Energy Bytes

0:00 And even then, right, there was some of the coding that I was doing that was in Fortran 77. And then they had the Fortran 90 or 95, whatever it was called. Um, and it just showed kind of the

0:12 legacy of the code that existed across those projects, those projects have existed in different forms for, and you were using libraries that have been built, you know, a decade or more in advance.

0:22 When your advisor was in grad school, you know, he's carrying that stuff. And, uh, so I didn't think that much of it, because it was just what I knew, but, you know, yeah, fortunately,

0:31 whenever I would tell it to someone, I was like, holy cow, like people are still using that stuff. No, that's, uh, I had an, a professor in undergrad, some of them also mechanical. And, uh,

0:42 I don't even remember what she was teaching like dynamics, I think, or one of those foundational classes, but she was a, she worked at Lockheed full time. I was in TCU at the Fort Worth. And so

0:52 she worked at Lockheed in Fort Worth. And, uh, she was in their ballistics program and I was like, What do y'all use to code missiles and stuff with? And I was like, What? And I was like, Okay,

1:04 so I can understand the whole management knows Fortran, so that's what you're going to code in peace. But then she also made a really valid point, which is like, Fortran is basically the lightest

1:16 version. Like if you can't get it to work in Fortran, it's not optimized enough, essentially. And it's like, that makes me shit ton more sense when you think about the grand scheme of like,

1:25 resource, and this is a thing that's going to blow up anyway in like money That's right. Yeah, and it's what I'd known about it on by no means an expert, but you know, Fortran was the legacy,

1:36 language of the scientific community. And so when you're doing, you know, high powered scientific computation and precision matters and speed of execution matters, then there was just so much

1:46 inertia for that community around Fortran. And so that's what we all used. I'm not talking so much. We always do this, you can't have the good conversations before the podcast, but we'll bring

1:57 that up. So it'll be fun to talk about, 'cause I think a lot of people, at some point in time have either been exposed to it or know someone who used to write a bunch of stuff in it, and again,

2:08 it's still being used. That's right. It's pretty amazing that it's still being used and all how old is Python now? Maybe 10, 15. Oh, it's been around a lot longer, but hitting like a critical

2:20 mask for the last 15 years, for like, and so, just to see, how many languages have come and gone just in Python's time? Yes. Like, parole and all you're like, R, R, R, R, R, R, R, R, R,

2:31 R, R, R, R, R, R, R, R, R, P. Maybe, I don't know. Depends on who you ask on the internet. Yeah, it's been an interesting week or two. SQL's dead, R is dead. Yeah, everything's dead.

2:42 And SQL's not dead because Snowflake and Databricks just bought by Postgres Yeah, that's crazy,

2:49 yeah, I guess we could probably lead off. That's a good one. Well, welcome to another episode of energy bites. Bobby Newen here got the rad dad. John has a good one. How's it going? Good to

2:59 see you. Same. Got to be here. Yeah, I mean, we're starting to get a little momentum. We are Jeff and a couple more folks in the hopper. So it's all good. Maybe we can get a get back on that

3:07 weekly cadence. But, uh, but yeah, no, I could say I'm lucky to have, uh, Jeff Kremel here with us from a Kremel strategy group, the owner, founder, whatever, you know, janitor, yeah,

3:17 it's all above. Yes But, uh, no, I'm just glad we get you out. Well, thanks for inviting me. I'm, you know, I anticipate we'll have a lot of fun having this chat. Absolutely. Yeah. Thanks

3:26 for working with us too. We've rescheduled on you multiple times. No, no worries. No worries. I'm glad we finally you're shifting now 15, 20 minutes here or there for run practice, but that's

3:35 right. But I get it was funny. Ironically enough, uh, I think Jeff and I, yeah, we met at nape a year or two ago, but then I think I saw you at a swim meet and then kind of. We got this going

3:44 again and then I saw, yeah, I saw Jeff had a soccer, you know, Soccer match or kid soccer game or whatever they weeks ago. So it's funny, the, the small world of like you sports, because then

3:55 I was sitting at Sugar Land, Girl Softball Association last night for a practice and Matt Harriman walks by and so it's awesome. Yeah, no, it's all good things. But yeah, but I mean, I

4:06 sometimes we start with some current events. So we've got snowflake summits going on right now. And I think a lot of big things coming out of that. But I think one interesting trend in the last

4:16 week or two has been, I think it was Databricks acquired a company called Neon, which is like a hosted Postgres SQL vendor. And then it came out during snowflake summit that they acquired crunchy

4:26 data, which is also another Postgres SQL thing. So now people are trying to take those OLTP or transactional workloads and be able to use them within snowflake or Databricks and kind of get the best

4:37 of both worlds. Yeah, I saw, I saw the snowflake one yesterday and I was like, Oh, that's interesting. Yeah, this is very interesting. I didn't even know about the Databricks one. Yeah, but

4:48 it's even interesting and maybe even not that you may have as much context on it, 'cause you don't nerd out on those things, maybe as much as oil and gas stuff. But from a strategy side, it's

4:57 interesting. Now you've almost got these different layers where Databricks and Snowflake sit on top of AWS and Azure, DCP, and they're basically AWS is happy to have them, even with competing

5:08 products, because they're just turning the meter on

5:11 their servers and all their services. But now it's almost like you've got this whole other layer of now Snowflake and them are sitting there, but now they're layering and other services and things

5:20 within that. And it's like this whole like, you know, just bundling. Yeah, no, I honestly first see that happening with all the language model stuff too. Postgres already has PG Vector, which

5:32 is a Postgres plugin. I think CosmoDB is what Microsoft is trying to push. Even Microsoft's index, their AI search index can be a hybrid, but once you get into the language model side, having to

5:48 manage a dedicated. vector store is a huge pain in the ass. And so having it integrated with a traditional database is also very, very handy. So I foresee, at least that's my hypothesis, is that

6:01 someone's gonna come along and start buying a, you know, acquiring the good vector databases that are out there and have to kind of figure out how to integrate them into the more traditional

6:14 database side so that everything is much easier instead of having to maintain two completely different repos, which is not fun.

6:23 So Jeff, how much do you delve much on the SQL side of the world or any database stuff or not? Very little at this point. Yeah, the stuff that I deal with is such small scale and it's really not

6:36 production type data sets. There's more ad hoc, you know, trying to, it's most of my effort, frankly, is around trying to identify which data sets I'm going to use, right? So the first step

6:46 for me is. typically posing some kind of question, I'll work with my clients to pose some sort of question. It's almost always a market facing type question. And inevitably, there's data sets

6:56 that exist that can inform us in terms of how we wanna construct an answer to that question. So step one is make sure we're posing a pretty clearly defined question. And step two is trying to

7:05 identify these data sets and go capture them and organize them and then build some analysis on top. But each one is, and I probably air too much on the side of believing that each exercise is its

7:17 own special snowflake. So I have the equivalent of a real loose database structure that I keep just in spreadsheet form for larger data sets that I'm working through. But I'm now getting to the

7:29 point where I have enough of these - when I use the word database, please put air quotes over to what I'm saying, because it's not a database. That's exactly right. So these largely just data

7:40 tables, frankly, that I have. They're getting large enough. And I have them over enough industry verticals

7:47 becoming more acutely aware that I need to, I'm not far from finding a structured solution. Whatever I have to do though is bring someone in that lives in that space and explore that for me. 'Cause

7:58 what I want is to not to jump too far ahead but I'm pulling all this market data. And the market kind of data that I pull is rarely complete. It's rarely perfectly clean. And so even when I'm using

8:10 APIs to go grab, say the data out of the Securities and Exchange Commission, that stuff isn't complete. It's not perfectly clean. So I have to go in and massage and reorganize. And so I don't

8:21 have a great system for doing that. Like I just have some notes that I take. And I have a sense of what changes I've made and all that kind of stuff. But it would be nice to have a sort of a raw

8:31 database. OK, this is the data as reported. And then a cleaned, modified, massaged version, where I'm very clear about what changes I made and why I make it. And I'm still sort of early enough

8:42 days and small enough scale that that isn't really holding me back. But at the same time, I think if I pass forward into the future and look back in the past and be, okay, what am I gonna sort of

8:52 slap my own wrist over? It's like, man, you played a little too fast and too loose with some of these data sets for too long. Like if you'd put a little bit of effort into structuring that and

9:01 organizing it for down the road. But right now I'm just kind of project to project and I'm still very much also in an experimental phase trying to understand, okay, what sets of questions, what

9:11 domains of questions are my clients most interested in? So I know what data sets I'm going back to over and over Again, and so I've purposely kind of avoided some of the structure related work and

9:22 instead leaned into the exploration type work. But there's never gonna be like a bright line where it becomes perfectly clear that like, oh, I need to go revisit this. So I'm living in a land of a

9:33 lot of uncertainty right now. Yeah, but I mean, it makes sense to not optimize for something that you don't even know needs to be optimized. Like again, it's the whole make it work, make it work

9:40 well, make it work fast thing. So I mean, you're making it work right now, probably making it work pretty well. If we need to make this faster where I can deliver insights more quickly, or if

9:47 you were able to productize in a way some of your insights or how you utilize the datasets, that's proprietary, then maybe you get more into that structured way of thinking of things. I know

9:59 this sounds like an absurd example, but it's something that causes me pain, so I'm not going to resolve it today. But going forward, even something as simple as, I will pull datasets. You can

10:11 pull it from the USAIA, you can pull it from the IEA, you can pull it from the International Monetary Fund, or the World Bank, or just countless sources, right? And say these datasets have a

10:21 geological or geographical, rather dimension, where even something as simple as what continent this comes from. You can see some disparity there, but then when you get down to the level of

10:31 specific countries, or if you talk about subcontinental regions, you'll get apples to oranges like no organizations to map is exactly the same. And so I'm going to If you need, I already have,

10:41 it's a Cluji, you know, kind of. Hacked up version of a map that allows me to get where I need to get with it But I know that going forward if I play across enough industrial domains and enough

10:51 geographical Regions like I'm gonna need something pretty slick that allows me to Translate in a reliable way from one region to another

11:01 Exactly, so today it's all largely a manual exercise. It's just it's it's not a super compelling example But it's one small example of I seem to have countless of these where I'm pulling enough

11:11 There's a wide enough diversity of data sets that I'm working with that I will need some translation that allows me to fluently and fluently Communicate between those and today. I'm just kind of

11:22 relying on brute force because it's small enough scale But going forward is gonna have to change like I said I think even when you're delivering some of the stuff for a client or even just doing I

11:30 know you have some really awesome analyses that you just public you know post in public but like you Like you said some of the data is not clean enough Anyways, or you've got a you're at some level

11:39 even if you've got that those data pipelines that are getting you 90 the way there that last 10, you're gonna have to like get in with a fine tooth comb and make sure these numbers are right because

11:47 when you're making recommendations potentially on millions or, I mean, potentially billions of dollars, like you have to, you know, you have to know and trust what you're delivering. That's

11:55 exactly right. And it's something, it's such a powerful, the way you said that. It's a conversation that I get into with my clients. Of course, it lives, you know, sort of the epicenter of

12:04 this challenge is in the data domain, but it extends far beyond just data, which is when you're trying to make any kind of decision into business world or beyond, right? There's a spectrum on

12:16 which you can live. And one extreme is I'm gonna introduce as little rigor in engineering as possible, right? I'll just rely on anecdote or heuristic or what are people around me doing and I'll

12:26 just make a similar decision fine. That's like the least structured, least rigorous form. You can imagine the other extreme is frankly just over engineering. I'm gonna wait to make a decision

12:35 until I'm convinced that I've pulled every relevant piece of data and I have processed each of those pieces of data in a way that gives me confidence. that whatever decision I make is gonna be, you

12:44 know, as good a decision as I can make in this moment. And of course, when you optimize in that decision-making process, the optimal lives somewhere in the center. And so I try to, when I'm

12:52 working with clients, give them a sense of, yes, all of my recommendations will be data supported and data informed. And yet, one thing that I am also sensitive to is I do not want to introduce

13:03 so much rigor, so much data, either in the volume of data or the analytical machinery through which I process that data, where your decision-making process slows down, where we're not meaningfully

13:13 reducing the uncertainty, like we're just going through an exercise because we're engineers and we like having clean data and we like having relatively precise answers, and but that's when we're

13:21 making capital allocation decisions on the order of tens of millions, hundreds of millions or billions of dollars, you quickly get to a point where you're over-engineering. The uncertainty, you're

13:29 not reducing it by playing with the data. You're adding more variables now, and now it's like, which of these variables was wrong when my model's wrong? I mean, I think we kind of ran into that

13:37 at Grace and Mill, I think one year we kind of missed budget and I was like, well, why? And it's like, well, you know, the. Finance's model is wrong or, well, no, your inputs are wrong.

13:45 And like, we had to get to the point where now we were measuring the inputs and what people said was important, you know, but now if those were within a reasonable bound, like when we do look

13:53 backs, now's like, now what's wrong in the model or vice versa. But I mean, I think you had a good post. I think we engaged back and forth a little bit on it of about just about forecasting in

14:01 general, like, you know, it's a necessary evil, but it's like, they're never going to be perfect. I mean, the whole all models are wrong, but some are useful thing. Yes Well, your example is

14:09 powerful because, uh, and that's a one-on-one conversation with a client recently where he was saying he had a nice little anecdote where he's like, you know, they were looking for oil price

14:19 forecasts out years in the future. Uh, and he says, you know, someone needs to come take my pencil away for me. Cause I'm sitting here making1 per barrel adjustments out in like 20, 29. And

14:29 this is too much. Now your, your example is really, really good because, you know, like you say, so we're going to introduce some rigor, some machinery around this process And what that does is,

14:39 by no means does it ever give you a perfect forecast. But because you have that infrastructure in place when you're making this decision, you can now recognize either in advance or at least after

14:47 the fact, okay, which of these parameters really is important, like really is consequential in determining these outcomes or, you know, any deviation relative to plan. And so now we can go back

14:58 and have more structured conversations. Or if we need to do, you know, a new data gathering, data exploration exercise, we can go do that with the confidence that if we do capture data in this

15:08 domain, it really is going to help us achieve better outcomes. And there's other parameters in there that can be as noisy as you can imagine. And yet somehow the outcomes are just not at all

15:17 sensitive to that. Then great. Like we don't need to waste our time. Exactly. We don't have to waste our time going through all that. And I don't know how much you get into like machine learning

15:23 or data science stuff. And I dabble in it when I have to. But like, I know, I think we did a XGBoost model, but like the great thing about some of those, they give you like a variable importance

15:32 thing. And then like as you're working through it, like you can say, actually, I don't actually need these. So now you pull those out and now you can now retrain it on these five variables. and

15:39 you can get really down to what's actually moving the needle for this. And then I guess really what probably always comes down to the Pareto principle, right? You know, 80 of it's coming from 20

15:48 of the inputs or vice versa. You know, but that's right. But in this case, the machine learning algorithms are the ones that are helping you, you know, build the confidence that we understand

15:57 which data sets really do move the needle, right? So before those algorithms existed, yes, you can, you know, do some analysis to help get you there. But in a lot of cases, it's just, you

16:07 know, intuition Like, I feel pretty confident that these particular parameters matter, and I have a sense that these don't. But now we have more confidence because of the algorithms that exist

16:16 that can really point us in that direction. That's something that that's a theme that if I keep using the word confidence in this conversation, it's a theme that I've settled on with my clients, is

16:24 that the way that I use data, the way that I try to then, you know, inform that the insights that I communicate and everything else that I share is, you know, I'll tell clients, you can get

16:34 value from working with me, even if you end up making the exact same decision.

16:40 that you otherwise would have made if you had not retained me. And the difference is, I can bring this data to bear, these data-supported insights, organize our thinking, ask some questions, and

16:49 now you have more confidence to make the same decision you would have otherwise made. Well, what's the value of being able to make a decision more confidently? You can make it more quickly, you

16:56 can enroll stakeholders in your decision more effectively, whether it's board members or shareholders or employees or competitors or vendor or whoever, right? You can enroll them because you have

17:05 more confidence, move more quickly, capture these gains, more reliably. So even if you were gonna make the same decision, having using data the way that all of us use data to help you build

17:15 confidence in that decision, it doesn't necessarily mean that you must change your decision to have gotten value out of it. The added confidence comes with a lot of value. I would say, even

17:25 sometimes like the wrong decision, but done with, you know, like gumption or like with like, with confidence behind it, can actually be more effective sometimes than like a lukewarm, even if it

17:35 was a better option, but you kind of had a lukewarm. you know, reaction or confidence behind it, then it may not actually go as far, so. That's right, and you know, so my background, right,

17:43 so my education's all, you know, bachelor's, master's, PhD in mechanical engineering. So I have this very academic, you know, fueled background and brand, and I live in the world of data. And

17:52 so I know that, you know, at first pass, where I can be very quantitative and analytical, and you know, the work that I can do, probably, you know, can come across as cold, just, you know,

18:02 very objectively built around that data. And yet, so one thing that I try to make sure when I'm engaging with clients is, there's this human psychology element that lives adjacent to all this data

18:12 work. And I think that's the reason why what you just said is true that, you know, when you make a decision with confidence, right, a sub-optimal decision is different than a blunder, right? So

18:20 a blunder is, well, we'll set that aside. But you're talking about a sub-optimal decision that you make with, you know, confidence compared to something that might be closer to an optimal

18:29 decision, but where you lack confidence. It is that human psychology. There's a lot of employees, there's a lot of, you know, partners that rely on You know a communication of the vision of like

18:38 why are we pulling this law? Why are we deploying capital in this direction? Why are we standing up this new? You know IT infrastructure the way we are and if there is a compelling and vision

18:49 shared with confidence People have a lot more that they're just quick to buy in they're quicker and then you know even if you take that part out a lot of folks just feel more Excited they feel more

19:02 validated You know they themselves have more confidence in the management team that like we understand why we're making the decisions that we're making So it's you know for as much as I focus on all

19:10 the data and the knowledge and I love that stuff There is a human psychology element that again is adjacent to that But when you marry the two of them together is where things really start to become

19:20 magical Yeah, I think I think you were getting into like being like really detailed but like One thing I noticed when I first got into oil and gas working with a lot of reservoir engineers Which is I

19:29 think falls in somewhere where there's a lot of planning and there's uncertainty and all these things but I mean, both of you are mechanic engineers, right? And I feel like on like the true

19:37 engineering side, you can actually like design a system with a confidence cause the rules of physics apply and mathematics. And I can do this and I can that's asset and like give you a really good

19:46 idea of, if I put this pipe diameter with these conditions and PBT or whatever, like it's going to do this. Like I know that, but now you get into reservoir engineering and planning and now

19:55 everything's like the confidence intervals and uncertainty and statistical probabilistic type stuff. And it's like a, I mean, that's engineering, but it's a totally different mindset It flips

20:05 everything we did in school on its head in most cases, right? 'Cause it's like, I mean, I struggled with that when we were together, when I was putting together product business plans and stuff

20:15 like that. And it's like, well, it could scale like this or it could scale like that. I just need to, I just need to have a way to justify some kind of hypothesis either way, right? Like, and

20:26 that was one of my biggest takeaways of it is it's like, you can, the decision ultimately could be wrong, but as long as you had some, strategic enough to have ways to justify that, then coming

20:37 back and looking at why was that decision wrong? Well, it's because we made this assumption that wasn't correct. And it's like, okay, that's fine. We won't make that assumption again in the

20:45 future. Whereas in the physics and engineering world, it's like, here's the formula. There's the answer. And the only change in the answer is going to be your units or your sig figs. Like that's

20:56 it. And so that's the that's your variance that it might might be a decimal point or, you know, a tenth of a percent or something like that. But the flip side of that is we also designed for

21:06 failure too, right? Like look at cars and things that are supposed to fail or designed to fail over time, unfortunately. But it's yeah, it's it's very counterintuitive to them to what we are

21:19 trained on basically our whole career. Exactly. Right. No, you need the answer. The answer is there. Go find it. You can get there a bunch of of different ways depending on what you want to do,

21:28 but the answer is there too. The answer is definitely not here. You got to just figure out what the best guess is based off a thousand variables, go figure it out. Well, and what you said about

21:38 assumptions is spot on. So it's something that I lean hard into, is that, okay, I'll create a multi-year oil price and recount forecast. And it's not that my crystal ball is somehow better than

21:49 anyone else's, but because I'm clear about what assumptions are embedded in this, then we can interrogate those assumptions and wind up with something that we have more confidence in. 'Cause again,

21:58 if you don't do that, make a decision, you go back after the fact, it's like, I don't know, yeah, my gut just told me that that was gonna be the right call, it turned out not to be the right

22:04 call, so what do you do differently next time? It's hard to say, but when you have this clarity of assumptions and then get back to the data, 'cause I'm sharing with clients, right? The clients,

22:14 I'm working typically with senior executives in energy companies, right? So obviously very capable, well-educated, smart people, not always super deep engineering or analytical backgrounds, but

22:25 to a person in part because they're self-selecting, having worked with me, they care deeply about data,

22:31 And so they do want confidence and clarity around what data sets are we using and why, like what do those data sets tell me specifically? And then when I share that with them, that gives them a

22:41 chance to understand, okay? A, do we think just working with the right data? B, are the insights that he's pulling out this data set, does that jive with what my team and I would have pulled?

22:51 But again, it gets back to the fact that that clarity exists for them, it's not a black box. It's not just, hey, Jeff happened to have a bunch of informal conversations with smart people and now

22:60 believes X and he's sharing that belief with you. I can show my work in a sense. Yeah, that's exactly what it is. It matters. It's proof of work, right? Yeah, transparency. Yeah, 'cause it's,

23:09 again, reviewing it without having any of that written down anywhere or even thought of, right? 'Cause you can make a decision without going through that process of like, oh, well, what are the

23:19 risks? What are the uncertainties? What are the knowns, all of that stuff? And then, yeah, you're really screwed 'cause that decision gets made and then nine months later, you get the results

23:28 or whatever It's like, oh, well, that sucked. That wasn't right. why? No one knows, you know? That's, I think that's one of the big things with

23:37 the industry specifically is because everything, the first thing everyone thinks about, especially management and up is risk. And then the next thing is uncertainty, right? And they're basically

23:47 tied together. And so coming from the traditional engineering world into the industry, you have to be, you have to understand that that's how this works, right? Nothing is certain. It always has

23:57 a risk. There's always an uncertainty. It's just about managing those and understanding it before you go and make the decision, right? You know, I mean, you think about that with any big

24:07 engineering project, right? Like, it's all risky. The best decision if you wanted to think of risk was to do nothing, right? Like, that's the least risky thing to do, but that doesn't spawn

24:17 innovation. That doesn't build crazy bridges or new things. And so it's, yeah, it's pretty interesting. Yeah, I mean, like, I think I wrote that blog, but even said it before, like, you can

24:27 put the the whole way down the fairway. Right. You just, you wouldn't get there very fast. I'll keep you in the fair way though. Well that's part of not to preach to the choir here too much, but

24:37 that is part of what I have appreciated most having been drawn into oiling gas. When I launched into oiling gas by no means was it because I found it super compelling or a tricolon, that's where the

24:47 job was at the time and so I took it. And one thing I very quickly came to appreciate, and we're seeing it today when you talk about the context of the energy transition and what role do you expect

24:58 oiling gas companies to play in the future around oiling gas proper, but then more broadly. And very quickly people bring up that these oiling gas entities have an enormous amount of experience in

25:09 pursuing these large scale capital projects that have considerable risk, right? In an oiling gas you live in a world where the consequences of failure, you're talking loss of life, you're talking

25:19 considerable environmental harm, you're talking the kinds of consequences where is what you're saying, that it makes sense why every big decision you start kind of with the assessment of risk,

25:30 because when things do go as badly as they could go, it's truly catastrophic. And yet, right, for as consequential as failure is in this industry, it has not caused the industry as a whole to

25:42 putt all the way up the fairway right there, still enormous projects and victories that have been achieved. And that's in celebration of this industry has found ways to understand risk, articulate

25:54 it, measure it, mitigate it, and still pursue these large high profile endeavors. And man, it's a super exciting thing to be a part of. That's the interesting part though from the engineering

26:04 side, right? Is it's like you design for the risks, right? And like, that is the true engineering marvel of what we do. And it's like, yeah, you're designing something that's five miles away

26:14 from where you're standing to control it. It's like, and you can't see it, you can't touch it, you can't simulate it. I mean, you can kind of simulate it. But it's like, that's why I think we

26:22 have so many really brilliant engineers because the problems we tackle are truly kind of magnificent

26:29 I also also find it just completely ironic, though, like the dichotomy in the industry is wild, right? 'Cause you have these like,

26:37 you know, frontier pushing companies, but then you have a lot of companies that are like what you mentioned earlier, where it's like, well, what are our neighbors doing? We're just gonna copy

26:45 that. Yeah. And it's like, but they both work in some form or fashion, right? A lot of the time. Yeah, these are RD department. Yeah, that's probably 80 of the industry, you know. Like,

26:57 it's, yeah, I'm curious to stand It's speaking of news. I'm curious to see what the Utica does now that EOG's buying players up there. Yeah, Encino. But, yeah, it's such a fascinating, the

27:10 whole system, right? 'Cause there's so many components to it. And then some of it is so counterintuitive, yet still somehow manages to work. And then - That's right. So let's talk about your

27:21 coming, 'cause we never even really formally said when your company does, yeah. I think, you know, the context clues, people will probably put it together pretty quick, you know, what do you

27:30 do with criminal strategy group? And then maybe let's talk about like a typical project if that exists and like, what does that look like? And then maybe the tools that are part of your tool belt,

27:39 you know, and getting to that answer. I mean, that sounds like a lot of Excel, but you're talking about pulling from APIs and all this stuff. So we'll love to kind of dive into all that.

27:46 Certainly. So the criminal strategy group, I called KSG, I launched it at the beginning of 2024. So it's been, you know, about a year and a half as we sit here that I've been doing this. And I

27:55 largely work with executives at energy companies to help them use data to inform their corporate strategy. And this happens kind of in two different ways. One is much more relevant to this

28:08 conversation. That is the sort of market research fueled corporate strategy support. And, you know, one small example, we can go through more depth. One small example is, you know, I've worked

28:18 with a big

28:20 oil field equipment supplier that was looking to make a capital allocation decision that had a five year return horizon This was the case where this person was looking for - oil price forecast or

28:30 account forecast out to 2030, knowing all the uncertainty that exists there. But you have to make a capital allocation decision today. It does have this five-year return horizon on it. And so you

28:39 can either choose to engage formally with a forecasting exercise, or you could punt. And but you still have to make the capital allocation decision one way or the other. So I helped build up a

28:49 data-supported forecasting infrastructure that gave us a sense of what oil prices and rig counts might look like in the future. That in itself was valuable, but then also, and this is where I was

28:57 talking about the human psychology element to what I do, just the conversation being able for him and me to sit down and talk through, because he has a whole team of folks that work on this kind of

29:06 stuff. They had an in-house view, they have some third party data subscription services that offer them some look out into the future. The difference was that he and I could sit down and talk

29:16 pretty deeply through this is how these markets function. I could explain this is how I've built the tool. And so knowing how sensitively pricing depends on supply relative to demand. This is how

29:27 we expect those things to evolve going forward in time. And so that way, when they then make their capital allocation decision, which I really wasn't involved with, it was more just helping them

29:35 build the forecasting piece that they could use to make that decision. He has a lot more confidence and perspective. And there's someone like me who - I don't know the particulars of the capital

29:45 allocation program that they're assessing. I don't know the internal politics who's landing on what side of what, right? I don't know any of it. And so I can just come in with a third party, more

29:53 or less objective perspective that they can then layer in and have some confidence that's like, OK, we're not. Yeah, if you're wildly different than what they were thinking in house, I was like,

30:01 oh crap, maybe we need to revisit this. That's exactly right. And that comes with enormous amount of value. And then if I'm spot on with what they were thinking, that comes with value. And so

30:08 there's really, again, there's no way that if I'm bringing the kind of data and crafting those into the insights that I'm confident I'm crafting them into, there's no way you don't get value from

30:17 it, no matter what the ultimate outcome winds up being. So there's this market research-fueled strategy support The piece that's less relevant for our conversation is a learning and development

30:29 piece where I build and facilitate workshops for either aspiring leaders, so like leadership development programs. And I've built some workshops to teach them how to improve their own strategy and

30:39 finance acumen, how to build their own market awareness. But then I've also done programs for existing executives and even for board members. There's cases right where board members get selected

30:49 because they have their backgrounds in IT or it's in HR or something like that. But they know sitting on the board like I want to be able to participate in some of these more formal financial reviews.

30:58 And so I need the vocabulary and an understanding of kind of what models and frameworks exist to have these kinds of conversations. It doesn't require a year long effort to study this. Like you can

31:09 learn that relatively short order. These are very bright accomplished people. But for me to package it in a way where they can learn it pretty quickly and then we can engage and ask questions and go

31:16 forward. So there's that learning and development piece. I have a sub stack newsletter that I use to try to more broadly for any energy professional to help them improve their finance strategy

31:25 acumen. So that all lives in my learning development routine. What's the handle? Give a shout out to. Foundations of energy. On sub-stack. On sub-stack. Yeah, that's a good read. Well, thank

31:33 you. Yeah, it's a lot of fun. And

31:36 it keeps me sharp. You know, what we were talking about here when you were kicking off with the newest developments in the world of data engineering. You know, there's so much happening in energy.

31:46 And for as much, for as excited as I get about it and for how much energy, no pun intended I have to. Yeah, so it's like, okay, that what I need to do is, yeah, there's some some high

31:56 consequence developments that I force myself to really lean into unpack, build up a foundation, okay, why is this happening, the way that's happening, introduce the vocabulary, introduce the

32:04 context, throw some data at it so that we can, this is not just a bunch of hand waving and all of us running our mouths, but there's real data to suggest that something interesting is happening

32:11 here. And then I do that. So that's all in that learning and development domain, which keeps me sharp, it's a lot of fun, it helps me engage, you know, where this consulting work is more very

32:20 senior executives in the energy industry world. This learning and development stuff, get to interact with, you know, all. All different levels in the industry, which is fantastic. And so, yeah,

32:30 that was one example I gave you, the forecasting that there's other examples we can walk through, but it's typically the case where an executive comes and they have a sense of at least what question

32:38 they're trying to ask. And what's interesting is no one has come to me yet, like not having done their homework. They do have an internal view of something. And what's really interesting and makes

32:49 it even more fun for me is that they recognize like, okay, to dot highs and cross T's here, we really do want a third party perspective. Yeah, and it makes a ton of, like even myself, as

33:00 someone who is dealing with all of our data and lots of aspects

33:07 of it, I recognize that I have biases that I'm not even aware of, right? And so it's like,

33:15 that influences how you deal with the data and what decisions you make and how you go about making those decisions and all those things and typically when you're in the weeds of it, a lot of the

33:25 times you can't see the big picture piece. And so having a third party that doesn't have any kind of bias or, you know, there's not a horse in the race or anything internally, you're not, you're

33:36 not going to be penalized going one way or the other. Whereas if you're internal and you're going against what your boss thinks is right, then there's a whole political piece to that, right? And

33:46 so it's a, yeah, just like that's honestly how most big decisions should really truly be made, in my opinion, just because I can make the plot look exactly how you want it to look based off what

33:60 you want to represent. And that's the biggest, like we saw that with politics, you see it all the time with news stories. You saw it in the show revolution and then like, Yeah. I mean, you had

34:08 Equinor, Sadowoils, like Mayacopa, and like, you know, I make the price decks, say, 120 bucks. Yeah, yeah, like. Or use Aubrey's strategy and just post all those big IPs and don't worry

34:19 about the decline curves falling on their face a year down lower.

34:23 Yeah, it's, no, it really is. Like, anybody that has dealt with data in any kind of capacity begins to very quickly realize, like, I can make this read exactly how anybody wants it. You just

34:36 tell me what you want. Yeah, I mean, so that's an outlier, it's not a real data point, you're like. Yeah, so that comment, I was just about to say for both of you guys, 'cause you live in

34:42 this world even more deeply than I do, is that one sort of small scale soapbox that I end up getting on is there's an abundance of data in the world, right? And so, you could almost like, exactly

34:55 what you're saying, John, you give me a narrative, I will go find a handful of, and then it's not just one, I'll find a handful of data sets and I'll triangulate in a way that supports exactly

35:04 that narrative. So what do you do with that? Okay, one option is you can just be like, I'm just not gonna do any of that work. I don't think that's a great answer. But the other is like, okay,

35:12 you have a hypothesis, you're trying to answer a question, you pull me in to support in

35:17 almost all cases, my clients will not share their, I would rather not know. I was gonna say, it's better not I don't want to know the hypothesis. And so then what gets interesting, not just the

35:25 conclusions, okay, what does Jeff recommend or whatever, but why did you, Jeff, why did you choose the data sets to look at that you did? And when you make assumptions like, why those

35:33 assumptions and not these others, will you wear this other data set or why did you, make whatever decision you did or didn't make? That in itself, there's enormous value in just interrogating that.

35:43 And like you said, being so deep in it, you already know what data sets you care about and which ones you don't care about. But occasionally it is kind of interesting to like, let me pull someone

35:51 else who lives in this world as deeply as I do and they may have different convictions around the waters on or whatever else. Yeah. Okay, so now like, so someone's come to you with a hypothesis or

36:01 a question that they know they want or you've kind of agreed on the question like, now you go find these data sets and where are these, you know, living? I mean, are these, you know, do you go

36:10 get an export of a website? You sound like you go to some APIs like they were like, and what does that look like as far as aggregating that? Can I preface that with what are generally speaking are

36:20 most of the things people are coming to you about. I assume they're tied to some kind of capital allocation at some point in time, regardless of if it's an operator or a service company or whatever.

36:31 There's lots of money being spent to run and maintain these businesses. And so is that generally the case? It's, hey, we're gonna be scaling up this product or be scaling out in this field or

36:44 whatever. Is it, okay. Yes, 100. 100. So yeah, I'll get, it's exactly that's the premise of all the questions that I get And so I'm pulling data sets right, knowing I'm trying to study the

36:56 implications of some capital allocation program. One big data set that I get is a bunch of data out of the SEC, the Securities and Exchange Commission. And so there's a

37:06 framework called XBRL, the Extensible Business Reporting Language Setup. And there's a nonprofit entity that manages all of that. And I think I forgot what year it was, but all publicly traded

37:16 companies in the US at least are compelled to file electronically, in addition to the kind of legacy. analog stuff. And so they're through XBRL, there's an API where you can get in and pull, you

37:28 know, as much data, frankly, as you want to pull. There's, are you pulling any data from like Fred or any of those other? Yes, exactly. So I'll do from Fred, I'll do, you know, directly

37:38 from the EIA, they have their own API so I can pull that stuff that way. And, you know, from each of those, right, there's different versions of that example I gave earlier about just the

37:48 geographical regions that very simple, in fact, the simplest, different oil and gas ENPs, this is kind of trivial, but it's an example that has some importance. You know, they'll call revenues,

38:03 one of them will call revenues electronically. The other one will call them, say, net sales. And it's the same exact thing. It's just different companies have different taxonomies, different

38:12 vocabulary. And so I need to maintain my own map so that I know when I'm pulling net sales or revenue, the exact same thing. but it gets a little more nuanced as you go down through each of the

38:22 financial statements and how stuff is bucketed and segmented and all that. So I'll pull data from APIs either for

38:31 XBRL or through Fred or through yesterday. Sorry, you're pulling the data from the APIs, what are you using for that? Almost always, either into Excel or into Google Sheets. Okay, but like

38:39 you're connecting, I mean, you're making the API calls from like a get data call. Yeah, in the case of XBRL, there's an add-on that I can do directly in Google Sheets and it will make that call

38:48 So I can go in, so what I end up doing is the API call, right, is they give me a function, I forget the name of the function, but I now, you know, craft this URL. It gives me a custom function,

38:58 I feed that URL into that custom function, and then it gives me all the output on the back end. And so not to bore you with all my details, but to build - That's what we're here for. OK. But to

39:06 build this URL, right, is that there's a bunch of different parameters that go into it, where I can tell, OK, what field do I want, what year, what period, you know, is it first quarter,

39:15 first half of the year, third quarter, whatever This is like an O-Date API, almost. I don't know. Okay. I don't know. You can add filters onto the URL through OData, so I'm wondering if it's

39:24 in - Yeah, so in this case, I'm manually constructing, so I'm telling

39:32 it what the time period, what the name of the financial field is. Some of these fields have dimensions, so in the case of revenue, when it doesn't have a dimension, it's just a consolidated

39:41 number, so you'll get corporate revenue for ExxonMobil. Well, that revenue field can also have dimensions, and the dimensions might be it's different geographical regions So I can build up the URL

39:49 where I specify, okay, I do want this regions results or whatever. And then I build up the URL that way, call that function, and then it gives me a big dump. And this is where I should be shamed

39:59 to admit this publicly, but it's just what I do, is I'll take that big dump of data, copy and paste it into just a big data table that exists in Excel or Google Sheets. And then I'm creating pivot

40:08 tables off of that, and then using my git pivot data functions to pull in that stuff in a bunch of different sheets in a bunch of different ways. Where it gets a little complex, and where I would

40:15 love to talk with people know what they're doing here. is that when I pull a lot of this raw data, I then almost, very frequently, I should say, almost always, very frequently. I will have to

40:25 do further computations around that data to get what I want. A very small example is when you pull stuff off of a balance sheet, the balance sheet is almost always presented where you get the year

40:38 to date data. So in the first quarter, you get the first quarter balance sheet data, in the second quarter, you get it as of the end of the second quarter. And so if you want to see how the

40:45 balance sheet changed in the second quarter, I need to just do a delta of what I got at the end of June to what I had at the end of March. Fine, just a little bit of arithmetic or whatever. But

40:54 that kind of stuff adds up as I'm doing different computations to it. So the API is feeding me good raw data. That's not always useful in the form that exists. I have to do some post-processing.

41:05 And it gets just a little complex based on the time frames with which the data becomes available, whether it exists in a dimensionalized form or in comprehensive form. And so that's what takes me a

41:17 fair bit of time is organize my data tools so that I can perform all that analysis. And at the end of the day, I get to something that's like, okay, here's what CapEx looks like for a bunch of

41:27 ENPs by quarter, going back to 2014 or whatever. And then, once I have that chart, writing about that chart takes no time. It's almost clear what, okay. But it's just hard to get to exactly

41:38 what I want to get to. That's all the data work in advance. Yeah, that's what I like to tell people. And they're like, oh, it should be easy There's an API, you're like - Have you seen APIs?

41:46 You go do it. No, we were working on an integration right now for a client and they're using a production, I won't name the name, but they're using a production accounting, or it's just a

41:58 production software, right? They're field guys, they're going out there, they're doing tank levels. Okay, like the well tests and all that stuff. And it just goes into this very simple,

42:08 seemingly database. And they sent us the API documentation

42:15 items, there's like stops, routes, and something else. Okay, yeah. And so I was like, oh, there's not like a wells table, like a master wells list that these are all, nope, there's not none

42:25 of that. There's no API numbers included in any of the stuff.

42:30 And then like, it's just a mess, right? And we ended up logging into the front end of the site, right-clicking, going to Inspect, and then using the Inspect API calls to back out exactly what

42:42 the actual API calls we would need to be doing in order to figure out explicitly what they were, 'cause the documentation was such shit that it would have taken forever to figure it out just that way.

42:53 But yes, this happens all of the time. Anytime you use an API, you immediately go, Why am I using this API? I just want a parquet file. Just give me all of it and I'll figure it out at least.

43:04 Or give me database access. Yes, good lord. That's right. That's, yes, it's something that until you do it, You just really have no idea how painful and annoying and teeth. pulling. It can be

43:15 exactly right. And then yeah. And then the next step of that is it's like, okay, well, this data is good, but I need to do a bunch of stuff to it ultimately to get to my end results. And then

43:26 it's like, okay, well, where do I do that now? And I was actually going to ask you, do you do any of the, do you use any of the scripts or the scripting functionality in Google Sheets at all? I

43:37 don't, I don't. And you know, when you were asking that question, it's it. This is something where I did use generative AI somewhat recently is that, you know, I have all this data and I know

43:47 that I, in principle, should be able to get Python to do a lot of what I need to in a much more graceful way, right? Because my data tables can be, many of them are, you know, 30, 000 lines or

43:57 more, and I'll have, you know, a dozen columns, 15 columns, whatever it is. And so even that is slow enough, or if I'm doing it natively in Google Sheets, it'll take, you know, a few seconds

44:05 of not a few minutes occasionally to update. And so I know there's, you know, these, you know, tools out there that'll do that. So I got on to the generative AI tools and asked to - write some

44:13 code for me. Here's what my data looks like. I gave it a sample Excel file and said, you know, write something that does more or less what I need it to do. And I was slowly iterating toward where

44:23 I needed it to get. But it was also became clear. It wasn't quite where I needed it to get. And that the moment that I got it where it felt like, okay, it was going to be where I needed each

44:31 vertical was going to be nuanced in its own way, you know, whether it's chemicals and ENPs and oil field services and refiners and I just realized, okay, this is going to be too big. Like, I'm

44:40 relying too much on this generative AI tool I don't have enough of my own skill set embedded here to, you know, wrap this up. But going forward, if I, you know, professionalize this and, you

44:49 know, grow into what I think it can become, I'll have to take all this much more seriously. Well, now I ask because I think it is incredibly slept on in the Google suite that, you know, all of

45:01 those tools, Google Doc or Sheets or forums or even your Gmail and stuff, they They all basically have like a code, a dev environment back end. and I'm, God, I can't remember the name of it

45:15 'cause I haven't used it in a while, but I used to use Google Sheets all the time as my, like, POCs. I could give - The G-Script or something like that, or - I'm already starting to jump Davey's

45:24 about it. Yes, God, yeah, Davey's built a whole company using it. It's basically JavaScript, isn't it? Yes, it's just JavaScript-based, and you go into Sheets, and you go to tools or

45:36 extensions, and you click into it, and then it pops up a window, and it's just a little IDE, but you can write your functions in there, and then you can, I was writing API calls in there that

45:46 would return into my Google Sheet, and I was just using my Google Sheet as a pseudo database to be able to look at it, clean it, figure out what ETL or any of that stuff. I was like, Shit, this

45:55 is pretty slick, and I did it all through GPT, because GPT can write it very well, and so I assume Gemini now is probably very - Yeah, I mean, it should be very - Better read it. Yeah, right.

46:06 But yeah, it's like those things start making it like, Oh, do I really need to have, like, figure out how to stand up a production app. and do all of the shit, or can I just quickly iterate on

46:16 these one-offs? That's what I was doing, right? Like I was just prototyping different stuff, or I just needed a data set to do something with it, and it's like, I don't want to write it in

46:24 Python, and figure out how to get it ultimately to where I need to use it, or I'm gonna have to share this with somebody anyway. I can just share it directly from a Google script, right? God,

46:32 that would be huge. That's, I'm gonna definitely support that. Yeah, we'll go play with it after the podcast is over. We'll show you some of the stuff. Yeah, I'd love to check that out. That's

46:39 pretty cool, but even more externally, we had Jacob Mattson with a - Mother Duck, I was, the whole time you were talking, I was like, We're gonna get you on Mother Duck, and you're gonna be

46:48 like,

46:49 This is awesome. 'Cause now DuckDB has like the DuckDB G sheets, where I think you can write two Google sheets from there, but you can also pull from Google sheets down into DuckDB and then massage

46:57 it. Yeah, so Mother Duck has just hosted DuckDB, but I've used Mother Duck pretty decently, this last, whatever, six months, just for some prototyping stuff. And it's incredible, 'cause you

47:10 just point it, you can give it a JSON, you can give it an XLS, you can give it a CSV. You don't have to write a single line of code and it'll automatically generate the table for you. And then it

47:18 also has a language model trained for its flavor of SQL so that when you're trying to make a call on the table, if you mess it up, it'll automatically correct it and be like, did you mean this?

47:29 And I was like, yep, of course I did click. Oh, there's my table, perfect, it's very nice. I will say the only nuance part of that is there, the mother duck API is very different than a

47:41 traditional database API. It is not just a database, like your typical string, username, password,

47:48 it's a connection, not the API, but connections. It's not just a traditional connection string, which has been interesting to figure out. But because they flip between, like, you can run, like

47:59 it'll run the compute on your laptop if, where it can, and then if it needs more compute, it'll offload it to the cloud, but it's kind of a seamless kind of - Yeah, it's basically built for

48:08 analytics, too, or doing. transformations and stuff on the data. So it would be a cool one to experiment with. It's very easy for anybody to just jump in and kind of start using if you're mildly

48:20 familiar with any kind of database, which like, wonderful. I don't know, I mean, I don't write SQL, especially now. Thanks, GPT.

48:29 But yeah, it's a very handy tool for just like, hey, let me see if I need to see this data in the database and then do something with it Wonderful. But I guess getting to your stacks. So now,

48:40 like, I mean, like, you got the data in, you know, through any kind of API call or even like, you just take a Excel that you were able to export and slam it in there. But I've seen some of your

48:51 charts, like on what you post and they look really nice. I mean, is that are those Excel charts or cell charts? Okay, so yeah. So how do you have that set? I mean, are those do you have a

48:57 bunch of templates like that, keep it really nice and like that you can just roll into or sometimes I do other times I'm creating them on the fly. It's one of the reasons I like Excel so much is

49:05 that, you know, now it's It's very quick for me to format a chart and get it where I want. It's, I mean, it's the reason that I use Excel, frankly, is mainly for its formatting. Otherwise, I

49:15 could live exclusively in Google Sheets. Yeah, that's where Google looker is terrible, guys. Please fix that. And even the native built-in charting stuff in Google Sheets is awful. Yes. So bad.

49:28 That's where I was, is that, you know, I, all the customization that I can do in Excel on the visualization that most of that is not available to me. And Google, if it was, I would just stay

49:38 completely in Google Sheets, but, so I pulled into Excel, do those visualizations, and then for my clients, my deliverables are almost always, you know, just written reports, and, you know,

49:48 they may have a chart or a handful of charts in there, but it's not a super deep data reporting exercise, 'cause these are senior executives that, you know, don't need a whole bunch of that, but

49:56 all of that back, and as you guys know, all of that back end is necessary for me to then write andor speak with confidence about this is, these are the insights that you need to have in order to

50:07 support your capital allocation decision Yeah, about to make. Well, even though I think doing things in Google Sheets or Excel, if you need to provide that artifact, there's most people can step

50:16 into that and if they wanted to see the logic. And I think, say, on the Python side and stuff, no books are great. And I think where you can add the markdown explaining your steps and people can

50:26 see the code. But at the same time, that code is not - people know what a summit is doing. They know what Avidlokov is doing, or Air XLOOKUP, or whatever you're using. So it's kind of a lingua

50:35 franca, especially when you're talking to a lot of those executives They're either engineers or finance folks, usually. And they know one or the other. They've all had these Excel, like - That's

50:43 right. So I think there's definitely some benefit to that, or I guess not a black box for them. That's right. And that part matters. And so even though there is all this machinery in the back,

50:53 and like I said, it really is the optionality, the flexibility, the ease of sharing it, even though that happens rarely. But knowing that in a moment's notice, I can pull that stuff up and walk

51:01 people through it. That's why I end up doing it, even though there's a cost. But these scripting type tools or whatever that might - 'Cause that's really the pain for me is. The data comes in via

51:11 the API the way that it comes in, and

51:15 without getting super involved with the nitty gritty, but seemingly small decisions around the structure with which the data comes in means that, man, my data transformations that I have to do on

51:26 the back end become much more involved to make sure that I'm getting kind of an apples-to-apples compares than I expect to get, knowing all the variability that exists and how different companies

51:34 report, or like I said, just even smaller, how the data I'm getting from the IMF versus the World Bank. So there's just a lot of that that goes on in the back end. When you almost immediately

51:43 always have to to clean it up or to validate it or to do something with it anyway. And so it's always like, oh, well, just having it here is nice, but it's not remotely close to how I need it or

51:55 what I'm gonna end up, what it's gonna end up looking like, right? It's exactly right. Now, not to try to rationalize into my own pain here, but one advantage of spending that much time trying

52:07 to transform it and clean it and make sure that it's complete, is I do have much more familiarity with those data sets than I would. If I literally could just click a button and it gives me CapEx

52:15 for every EMP by year. Great, I would get the insights out of that. And yet, I am strongly convinced. I do learn things about the structure of these markets by studying this data as deeply as I

52:26 do. When you get that deep in data, I'm convinced when you

52:32 get that deep in the data, you see these patterns or these things that you would never have seen had you not done that. And then of course, you also have the whole like justifiability and like true

52:45 understanding of that data and stuff. The thing that's crazy about it. So like we're doing a bunch of data stuff with the language models and all of that. And it's like, you could like our

52:57 generation has always generally been trust but verify type situation. But you can see with the language models now it's just like, they're pushing you to just trust. And so it's like, even when

53:08 we're building out some of the workflows and the automations we're doing, it's like, well, yes, it could, the user could ask it to generate this and then it could just spit it out. But that user,

53:17 I know, will want to know exactly the steps it took to generate that and what values it used to generate that and what math it used to generate that in addition to the, the output, not just the

53:28 output. But it's like, is that where it's going to end up though? Like that's terrifying for me personally as someone who's dealing with state all the time because there's just, there's so much

53:37 noise and there's junk or there's like, you know, it misinterprets it or it's like, oh, I didn't have it structured to for UTF-8. And so it broke everything. And like there's, there are all

53:46 those nuances and stuff that kind of terrify me about like, I mean, I guess it's also, it's, it's, I feel like an old man saying it because it's like, Oh, well, that's the same thing of like a

53:56 calculator or like, you know what I mean? Like there's an analogy on both sides of it, which I was the first one to be pissed off in high school and teachers like, you can't use a calculator.

54:05 Right. That's not how the real world works. Why? Oh, my engine. Oh, God. All my undergrad engineering classes. Why do I have to memorize these physics equations? Or why? Why do I are you

54:15 forcing me to put them on a one-page sheet of paper when the tiniest, like the amount of time I spent doing that was just unbelievable. When that's not the real world's, you know, scenario, I can

54:25 look it up, right? But what you just said, it's also important to know the bounding, like the data, what actually happens, what's actually happening, where the errors are in the data, so that

54:35 when they say, well, why does the chart look like this, you can be like, oh, well, because X. But if you didn't explore it, you would have no clue. It's exactly right. And I think what you

54:43 just said, it at least struck me as being very powerful in the sense that, you know, like you said, you have these objections to using or not using calculators and now to using or not using these

54:52 language models. And the fact that that that you specifically all I certainly share a lot of that with you, and I think people in general share that shows that there is this kind of unresolvable

55:03 element where human beings do crave the clarity and perspective and confidence that interacting with other human beings or with these full data sets that once you start hiding stuff behind, you know,

55:20 an insider behind a machine wall, it makes human beings very uncomfortable. And so when you talk about, okay, what jobs are going to be available, how will people create value in a world that's

55:29 increasingly reliant on AI? That bit of it, like this human confidence building interrogation exploration bid, will never go away. And that's what human beings can offer each other. I hope so.

55:41 I'm very, and maybe I'm unduly confident about it. We're engineers. We have to know the why, right? That's right. There are lots of people who do not care about the why. They just want to know

55:51 the answers so they can move on about their day. And that's what scares me sometimes, right? But then there's just a lot of almost mental atrophy. Yeah, no, absolutely. That's a huge worry of

55:60 all of this for me personally at least, is it's like. I like SQL, right? Like I never truly learned SQL. I learned a couple of things, but it's like now I don't necessarily need to learn SQL

56:11 'cause I can get an LLM to do it for me. Or we have been testing this LLM agent that sits on top of your SQL so that my marketing team who definitely doesn't know SQL can just ask it questions and it

56:23 writes the SQL and gives them the answer. And it's like, but even then testing it, right? Like fortunately it gives you the original, or the SQL query that it writes so you can test it outside of

56:34 the platform. And it's like, oh, well, gave me one answer here. I took the exact same SQL call, put it into my database, and ran it, and it's a different answer. I was like, hey, that's not

56:45 good. I do not trust you now. And so it's like there's having to, like you have to go through that in order to build the trust in these things. But I'm also worried that it's like, we're gonna

56:54 get to that point where you trust it so much that it's just like you quit closing that feedback loop every time. Yeah, it's a scariest place to be. And to call back to our earlier part of the

57:04 conversation about if we live in an industry and work in an industry where the consequences of failure catastrophic, right? So if anyone should care about dotting eyes, crossing Ts, being that

57:14 specific about us, it's us in this space. And I think it is one of the reasons I'm as optimistic as I am about, yes, the future of work, there will be a lot of evolution. But I think there is,

57:25 you know, again, this sort of unresolvable curiosity that, I'm with you that not everyone is wired in a way where they're super curious people. So I don't want to come across as that naive, no,

57:36 we're all curious. But in a world where AI is increasingly dominant, those people that are themselves most curious and are willing to diverge from the path of least resistance, I think that the

57:49 work that they choose to do will be more valuable. I think the connections they form with other people that are wired similarly will be more enriching for them And then for folks that don't give a

58:00 care about. any of that and are willing to follow the path of least resistance and finally that's available to them. But I think there is this very rich future that's possible and maybe it is

58:09 possible for a smaller subset of folks, but it is those folks that are curious and interested enough in the rigor of answering these questions thoroughly. I think that will continue to exist. Um,

58:19 do you want to jump to the, the Q and A and the like speed around? Yeah. So Jacob, I was on, you know, they started up energy 101 again and he has taken a lot of cues from other podcasts and

58:32 stuff. I think we kind of got that from them too. Really. Yeah, we did actually. It's kind of funny. I forgot about that. Um, one of the things I was going to ask you is like, would you, so

58:41 he, one of the things he was doing was, would you rather questions with? I thought was another kind of nice compliment. Would you rather see plus plus or for Tran? For Tran, I love it strongly.

58:52 Yeah. It's just, it's, it's what I'm most familiar with it, um, for the kind of work that I do that is so computationally heavy, it's really stripped down and so I feel like I can quickly get to

59:02 what I wanted to get to there. I learned C enough to be somewhat dangerous with it, but I never got to the comfort level that I have with Fortran. What's your favorite restaurant in Houston? Yeah,

59:13 that's a good one. Good one. Man, favorite. Reframe that. Where would you tell someone visiting Houston to go eat? That's another. That's another good one. Way down. Wait up.

59:25 Zochi is one of the ones that I really enjoy there downtown. It's a lot of fun It's like the upscale Mexican or - Yeah, Mexican one. Yeah. What am I thinking? Oochi. Oochi. Yeah. And I just,

59:36 you know, it's downtown. It's right next to, you know, it's close to Discovery Green. And so just the part of town is fun, but then the ambiance and the restaurant is fun. It's, you know, the

59:46 nice music, great food, great drinks. Easy to have a good time in there. Yeah, it was, he was a beard nominator, or winter, right? I think so. Yeah. Yeah. Yeah. Gotta love that you're,

59:58 your city has multiple James Beardwinners and multiple restaurants. It's from that one of those people. It's Michelin stars here. Yeah, talk shit about Houston and how bad it is. It's like, yeah,

1:00:06 we're the best food country in the, or food city in the country still, so. That's right. What is your favorite scripting language, is it? We can include Power Query or any of the VBA. VBA.

1:00:20 Okay, so this might get really embarrassing for me, but when you ask that question, I immediately go to bash scripts. That's not nice, that's in Linux Yeah, yeah. When I was in grad school. Or

1:00:29 Mac, D's Mac,

1:00:32 yes, but for my day-to-day, I used Windows back when I was in grad school. I started in Linux, and then our lab kind of transitioned over to Mac, and so it was always Unix-based stuff. But when

1:00:43 we were a full Linux shop, I was the system administrator for both for our machines, but then also, 'cause I was doing computational fluid dynamics work, we had a Beowulf cluster. And so it was

1:00:53 in the basement of an adjacent building And so occasionally I'd have to go over, walk over there to reset machines, or if we were to add nodes to the cluster or whatever. So I would manage our

1:01:02 cluster. And so I would also do all the firewall, the IT security stuff. 'Cause occasionally, there at Caltech, some of the undergrads are trying to goof off and hack into things. And so I have

1:01:12 to try to make sure that we're protected from all that stuff. But on our machines in the lab, I had just real simple bash scripts that handled a bunch of routine tasks. So I know it's not really in

1:01:23 the direction you think of when you ask the question. No, I personally love that Like I use command line over anything else when I'm doing scripting, because it is the most like efficient way or

1:01:32 just fastest way to do it in my mind. And I prefer bash over command prompt or - Yes. PowerShell, so. Yeah, yeah. Yeah, yeah. PowerShell fucking, yeah. You know, I graduated college in May

1:01:43 of 2003. And so my first two or three years, I started in fall of '99, finished out three. First two or three years, I

1:01:50 had, you know, just a standard Microsoft desktop. But for some reason, I forgot why, but I kind of got on my high horse and decided, you know, I'm gonna wipe my machine clean. I'm gonna go

1:01:59 straight into Linux and learn how this thing worked. And so I was one of those Linux enthusiasts where, you know, every two to three months, I'd wipe the machine clean and put on a different

1:02:08 flavor of Linux. And so I could experiment with all of them. And then, you know, as a result, I just got really used to the, you know, living on the Linux command line, yeah. So everything

1:02:18 was command line based. And then when our lab went over to Macs, it was largely for the, you know, other software tools that Macs had to make the presentations look nice and, you know, some of

1:02:27 the animation stuff and all that. But it was nice 'cause I had that Unix backend that I could still live in. And so I transitioned to back just 'cause, you know, industry is the way it is.

1:02:35 There's a lot of Microsoft stuff and I use Excel. And so, but there was many years, probably close to a decade, maybe a shade over a

1:02:42 decade where almost exclusively Linux, a little bit of Mac, but very little windows. Yeah. That's, yeah All right.

1:02:51 I installed Linux one time on the laptop when I was in like the high school or college. I don't even remember and I was like, oh, this is scary. It's intense. But when you get into like Python or

1:03:01 really any coding language, everything works better on Unix based. Oh, yeah. The next, you know, whatever. But I'm still not sure that my gaming laptop, my Windows gaming laptop was ultimately

1:03:10 the right decision for me last year. But

1:03:14 as long as I stay in the command prompt, I'm okay going into WSL as I avoid that like the playing Because Linux was intimidating, and yet even 20-ish years ago, you know, I burned it onto a CD and

1:03:29 used it to install it that way. And I was surprised how adept it was at pulling in a range of different harbors. I would build all my own computers. I was ordering motherboards and, you know, my

1:03:40 own RAM and video cards and all that. Tiger Direct. Yeah, that kind of stuff. Yeah, that's exactly right. That's exactly right. And so I was impressed with how it could recognize that It was

1:03:49 well known if you were in the Linux. It was well known, there was certain classes of hardware, it's like, don't play around with that, yeah, it's not worth it. But all these different flavors

1:03:56 of Linux and how it almost got to the point I remember 'cause my grandparents had just a little machine they were using just for email and like the simplest form of web searching and web browsing.

1:04:08 And I remember my grandfather just getting so frustrated 'cause that thing would just get eaten up by viruses and like I'd had to come repair it that I never ended up doing but I remember being 16,

1:04:17 17 years old, whatever it was And man, I should just wipe everything clean, drop Linux on 'cause you only need to click on these two things in order to do what you're - Right, I just went to the

1:04:25 browser. Exactly. And I don't have to worry about all this virus nonsense on the back. Yeah, before the Chromebook came along.

1:04:34 It was just - Yeah, that's what it's like. Exactly right. It's exactly right. What's your go-to kind of vacation spot? The one we've been to the most often is Cancun. It's just so easy to get to.

1:04:45 It's so easy Yep, it's, and I'm someone that, Two or three days down there is perfect. Once it gets a little longer than that, I get a little. One, it's an easy trip for two or three days too,

1:04:56 right? You're not waiting in line, you're not traveling for half of a day. That's right, you know, the quote that I heard when we went down there with some friends one time was, you know, you

1:05:04 can be toes and sand before lunch. Leave that morning toes and sand before lunch, which is fantastic. And, you know, like you guys, right? I'm wired where I am kind of just always on. So

1:05:14 having two or three days to shut down a little, drink, relax, swim, hang out. My son's made fun. I mean, we took a family trip down there with some neighborhood families. And the resort we

1:05:26 were staying at, they had an inflatable obstacle course just off the beach into the water a little bit. And so they were having people race on this obstacle course. I went out there and did this

1:05:35 race and it just damn near killed me. So it's in the water trying to swim and jump and climb. And you know, I've seen so many videos of people doing this, Well, it's like, it looks like it would

1:05:48 be like fun and easy, like it was on land, except everything is wet, and then the minute you fall, you're in the water, and you're more wet, and now it's just this infinite cycle. That's right,

1:05:57 that's right, my kids were, one of our friends was recording me doing this, so I come back, they want to show me this video, and I can hear my oldest son at Oneport, 'cause I just fall off the

1:06:09 day like I go into this little pool that's inside the middle of it, he's like, Oh no, is daddy okay?

1:06:15 That's how bad I looked out there doing this, is that when I take a dive, my son's like, Oh no, is this gonna end badly for Dan? That's hilarious. Y'all ever go to Isla? Yeah, yeah, yeah, a

1:06:27 bit a couple of times, love that. Yeah, that place is cool. Yes. There's only golf carts on the island, and you can literally drive the island on a golf cart in like 30 minutes maybe, but we

1:06:39 went there one time, and they have one like grocery, It's like a Walmart or like a Walmart Superstore equivalent kind of. It was like, it was just like the fruits, vegetables, and then right

1:06:52 into like mattresses and appliances on the same aisle, you know? It's like, you know, like the traditional American grocery store. It's all the cold stuff on the outside. So we're just walking

1:07:02 and I was like, where did we just take a left her? No, right into mattresses next to like fruit. But, easel is a great spot. You got one more? No, I think we're good, man. Yeah, but that

1:07:13 was been great. Thanks so much Now, I'm glad you could make it work. Yeah, I had an absolute blast. I knew I would and it was even more fun than I expected. Not much. So thanks for hosting.

1:07:22 Absolutely. Thanks for coming on. Where do people get in touch? How do they find you? The easiest way is on LinkedIn. I'm at Jeff Kremel. It's like Jimmy Kimmel, but with an R. And then you

1:07:32 can also find the Kremel SG website, kremel SGcom. And then you can sign up for my free newsletter there. I set up a free newsletter so that, you know, people follow me on LinkedIn. I share

1:07:41 everything on LinkedIn. But then I have a newsletter that goes out each weekend. It has links to and summaries of all the research that I post on LinkedIn, so people don't have to find you, dig

1:07:49 through and find everything there. And then it has links to the sub stack and everything else that I'm working on. You've got a sub stack too, right? That's right. Yeah. So it has links to all

1:07:55 that stuff. So if you find me on LinkedIn, I'm certainly sharing all that kind of stuff there. And if you go to the criminal SGcom, then you can sign up for newsletter stuff there. You can also

1:08:04 find him on Collide. He's very active on Collide. Yeah. Collide's a lot of fun In fact, I texted you a while back that, you know, I joined Twitter circa 2009, I think, and I never found my

1:08:17 voice over at Twitter. I think, you know, for kind of the reasons you guys know from even chatting here, this real short-form text communication is not mine. And I'm not good at the hot takes,

1:08:27 like I'm much better at kind of building up to what I find to be sort of an insightful argument and sharing that. And so from day one at Collide, I found not only more engagement in terms of from

1:08:38 the vanity metrics side of deal, but it seemed to be more the kind of audience of people that um, comfort in, which is people that are thinking deeply about things, people that have a pretty wide

1:08:50 variety of experiences. These are typically not folks that are looking for an echo chamber environment where, you know, there's, you know, calls and call back. Like it, collide does not feel to

1:09:01 me like I'm at a political rally or like I'm sitting, you know, watching a cable news talk show kind of like, I don't get any of that environment off of it. Um, and yet it's people that care very

1:09:09 deeply about the energy space writ large. And so it's not like, you know, people are just avoiding talking about, you know, topics of consequence. And so, uh, yeah, the, the, the collide

1:09:18 bit is nice for me because it's, it seems to have its own community. That's not really replicated in a, um, in a sense elsewhere. And then compared to my experience on Twitter, it's like, man,

1:09:29 collage is so much more enjoyable to engage in. That's awesome to hear. I appreciate that. Yeah. Well, thanks for building it. Thanks for maintaining. I, you know, it's one of those deals,

1:09:37 you know, even me being as involved in social media as, as I am largely through LinkedIn, but then, you know, having my

1:09:45 I do have a sense of how much work goes on on the back end of that. But even that, I know that I don't fully appreciate what's going on there. And I also know it's one of those deals where it's got

1:09:55 to and maybe I'm putting more of it in your mouth unfairly, but it's gotta feel like sometimes, this is all coming together exactly the way we want. And then other times it's like, man, like this

1:10:03 is, it's gonna require more of a grind. It's gonna require more effort. It's gonna require more commitment to see this thing out to where we're trying to, in different parts of the platform,

1:10:11 probably give different reactions at different points But I'm always very impressed by folks that choose to try to build a community in that way, 'cause it's not one of those deals where you turn the

1:10:20 light switch on and which shows up. And you're gonna go. So now we're gonna go. You gotta build it. Yeah, no, it is a grind at times, but it's awesome to see. Like we just, I think a couple

1:10:30 weeks ago, across 7, 000 users. Wow. And so, and we're over, I think we set it all time high for monthly active at like 14 or 1500, I think. And so seeing the fruits of your labor, even

1:10:41 though it takes years to get there, pretty awesome and fulfilling. And then, yeah, having you say, people like you say such nice words, you're like, I've had people be like, oh, like meeting

1:10:54 someone for the first time in person that they met on collide, right? And like, that's really cool to kind of see in real life and stuff. But I'm sure I'll go back to my desk and there will be 20

1:11:04 requests and bug issues. Yeah, all of the typical stuff. So it's empowering when we get positive feedback like that to slug through all Well, and that's why I want to share, 'cause I know that

1:11:15 there's a tedious bit to it, but like you said, you guys are all mission driven and you know what this mission looks like, what the other side of it is. But you also, like you said, this is a

1:11:24 year's long engagement and there is no finish line. You're never gonna get to a point where it's like, oh, it turns out we've knocked this thing out of the park and we can move on to like, no, no,

1:11:32 it's a continuous improvement. Yes. Or deal. If you don't, if you don't like conclusion, go into development. Um, if you do, if you need conclusion, do not go into software development. It's

1:11:44 just hard and fast. That's, that's my pro tip for the day. Uh, it's, it's never over. It really isn't. So awesome. Well, cool man. So we did that. But again, thank you so much. And, uh,

1:11:55 stay in touch. Yeah. Thanks for having me. Well, we'll definitely have you back on again. Appreciate you guys. Thanks for, uh, for tuning in.