] give technical advice. No, I think that's the It's just like with all these things, right? Like it's a trust but verify in the early stages and then until you are comfortable enough that you feel like you can either trust it or even like I mean it I liken it to it's the same as a Google search, right? Like if you know how to Google search you normally get better answers than if you just type in a full sentence in the Google search. Same thing. I actually read somebody one of our advisors sent me a paper this morning and it was talking it was about specifically getting rags not to hallucinate to be 100% [laughter] accurate. And the the interesting like nuance that he found was that you know, the way these work is just model their word predictors effectively. But they all go based off like you know, these things understand parts of speech and nouns and verbs and all this stuff. And he did all these experiments and it's basically if you can do noun filtering before that's his hypothesis. We're going to be looking at it and playing with it to see if it how much impact it does have, but he used some example about uh so-and-so Queen the Second versus Queen the Seventh. And you know, so they both have V or they both have II in Roman numerals and Queen and it references one instead of the other even though the other one isn't in the prompt and there's but it goes back to you know, the root document in the rag model has the modifier to the noun and that noun right next to each other. Therefore, it goes and it finds the next what it thinks is the next thing. And so it's like you're right. I think we're we're quickly approaching much higher levels of confidence with these things. People are getting more accustomed to them. They're understanding like hey, they're not going to be right 100% of the time. I need to take this with a grain of salt, but it can be an efficient way to do X, Y, and Z. And so So, very it's fascinating. It's going to be fascinating in that space, but I was thinking about the other day and it's like LLMs are almost a like a forced uh product of the volume of data that we have, right? Like the volume of data continues to grow and then ML was the first stab at that on the numerical data side of things, right? And now LLMs are essentially the text version of that of like how do we sort through all of this data that humans literally can't interpret or understand to try and figure out all these relationships and connections, but it's going to be very [snorts] interesting to see how it all kind of shakes out in the future. Yeah, that is well, you know, it's a great example. I saw something over Labor Day weekend. I had to go and recreate it myself cuz it was so bizarre, but somebody posted on Twitter that the newest version of Google's AI-enabled search was saying that the Austro-Hungarian Empire had a space program. And I thought, well, this can't possibly be right. So, I went and typed it in and sure enough, it informed me that in 1889 the first Mars expedition of the Austro-Hungarians landed and you know, and then of course I had a little grin and I realized that there was a tabletop game from the '90s called Space 1889. So, what it had done is it had read something and treated it the same level I I assume credible credibility as as you know, real history and it it kind of struck me that you can have the best search and the best LLM, but if it's still garbage in, garbage out. Oh, it always is, yeah. Well, that's what happens when you license all of Reddit's data and you don't [laughter] actually do much ETL to it. [clears throat] Or you think, oh, well, if it's got the most upvotes, surely it's the best answer. Except humans are are trolls on the internet and yeah, now it's like, oh, if you want a cure for cancer, go eat rocks or jump off a bridge. Like all these crazy terrible answers and it's like as someone who's been working on the LLM side, that to me is just a direct result of and it's one of two things. It's either the manager is saying, I don't care, you don't have the time, push it. Very possible. Or just being there's some kind of Yeah, I I don't envy the the job of having to figure out how to do that with Well, you you can't possibly predict the questions that are going to be asked and validate any answers you're going to give. So, it's like you're having to programmatically try and eat the ocean and they just did the best they could with what they had. And so, it's it's a wild thing, but it does it's at the end of the day every single math professor and engineering professor I've ever had garbage in garbage out. And that is the basis for all of this stuff. And you know, we are at Corva never going to solve all these things. But, one thing that we we do feel like we can help with is data quality on the way in. Um and so, you know, first off, we're never going to have the staff to do all the different projects that uh our customers, you know, be they operator, service company, drilling contract like we're never going to be able to do everything. But, our thought is if we can get the highest quality data, whether that be surveys, mud reports, BHAs, whatever on the way in uh and do all the correction as it happens. Cuz as you guys know, if we if we forget about it, 2 weeks later it might as well have not even happened. Like that guy is gone. It's that's it. And [clears throat] but then make that data available in a way that uh machine learning and AI algorithms uh can consume it like in a parquet file or something. So, kind of you know, comes in, we put everything in [ __ ] You tell us how you want it out, right? Uh and so, we feel like that's one thing we can do at least in the drilling and completions domain to help uh our customers eventually eventually leverage AI in like 50 different ways as opposed to right now just drilling, just parsing PDFs, just building charts. Yeah, cuz I mean if they're you know, and I'm assuming a lot of your customers are probably using multiple drilling vendors and all this I mean so, I mean they're going to get the data in different formats from them. And completion we talked numerous times about completions and like we're on the drilling drilling side. They're going to start commenting to tell me to quit ranting about how bad completions data is. Well, then do better. I was going to say. [laughter] Just do better, guys. Fix it for me. But again, that standardization of a format, you know, again, now things programmatically downstream of that can work with it. Again, even if it's not optimized, we've talked about Witsmo, even like, is it great? Yes, no. I mean, like, it's great for some things, poor for others, but I mean, it's a standard nonetheless. People have been able to build things on top of it. Um, but then, yeah, if you're able to provide clean, um, data in a similar format where you can go back historically, and if you want to do that train machine learning algorithm, like, it's takes one data munging, you know, step to do, not, "Oh, here's my Corva one, and here's my neighbor's one, and here's my, Right. uh, H&P, you know, you know, our Payson, or you know, whatever, you know, wherever they're getting the data from. Um, if they can get it from one source in a similar format, then Well, that's the, again, 80% of the way there, right? at the root of this. I think I feel like the industry got really turned off by ML and AI, and insert buzzword here, because we were promised all these things, which those tools can do if you have good quality, structured, clean data. But guess what? We haven't had any of that historically. We haven't needed it in that certain aspects, because we couldn't use it in these ways, right? But now, it's like, oh, [ __ ] We've got all this data, and it's all crap. Well, yeah, I mean, that, or some engineer tech at our company built an actual useful model on the data, and there's a couple problems there, because A, it was probably too narrow focus, and it was for, you know, one pad or well, and it doesn't generalize. Or B, well, that's great, but you [clears throat] actually didn't capture all the steps you had to do to get that data to a clean, ready state for that. And if you actually want to productionize that, yeah. You were just running Python scripts on your laptop, never get pushed in. And now, if you want to productionize it, now you actually need a data flow that's consistent in the format that you were getting it, and it was needs to be cleaned in the same way. Right. Um, Well, yeah, oh, I built it on a one-second data, but I only have five-second data. Uh-oh. Uh, well, you know, the other thing, kind of, that we've encountered, at least with some of these products, is that even if we get that model great in bench testing, like there's there's so many variables that the only thing we could figure out to do, and I say we in a very broad sense. I've been here 13 months, so keep in mind the prior 6 years were people, not me. Uh but so the team figured out the only way to make it actually work and stay working is to have an automated correction factor. So we're This is a real-time application. It's drilling. There's dynamic stuff coming in. The only way to keep it dialed in was basically do a 75-ft look back every foot. Right. And Okay. sort of dial up and dial down the weights and everything. Exactly. So we're doing a 1-ft summary and looking back and every foot that is feeding back in to try to keep the model on. And if we don't do something like that Yeah. you know, oh, one one well on the pad was perfect and the other two we basically they had to turn us off because the predictions were so bad, right? So you know, to your point, we can do all the data prep. We can build the initial model. But then if we don't have something to try to keep it dialed in, at least in these applications, in the real world as we're using it, it still won't work. Yeah. Yeah. [laughter] Yeah. No, I think that that window, I don't even know the proper technical term for it, but that whole windowing concept of being able to win or span time periods and look at all the fields and variables over that and then use that moving forward. That stuff is is a game changer for our industry because we have so much time series data. And it's Bobby and I have gone down that rabbit hole of like picking stage start and stop times using a model and it's like I can visually easily tell you the beginning and end of a frac stage, where they shut down, everything. But programmatically getting like trying to code that in on 50 to 100 streams of you know, 1-second ASCII data is much different. [laughter] Right. And then oh, wait a minute, you built it for land unconventional and you want to run it on a offshore rig that's dual Derrick with two different streams coming in. Oh, wait, yeah, we're going to have to that. Yeah, or we're we're simul-fracking now and we need both feeds in the same place because we want to see the communication, if there is communication. Oh yeah, then we needed FDI analysis on top of that. Right. Oh, hey, can I uh run this for well control? Well, [laughter] Yeah. [clears throat] Yes, but um we we're going to need we're going to need good data. We're going to need next to zero latency and you are all going to have to agree on what the definitions are and how many false positives you're willing to accept, you know. That's That's So, right now we've got a project to do um predictive alerts for stuck pipe. So, all three of the most common mechanisms. And you know, thank god we've got an operator who's helping us uh and cuz the most important thing for that is labeled data, you know, you tell us when was this event, what do you think the triggers were, what kind of mechanism was this, you know, just had to do with the torque velocity, this what was causing it and all that has to go in to the feature engineering and so on. And then you have to say, well, how many false positives and how many false negatives am I willing to accept? It's like, well, we're willing to accept zero false negatives and up to n false positives, but telling me I'm stuck as I'm stuck doesn't help. So, now [laughter] I need I need to get between 5 and 30 minutes before the event right and that just drives a whole another thing. And so, now we probably need good temperature data and we probably need good mud data and it's it's a lot of stuff. So, there bias in the labeling? Exactly. Exactly. And so, we can, you know, we're kind of inching out and I won't speak about all the completion stuff. Jessica probably talked about that when she was here, but you know, it can be done with enough [snorts] data and enough uh help from folks who are experts who can say, okay, that was definitely uh event category X and I would look at, you know, the cuttings build and these other things and if you put those together, we can build a model that'll be better than guessing. And the goal is to get, you know, 80% uh you you to 30 minutes before they happen in kind of a probabilistic manner, right? And there's no silver bullet, so okay, we got different mechanisms getting stuck. We basically have to run all these things in parallel at the same time. Now, that's another a very good point because we ran into that, Bobby and I did, on the the frac interference stuff, right? Is it's like, okay, you know, we were working at a gauge company, we put the gauges on the offset wells and on the frac well, you can see the treatment and then you can see the frac hit on the child wells, but then it's like, okay, well, now what? Yeah. So, we know what's happening. We don't know sometimes it's good, sometimes it's bad. It depends on formation, depends on where you're at, it depends on spacing, it depends on how bad, you know, [laughter] Colin Colin's Twitter handle is frac slap because he heard that mentioned in a paper about frac hits and he the guy was quantifying, I think it was George King, was quantifying, well, you know, you've got a frac hit and a frac bash and a frac slap and they're different degrees of hits and stuff and so it's like, that's the the tricky part about a lot of this stuff is it's like, we've got all this data and we know these things are happening, but we don't know all the time what to do about it or Yeah. And should we even make a change based on this because you know, we think it might be bad, but Right. Yeah. Well, the dream, right, is that you get your AI models to predict the stuff before it's going to happen and then you get your automation to mitigate it before it happens. Yeah. And uh but I think the baby step for us is, well, let's just do a a red, yellow, green indicator of how probable this bad thing is is about to happen and then, you know, once we believe that as a group, as a collective of operator, drilling contractor, uh you know, software provider, then we can decide, do we really want to trigger automation to try to mitigate it? Uh and but, you know, the other funny thing about that is, okay, great, AI, but now, how do we alert people in the middle of the night? Mhm. Because if it's if it's just in a banner, Right. then I can ignore that. And if it's just an email, then I can ignore that. And if it's a text to the DD, but he's asleep Mhm. uh in his van, then is there a escalation tree for the alert? And then at what point does it call the phone of the vice president of drilling and say, "You're about you know, Right. And so, there's all these sort of bits of okay, we've got we've been able to create the insight with some AI and there was some ML. But how do we then make it executable to be able to prevent the bad thing from happening? Yeah. With that too, I mean because we know [ __ ] happens in the oil field. Um you know, you mentioned the cloud and obviously we it's not a secret that Corva uses the cloud, but I mean this predictive driller like is it running in the cloud or is it running on prem or going to be both because I mean we are you know, on the edge rather like cuz we know that connectivity is not a guarantee. And so like how do y'all handle that? Yeah, so it's running in the cloud. And so basically every rig we're on today has Starlink with cell backup. Okay. And that's probably 98 and a half 99%. Yeah. Uh and that's good enough. Sure. Um we we have had folks ask about you know, we're we're in the process of deploying Corva on prem or in specific um geographies as the requirement of that oil company for you know, data residency. So we're doing that piece. Uh but currently we do not have any version of this running on edge devices today. Yeah. Is is the is that always the plan you think to try and do it from the cloud or is there any interest in pushing some of the stuff out? And the only reason I ask is because you guys do have it's not like pushing it to an edge device would be a new product line. The product itself might be, but you have an edge device that you or you guys have historically on the mostly on the frac side I guess, but Well, I think it this is one of those cases where typically [clears throat] the at least for my products on the drilling side, the the drilling contractor, whether they be land or offshore, very paranoid about owning that IT OT barrier. Yeah. Uh and they typically have all the edge devices necessary and so it's more of how can we agree where is the API between us and you that you're comfortable with? Yeah. And then it's cleaner that way too because then if they have their operator make like let's say a limit change. Like, I'll give you an extra 3,000 weight on bit, right? If they do that on their HMI that's under their control, and then they push that via API to us, then we know. Out of your hands. And if we push something back down to you, you know, it becomes your clean responsibility at this at this, you know, IP address, basically. Right. And uh so, that's been the approach at least on the drilling side so far. No, that's smart. So, with that, you know, and I think there's there's similar things that are coming up with this with AI, you know, like Tesla's auto, you know, driving vehicles, but like where where does liability land on this, and where really interesting question. from like even insurance standpoint. I mean, I you know, I'm sure there's things you may or may not want to get into around some of this, but like just how are those conversations how are those conversations, yeah, with you know, the drilling companies and the operators and everything? I mean, I think I think all I'll say on that is that uh all of these companies that are involved in any one of these operations have got, in our case, a software as a service agreement, in their case, an MSA in place that governs. If you think about it, people have been using ROP optimizers for a long time. And so, this is sort of just an evolution of the existing legal structure between uh service company, drilling contractor, operator, third-party software provider. And I wouldn't say that there's anything particularly crazy or different about how that works because we're still handing off what is ultimately a recommendation. But at the same time, we've we're taking the the safety piece quite seriously. So, you know, if we lose connectivity, according to whatever we and the drilling contractor want to agree to, right, kick us out, right? We lose control. A certain number of seconds, if there's not a positive handshake, kick it out. Uh you know, layer two, okay, there's the rig limits that are built in the control system. You try to send something outside of that, kick it out. Layer three, there's the limits that that particular driller has keyed into that particular chair. If we try to send something outside of that limit, you're out. Yeah. You know, so, the idea is that we are we are in fact in control for on bottom drilling in the sense that we're sending the set points to be executed, but there are still three layers of safeties around that. And if there's some kind of fluke where we try to send 80,000 weight on bit as the next set point, it will ignore it and kicks out. Got you. No, that's I think that's a smart again, it's human it's augmentation to a to an extent even though it is controlling, it's not controlling without any bounds bounds or with out a human potential for an override or And they have one button to press. One button press to take control back. Got you. On top of the automated control. I mean, I think that's again, to me this is a very iterative thing as far as the AI stuff goes and that is like the true first step, right? Is augmenting human activity behavior making recommendations or predictions to prevent things from happening in the future. And [snorts] it's again, it's it's the cruise control where it's like, "Hey, I'm on the highway. That doesn't mean I don't have to pay any attention. I still [clears throat] have to pay attention. Now it'll slow down for me if the person in front of me slows down or breaks or whatever, but I still have to drive the the car generally speaking. But for those boring times, it's in control and it's within its bounds and its limits and keeps going." So then say more from a commercial side like is this it may be both, but is it the you know, the neighbors or someone like that like saying that we want to use the uh predictive drilling or is it an operator saying we want to use Corvus predictive drilling we want to you know, you to allow this on your rig, you know, just how does that work? Yeah, so [clears throat] the with neighbors it's a partnership. We've kind of co-developed the fully automated version of this. So the team had been doing versions of predictive drilling in kind of advisory mode for about five years and sort of the big breakthrough for us to get to actually in the control loop was an agreement inked with neighbors last February. Um but you You the reality is it's it is a joint sale to the operator, right? You know, there's there is an incremental additional cost, which I think the ROI is 40 to 1 or better, but uh at any rate, you know, putting that aside for a second, it's absolutely a joint thing because again, if the drilling contractor does not want to do it or the operator does not want to do it, it's not going to happen. Uh so, but yeah, you know, we we were lucky enough to have some great customers who are pushing to find ways to have predictive drilling work on a variety of other platforms, not just yeah. Let's let's unpack the 40 times thing cuz I'm curious. So, like, you know, again, a lot of people a lot of people, laggards, whatever you want to call them, will say, "Well, okay, so what? It It does it by itself." What is What does that mean as as So, I'm glad you brought that up. the GM or the the manager brought that up because I think it's very important to say, "Okay, we can all look at the DVD before and after, [laughter] right?" And it is truly impossible [clears throat] to say how much What percentage of that difference, you know, came from this software versus some something else. And going back 14, 15 years, my first job at NOV, we had a real-time optim- optimization center at Jack Rabbit Road. It's engineers in there. They would call the rig and they'd like, "Hey, it's time to Time to do a drill-off test. It's time to do this, time to do that." And this is back when people were spending like 25 days drilling a well in the Bakken. And we'd do a pad and we'd come back and we cut 3 days off. And like, "Well, you didn't do that. We did that. We're geniuses. We're not paying. [laughter] Tear Tear your invoice up, right?" And so, I think we've got a we've got to recognize that you should improve across the pad, right? From the first lateral to the last lateral, from the you know, improvements is expected. Um and there's no way that we can claim that this software is 100% of the difference cuz it's it's probably not. Right. Uh but that said, to the and you'll just have to trust me on this. We're going to publish a paper later this year, but you know, an analysis of the first 100 wells. But to the extent that we can normalize uh well drilled with predictive drilling versus a well that was not drilled that was to the best extent the same well design, same BHA, same tools, similar rig. Yeah. It's about 17 to 20% faster, basically, on bottom. And so, give us credit for half. Right. [laughter] Uh that's my pitch, right? Um and you know, part of it is because the automation with neighbors, they've got this deal uh smart drill part of this package where it automates landing the bit, you know, it does a smooth handover between sliding and drilling. Like that that is absolutely the Right. the automation's part of it. Yeah. Uh but so, when I say the ROI is 40 to 1, I'm not going to say how much it costs right now, but basically it's not a lot. And if you give us credit for half of the savings you see on that DVD when we're used, uh I think it's pretty compelling. But I I want to add one other thing on that, which is that from sort of like 0 to 50% utilization no no statistical impact. From 50 to about 70% utilization maybe single digit in- improvements. It's like consistently. Like you know, I can show it on a graph. Improvements in ROP. Don't really see any change to bit runs. But if we get over 70% and get into that sort of 80 to 85% of rotary footage drilled with this thing control, it's double digit consistently better than an offset that's being drilled without this Right. sort of system. Um so, the key is that's why I was saying earlier the key is we've got to convince the the guys Yeah. at you know, and gals at the rig to to run it. Cuz it won't do any of that Right. if they don't trust it and we don't turn it on. No, that's why I asked the question, cuz there's always this dichotomy of technology in the oil field, right? It's like, oh, well, that's really cool and it can save me this, but what does that mean? There's always that business argument at some point, right? Of like, how do I quantify this? You know, it's it's certainly the number one metric is always going to be what's avoided spread rate. I mean, even just cutting down the drilling days, like that's a simple one, right? But like those have consistently fallen since I got in the field, but I feel like they they I feel personally this there's no completely anecdotal. I feel like that time kind of plateaued the last few years, where it's like, okay, we can consistently drill whatever it is, 5 days. I don't even know what it is now. We can consistently drill at this rate, and so you start hitting that limitation of just what is humanly possible without augmentation. And so, this is the next kind of iteration on that. We can take another step down. What we're seeing in the stats is that we can take and it's not This is not a wonder weapon, you know, but there's a significant step and improvement from bottom rotary drilling. We can avoid trips, we can use vibration filter, stick-slip filter, differential pressure derating to avoid bit trips. Yeah. And which is, okay, now we're saving completely Right. flat time. Yeah. Uh so, I think between the two of those, that's that's going to be the primary economic thing. What we don't have enough data on yet, but what hopefully, you know, if we come back this time next year, we'll have enough data on, is are we actually drilling better wells, right? Cuz, you know, for years it's been, let's go as fast as possible. Let's pump as much crap as possible, you know, what's But is it actually a better well? And so, part of the effort that my colleague Jessica's leading up is, okay, we've got drilling data, we've got completions data, we have some publicly available production data, and maybe maybe Mr. Operator, if you share your real production data, we could figure out for you whether we're actually drilling a better well. Yeah. Uh so, I can't make that claim yet, but I will what I will say is, you know, uh we don't beat up the bit, uh we seem to go farther between slides, slide percentage seems to be lower with predictive drilling turned on, and that would all sort of be preliminary indicators that maybe by being more consistent with the application of the drilling parameters, we're drilling, quote unquote, a better hole, but Yeah. even jury's out on that. these days with longer laterals and I remember even back when I was at Conoco, there was a thing with SPE paper or you know, maybe it was geology paper, I forget what it was, but it was about [snorts] the you know, the sinks and sums and the tortuosity of the well bores and just like the impact on production cuz you know, it's it's great you went fast, but now the thing's all over the place and is it even in zone or whatever? But like again, but again, I think the longer the lateral, especially on more liquid rich wells, the more that matter it's harder to move that fluid. So, if you have any of these little sums like it's not moving through there and it's hard to get it out. Yeah. I I can't remember that was at somebody's office in the last 6 months and they just up on the whiteboard in the conference room, somebody written we can uh drill wells, we can't case. We can case wells, we can't uh frac. We can frac wells that can't produce or you know, cement. I'm just giving it to you. I just thought that was a a very interesting mindset to have in the middle of the drilling team conference room of that, you know, it's not it's not just about I know this is being sold as an ROP optimizer, but it's not just about drilling a little bit about consistently delivering wells that you can actually produce. Yeah. No, I mean the other thing I think about there is just like cuz one of the things Bobby and I worked on at one point was uh drill out monitoring stuff, right? And one of the primary reasons they wanted that is because they're doing it with workover rigs that still have analog gauges and then they stick pipe and then no one knows who is to blame because it's just on this little tally book and that's it. Yep. And so it but then I'm starting to then we go through this, I build this business plan and business model for it and I'm sitting there thinking, I'm like, [ __ ] man. You're talking about like $100,000 minimum up to the whole cost of the well if they can't recover any of it. And so like that's a really nice insurance thing. So, I assume there's probably some of that as well on the drilling side where it's like, hey, by implementing AI, we know that our percent chance of getting stuck or completely trashing this well significantly drop and I'm not saying that we have a high as an industry occurrence of that or anything like that, but one well, I mean, what's the cost of a well now? A few million bucks at least. Right. So, like that in and of itself, you have that on your entire system could pay for itself just by preventing one in an entire year, you know, like I know that's hard to quantify for for itself. Right. Yeah, I mean, it's hard for operators to quantify that, but it's one of those things that like that is the reality of what could happen. I I 100% agree with you and that's one of those things that I think the data is starting to show is true. Yeah. But you can't sell it just on that. Right. It's like, yeah, but I also want to see the DVD getting closer and closer to left, right? For sure. You know, so there's we kind [snorts] of the part of the issue of of {quote} selling um you know, optimization software like this is that you kind of have to sell it at all levels, right? So, if I go speak to a company man or super, he's like, look, the only KPIs I have are I want it to drill smooth. Mhm. I want it to drill fast. And if you do that, I'm okay. Yep. And then, you know, you go to the drilling engineer and they're like, well, uh I I want to avoid bit runs whatever and we get as you go up, it turns into dollars. But basically you have to have an a compelling enough uh solution at all levels for all those people to basically opt in. Yeah. And, you know, I I mentioned to you earlier there's there's some things where, okay, if I can drill a well 20% faster, that's great up. Like that's a pretty that's one compelling economic thing, but maybe there's some other things that are like one or two percentage point things that you wouldn't normally do. And so, I'll give you an example. Uh we added a feature at the request of one of the customers around tool joint mitigation, right? So, you know, you're drilling, you're going through the rotating head, the joint gets hung up on the rubber. So, Exxon published a uh um a procedure a couple years ago, maybe it was a year ago, I don't know, time's a flat circle. Uh recently where effectively it said, okay, based on where the joint's going to be, a certain number of feet before the middle of the joint, go into ROP control and get weight on bit up 10, 12, 15,000 out of the way so that you will not slow down every time, you know, three times a stand. And you can see, especially if you're drilling more slowly or slower. If you're drilling slower, let's say under 250 feet per hour, you can see it you can see it in traces right there. It's going you know, slowing down. And but asking somebody to do that manually is adding like 20 additional interactions for that driller per stand. Yeah. And so it's inconsistent. And so that's sort of thing where of course you could solve that with software. It's not even AI. Right. But are you going to do it if it's not tied to some other thing? So we were able to take something that maybe maybe the well is 3% 4% faster as a result of that and add it on to this other thing as an additional service, you know, or an additional feature which otherwise would have just been it's never going to get in my top 10, Right. you know? [clears throat] But again, then but over the course of a drilling campaign in a year or you know, further. I mean those are insane, you know, It's probably single digit to tens of millions, Yeah. you know? That's the thing that I I'll never get over with [laughter] anything on the service side is just due to the volume of at least unconventionals, right? Is it's like a 2% improvement is actually significant amount of money when each well is 5 to 10 million dollars and you have hundreds or thousands of wells. Like that's Exactly. If you can do it across the fleet in a consistent way without making things worse, Yeah, on an individual basis on an individual well, it doesn't look like that much, but 2% on 100 250 500 million dollars I mean the lid on your water bottle right there like Yeah. They've got thinner Right. two years cuz someone's like well what if we just made them half the size? It's still going to hold water in it. Take the extra olive out of the jar again. strategy I say how much money they save per bottle? I mean fractions of a cent probably, but times how many bottles of water are produced every year? I mean that's insane savings. They compound too, right? Like that's kind of essentially where it's the same compounding interest principle, right? It's like 1 or 2% by themselves not worth my time to focus on. I'm I'm faster to production. I'm a little bit cheaper so I can do more wells in the same amount of time and each one of them is incrementally a little bit faster to production. And more reliable and safer and all of the other I and that's going to be another really interesting thing, I think. I haven't even really thought about this till this moment, but from a safety perspective, like it is so hard to operate a service company in the oil field because the minute you have someone on a location, you're liable for all these things, and you need all these insurance requirements, and all this stuff. And so, it'll be interesting to see like as we start augmenting people or, you know, putting more responsibility on less people in the field, my polite way of saying getting people out of the field. Um it's going to be interesting to see how that impacts the industry and these companies as well because that's I mean, that's a huge amount of money that if you can lever- say, I because my AI makes us faster, safer, better, etc., we have way less safety incidents, we don't stick pipe on any of our well, all this stuff, right? Like, that could have another bottom line impact that no one's even really thinking about. it's funny. Uh previous company, we did some work for a offshore driller uh it's not doesn't exist anymore. And uh but they uh they had enough data that they they they were able to say that instead of pre-reporting the planned amount of footage drilled to the insurance company on a quarterly basis, and then trying to go back and revise it and ask for a refund, what if we could write a bot, a script that basically summed up the footage drilled as it was happening and submit that every month because it's just the only reason it was quarterly was a pain in the butt. Right. And to be conservative, the estimate always had the planned footage. Okay. And then the formula for what your insurance was going to be was a derivative of what was the planned amount of activity in the form of footage. And so, just the act of being able to automate this sort of very simple but onerous reporting task was going to save, you know, like 7 or 8% uh per quarter on their expected insurance costs, you know? And it's like, that's another one of those things where nobody is going to start a company to do just that. [laughter] But like, if if you've if you've sort of bought into the idea that we want to have a digital platform Yeah. for our drilling and completions data and we're going to bring it in and we're going to put it somewhere where it's accessible, what are all the little things that you could you otherwise would never been able to do, but now you could do? Yeah. No, that's that's that and then I mean throw on top of it and you again you haven't mentioned the price and I'm not going to ask you for it, but you've you know hinted at I mean like a certain level I have to assume this is a rounding error on AFE. You know, unless you're talking to the person that you're selling to, right? Well, because again I'm going to get to the year and like again this is just me off the off the cuff and you hear a hundred thousand dollars that's I'm not paying that you know, but then like sit back and like that's nothing on a ten million dollar well, Well, I mean I guess you know the the thing is are you willing to spend the equivalent of a few hours a month to save a few days a month? Yeah. And if we can make them if we can help them believe that that is what they're seeing then then it becomes you know, okay, this makes sense. Yeah. Yeah, and there's always going to be and this is just my life experience, but there's always people that are going to say, well, it's too expensive and then those are the people that will come calling you when they do finally screw something up and that'll be the easiest sell of your life whenever that does finally happen, but it's uh no, it's yeah, it's software is not always the easiest sell even if it's sexy and has really cool features [laughter] and bells and whistles, right? Like it's I guess it goes both ways cuz I mean like and I know plenty people that have production optimization software I can save you one barrel a month or one barrel a day per well, that's you know, and you got seven hundred wells seven hundred barrels a day times however many days and like this easily pays for itself. Yeah, and they're not wrong, but I mean like again then comes like can I even attribute this improvement to this and or is If you do that ten times then you're probably paying too much for software, you know. Are the production engineers and chemical cops assigned to that asset going to be in position long enough before they're rotated to even complete the sales cycle for that software? Yeah. Uh you know, that's a real No, very true. That's it's real thing and uh but anyhow, my sort of overarching theory is that there are a number of software things like that where if you can just lower the barrier to saying yes by saying it's like it's on this platform. You're already doing it. You know, you already have the We're literally You literally just have to say yes and we'll turn it on. Like then instead of only the top three things getting funded, maybe we can get the top seven things funded. And to your point, you know, what if what if it's like, "Okay, it turns out not only could you save 9 or 10% a year on your production chemicals, you know, you'll have half as many wells collapse because they weren't treated. You can afford to put a modem on all the pumps to see when the pumps are broken down. You can afford to put a tank level sensor on all your tanks and see, "Oh my god, let's automate the Like there's folks who some of whom who've been in this building at various times who are have platforms to automate ticketing to go and ask your vendors to come fill that tank back." And it's like it's all possible, but if we can only focus on three things cuz they're so hard. You never get the It feels like you never get there, you know. Well, it's an enablement thing, too, right? Like that's at the root of all of this. If you have good quality data, it enables you to do so much more than if you don't, whether it's And then you start bringing in the partners and the integrations and stuff. It's like, "Hey, everybody now has good quality data and look at what kind of cool [ __ ] we can all do together." Exactly. Which is what tech has been kicking our asses at forever. And I'm glad that people like you guys are finally starting to push the industry forward by having these, you know, partnerships and these integrations and I mean, even I mean, we haven't gotten into the the core of a platform at all, but even just being able to write your own code and have your own templates and all this stuff, right? Like that's very I think that's almost an expectation for, you know, our generation in the oil field and definitely will be for the younger guys. I mean, my [laughter] my dev my dev or I guess devs now, we hired another intern this week. I've got two interns at UAH and they speak a different language basically and you know, have all these They're just you know, they're young kids, right? And so it's like it it doesn't go backwards. The younger the oil field gets, right? It everything becomes more digital, more instant, more, you know, mobile, everything. And so, it's it's just so refreshing to see people that are actually, you know, going out, integrating, providing real value to real problems that, you know, the the entire industry really has. So, So, I was wondering, any um new features that are coming up that you can tease or even new products that, you know, maybe are in the works that you, you know, willing or able to something you're excited about or Yeah, so on the predictive joint side, we're just continuously iterating basically. give us give the audience an update on that as far as like how cuz I have followed you guys, I'm good friends with Kyle uh Doggett, and we've had Jessica on and stuff, but you guys have been, I mean, very active with the the the drilling What what is the appropriate verbage for this? Because I don't I don't want to screw it up and I want to make sure I get it right. there's the core of the drilling platform, which has got about 300 rigs on it. Um about 140 of our own applications, and about uh 40 to 50 uh partner or customer applications, where to your point, they've used the dev center to build their own stuff. And that's theirs, but we're the platform as a service platform as a service does, basically. And then predictive joint is, you know, sort of one application within that. Uh we talked a little bit about the predictive alerts, which is sort of the next step in there. Um in terms [clears throat] of what are we doing with the app? Uh incrementally, of course, we want to we want to integrate with uh other control systems eventually. [snorts] Uh we want to make it easier to share plans between uh rigs. So, a common thing will be, "Hey, what are those guys doing? They drilled so fast that day. Can I get the Can I make that one button where I can go and get their settings that are at at formation and depth appropriate adjustments, and with one button put that into my rig?" So, there's some things like that, which are not sexy, but like very functional. No, it's it's we've we've had multiple conversations about how like there's no there's no I use the term knowledge extraction because I can't think of a better way to put it, but there's nothing like that at the field level where it's like you literally have the drillers calling each other, telling each other oh yeah I hit I hit a gas kick at this, you know, depth or whatever, and then it's like but there's nothing actually capturing any of that information. It's just them talking on the phone or texting or whatever, and it's like why isn't that being spread out through the entire org so that everyone learns from each other and Yeah, we have a we have a application we're we've got coming out called Geo Drilling and the theory is let's get everybody every every person involved with every type of data you can possibly think of from the geologist to the driller and comments and hey I want to attach pictures of stuff and put it in one app that everybody can see. So that's kind of in a similar vein. Um Yeah, and you know, I think where we're going over the next 18 months is okay, we've got it, it can control. Now we're building uh dysfunction uh detection. So stuck pipes one of them, but just auto driller dysfunction in general. Like for instance, for whatever reason differential pressure is well suddenly low in the settings, we're starting to cause saw tooth in by having weight on bit or ROP overshoot instead of just having a little banner that says you're overshooting, let's self-adjust it now. The dream here is that we get to in this time next year, maybe we'll come back, uh bring some cool swag. Uh the idea is that basically you're drilling a well, you're drilling the surface, vertical intermediate, maybe the on bottom rotary part of the curve, and the whole lateral with this thing in control, and the only thing you have to do is put in the BHA tool limits, which we already collect for you anyway, and transitions of interest. So if you're going to hit Brushy Creek That's crazy. and you want to slow down to 225 ft per hour for the first 50 ft of Brushy Creek, you put that in, and there's no road map. It just drills. Yeah. Yeah, that's that's kind of the vision for the the predictive drilling piece of this. How many wells are y'all up to now with that? So I think as of this morning we've done about 135 wells. That's awesome. Um and we've done I guess that's probably around 725,000 ft. Um you know, we've we've fluctuated. We've been on as many as 15 rigs. We're currently on This morning we're on eight. We're on eight rigs this morning and the Bakken uh and Permian. We've drilled some in the Haynesville. Next month we're going back up to Canada. So you know, and we've got uh some step-outs planned down to Latin America as well. So Yeah. That's cool. I love seeing new useful tech actually being adopted by the industry. [laughter] That makes me makes me happy. [snorts] Well wrap it up. What time we got? Yeah, we're right at an hour or so. Okay. It always goes quick. Yeah, it sure does. We could talk about this for days. Um so at the end we kind of wrap up with just like a random speed round where we'll we'll pepper you with just some random stuff. Okay. Nothing crazy. What are What about some uh some books or or even podcasts or anything like that you'd recommend or you like to listen to or read? Uh [snorts] okay. Well, uh I just uh I'm part of the way through a uh single-volume history of the Assyrian Empire, which I previously knew next nothing about that part of the Bronze Age. That was actually pretty interesting. Uh and then uh went went down the the UFO kick recently. So I just read uh uh Lue Elizondo's Imminent. Uh and if if you buy into that, basically uh Roswell was real, they're here, and the army's been and the Air Force has been lying about it for 75 years to us. So that's kind of a fun little side hobby to think about. Real world at all. And is that realistic fiction or is that uh or is it or is it like just kind of a theoretical kind of like Well, how much time do you have? There could be a [laughter] whole 'nother podcast. Uh let's let's put it this way. There's some weird things happening. Yeah. And since 2017, a bunch of Navy videos have come out showing weird stuff. And so as as Senator Marco Rubio put it, either one, there's aliens, it's the biggest story in history. Two, uh it's all secret tech uh and uh or three, we've got a bunch of crazy people with top secret secret compartmentalized information stuff running around the Pentagon convinced that UFOs are real. So, any of the three of those are probably pretty interesting outcomes. Interesting. [laughter] No, that's uh I love I love going down those rabbit holes personally. It's It's a fun brain exercise. I also just think that part of the cons- part of the disinformation campaign is making things conspiracy theories so that everyone that Oh, and now it's not becomes a Oh, yeah. Conspiracy theorist and then all of them become actually The whole thing could be a sci-op. Yeah, come to find out they were all real. Your hobby's a sci-op. laptops and all the other fun stuff. Anyway, um But uh Favorite restaurant in Houston? Ooh. Ooh, all right. I uh I have to say I really enjoy Jonathan's The Rub over there on I-10. Uh and very Have you been to the other one? I I've only been to the one on and back behind it in Memorial because I know there's one off I-10, there's one back I I've only been to the the the one at uh I-10 and uh Campbell. Uh but that's my go-to on the way home. Love the food. Yeah, very reasonably priced. Nice. Well, where uh where can people find you? How do they get in touch if they want to Yeah. set you guys up? It's You can see me on LinkedIn. Um to do with drilling, uh I'm your guy. If it's something else, I'll find the right person in our company to talk to you about it.
Nice. Awesome, man. We appreciate it. Thanks for coming on this Thank you for having Thanks for walking over. Yeah, [laughter] we'll have you guys on the next year with everything. All right, sounds good. updates, man. I appreciate it. Guys, thanks for for listening. Make sure you give us a uh a like or subscribe, give us some comments, and uh we'll talk to you all next time. Thanks. Thank y'all.