| 00:00 |
I was walking with someone and they said, this is kind of like a graveyard. Like people died here. And so that impacted me quite a bit because I’d never been in oil and gas. And this was 20 years ago, so I’d never been in oil and gas. And it kind of brought home and it kind of drove home how important what we were doing was. |
| 00:30 |
And it’s easy after a while to dismiss it and get complacent, which is why I’m glad we’re doing this because it kind of and stories like this and anecdotal stories that we have and things that we share with hopefully that impacts people that are listening and watching. I hear you. Safety Margin, a Cognascents [music] podcast on process safety, asset integrity, and AI. |
| 00:57 |
[music] Lately, one of the conversations I’ve been having with a lot of folks is, are things worse today in the industry, whether they’re under the jurisdiction of process safety or EPA’s RMP? |
| 01:13 |
And are we seeing more incidents that aren’t just killing people but are near misses that are significant enough that we were maybe one layer of protection away from a fatality or multiple fatalities. And one of the incidents that happened recently it was January 27th to be exact 2026 up in Maine. |
| 01:40 |
It was a wood pulp facility and this one struck a little close to home because I was a chemical engineering co-op at one point and I was also a relatively young chemical engineer in an industry and you had two folks working on the second floor of this mill and unbeknownst to them during a shutdown process you had the mixing of chemicals that created H2S and they succumbed. |
| 02:10 |
And later on, once they realized they had the excursion and they took, you know, emergency response measures, they didn’t find these two individuals till close to 3 hours after they had already fallen. Oh yeah. And so the 20-year-old, I believe, died a day later, but then the chemical engineering employee died several days later when he was taken off life support. |
| 02:40 |
It’s tough. And when you look at those incidents, what comes to mind is the very name of our podcast, Safety Margin. And it’s that it’s that space between everything being good to that really bad day when folks aren’t going home the way they went to work. And I feel like it’s part of our job and part of our why to raise the clarion call right now and put our industry and put ourselves under the lens of scrutiny and say, “Hey, are we protecting our safety margins?” |
| 03:13 |
Definitely. So, welcome to safety margin. Here we are. All right. So, who are we? I’m Justin Daarud. I’m asset integrity, reliability, and play a little bit with artificial intelligence director at Cognascents. Happy to be here. I’m a mechanical engineer, which is fun because I’m always surrounded by a lot of chemical engineers in this company. |
| 03:36 |
Had to fix the TV stand the other day. We were setting it up, right, for a conference, and Jennifer, who we work with, she’s like, “It’s broken. It’s janky.” I was like, “No.” And Jennifer is a mechanical engineer. She’s a chemical engineer. Yeah. So, she’s like, “We got to buy a new one. We just can’t go to another conference with this.” And she rolled it over to me. I was like, “How? Who set it up?” Oh, well, John, me, Casey, Norberto, we were all at the conference. John, not Alamo, John. CCPS conference. You guys were all there. Remember the TV was kind of like janky? |
| 04:09 |
[laughter] Awkward. Okay. But what to be fair, right? One of the biggest differences I often tell folks is mechanical engineers have the luxury of touching the things that they engineer and protect and serve. Sure. We play in the abstract world of I don’t get to sit inside a column and see the separation occur. |
| 04:31 |
Right. So we’re very abstract when we start looking at TV stands. It’s very difficult for you guys. Sometimes I’m looking at [laughter] it and going, “Why isn’t it fixing itself?” |
| 04:40 |
Right. Yeah. Well, I fixed it. Yeah. It took and we don’t have to buy a new one. Yeah. What about you, sir? King, sir. I am King. I’ve been with a Cognascents for 10 years. Tomorrow. As of tomorrow. Congratulations. Where are we going for lunch? Wherever you want to go. Laas. Well, not wherever. |
| 04:59 |
Laahas. Laas. Laa. I am an API 510 pressure vessel inspector. That’s another congratulations. Thank you. It was a long journey, but I got there. Yeah. Some of the best are sometimes long. And I’ve got quite a bit of field experience. Yeah. And I was offshore for five years, which I think in that was one of the most interesting facets of my career. |
| 05:29 |
And I miss it. Well, we’ll try to we’ll try to get you back offshore, but we have enjoyed you being back in the office quite a bit lately, though. Well, you say that now. I was gonna ask before this started if you wanted to go back offshore. Well, there you go. There’s my answer. |
| 05:46 |
Don’t forget, we should ask his wife and kids as well. Oh, good. Yeah, obviously. And then to wrap it up, I’m John Perez, founder of Cognascents. Started this thing about 16 years ago. It’s my second startup. Chemical engineer by trade. Went to college as an English major. So, you know, took a bit of a detour there. |
| 06:02 |
But I’ve been in process safety for as long as I’ve been out of school. So, looking forward to being part of this crew. I know that Justin and I are probably going to be permanent fixtures. We’re still beta testing king here. You know, he might be in and out. His function really is to keep Justin and I honest so we can and interesting. |
| 06:26 |
Yeah. And interesting. [laughter] So, and our roles are really to make sure that King stays within the lines here. So, but let me ask you guys, why do you think we’re doing this podcast? Why do I think? I think [clears throat] it’s to create awareness in the industry, have a platform to try to have good conversations about what’s going on. |
| 06:50 |
I like the idea of our name, Safety Margin. That’s pretty cool. Good way to pull pull the idea together. I don’t think it’s for us to, you know, like sell stuff. I think it’s for us to get word out great information, right? We need people in our industry to think about what it means to have integrity, be safe at their jobs, and try to strive to continue to solve that problem. |
| 07:12 |
That’s what I think. Yeah, you do. What about you, King? I agree with Justin. [laughter] That can’t be how you handle the every question, man. [laughter] No, I like the idea of all of us getting in here and talking, in a more… what’s the word? Intimate setting. Yeah, but we’re going to post it on the internet where the but it’s still intimate between us. Good point. No, I agree. Yeah, so when we were trying to figure out the name, we actually went through a good number. Some of them we ran into copyright issues with, so we couldn’t go with like, for example, Risky Business. |
| 08:02 |
But [laughter] I still that’s still my favorite. But Tom Cruise might have issue with that. Yeah. Or the studio. But anyway, we landed on safety margin. I agree. I think it’s a good place to land. What I often tend to go back to when I think about why we’re doing this and actually why we do what we do. |
| 08:19 |
And you’re right. This isn’t a lead gen effort here. As we try to make all of our presentations at conferences not necessarily a sales pitch like other folks do. We really want to add to the body of knowledge here to the conversation. But for me I think about every day when I get in my car I am actually engaging in a social contract with everyone including King that they’re not on the road trying to kill me right and I’m not sure everyone gives pause to that moment they get behind the wheel. |
| 08:53 |
I know some of them use Teslas now and FSD, but you know, it’s awesome. Justin has a point, but you’re still getting on the road engaging in a social contract that other folks are working towards this promotion of societal safety, right? Mhm. And even we have that general duty clause in industry from OSHA where even companies have this service to their employees to if they know of something that is unsafe, they should correct it. |
| 09:21 |
We tend to enforce that though after the fact after something’s gone wrong, which I do find interesting. But either way, I feel like you’re right. I can’t wait for someone at Dow or Exxon Mobil to get through the corporate board, you know, red tape and the legalese and the legal counsel’s machinations to be able to do a podcast to be the captains of industry that have this conversation, right? |
| 09:52 |
So, interestingly enough, it’s left to the little guys like us to actually serve as, you know, that clarion. But I feel like it’s part of our noble duty, part of serving it forward, and just part of being captains of industry in our own way. So, I’m happy to do it. I know that there’s a bit of risk in doing it. |
| 10:14 |
And so, the disclaimer that I put out to everyone listening to this inaugural episode is seek legal professional help before following anything that we say, right? We’re pretty good, but you’re not paying us to tell you or give you any advice right now. So, don’t take it as, “Hey, this is the Bible. Go do it.” So, that’s our legal disclaimer and hopefully our legal counsel is happy enough with that. Yep. We should probably keep that part in the edited episode. Exactly. I agree. [laughter] All right. |
| 10:43 |
So, before we actually get into it, I know you probably didn’t read the script. I read the entire script. It’s been tweaked probably six times since last night. Good thing I didn’t read it. Yeah. But it kind of I wasted two hours. It follows the same flow, right? We kind of get into the why, who we are, why, but then we also get into u what’s going on out there in the industry, incidents, accidents, but then we talk about AI, and then at the end, we get into a conversation about asset integrity with a segue using armor and blacksmith and was he the inventor of the first integrity program? |
| 11:22 |
Interesting. Yeah. So, I didn’t realize that people oiled armor as much as they did. Oh, you have to Yeah. To keep it. So, is a single-piece breastplate better than chain mail? Depends on what you’re trying to protect. Right. Right. So, flexibility chain mail works. Yeah. Flexibility that maybe not against arrows, but if you’re getting into an armor fight, you’ve got a whole bunch of stuff going on, you probably want to have glancing blows protected by chain mail, right? |
| 11:56 |
But when it comes to corrosion mechanisms, surface area and yeah, you got to keep ring bits that are woven together as or maybe more vulnerable to certain things than just a single plate. Yeah, you’re going to have contact point corrosion. How do you know so much about contact point corrosion? How do you know total dirty TV boy? |
| 12:22 |
[laughter] How do both of you know so much about armor? We drink and we know things. Yeah, that nerds said that Peter Dinklage at Game of Thrones. Yes. So, see it’s the nerd part of us. So the next segment and we kind of introduced it in our first segment is are the numbers and are we seeing an uptick in incidents and so what’s your take on that? |
| 12:42 |
Well I think we have to think about our perspective. We are in the industry so we’re always kind of looking and thinking about process safety incidents how to prevent them and all that good stuff. But I think there is a proliferation of information, especially with all of the social media, the quick shorts that you might see on LinkedIn and whatnot. |
| 13:06 |
People trying to capture attention on what’s going on, whether for good or bad reasons. So, I think we are seeing maybe more publication of these types of incidents. I hope that they’re going down, but it seems like it’s not, you know, like it’s it’s more readily available. It’s easier to see on your streams and feeds. |
| 13:23 |
There’s YouTube videos on how all of these things go crazy. The CCPS kind of group and organization does a good job of covering that kind of stuff as well. So you can always find all that information everywhere. The stuff that I really like is when the incident investigations come out and they give the root cause assessments. |
| 13:49 |
I like thinking if that’s the correct way that they actually identified the right root cause. For instance, when the Philadelphia refinery incident came out, that one was really interesting because it was very much attributed to asset integrity and through wall thinning of a damage mechanism on a piece of pipe which failed, jet fire issues, refinery blew up. |
| 14:15 |
When did that happen? Oh, now you’re challenging my date in memory, but I think that was around 2019. Okay. Somewhere around there. Few years ago. Which one? That was the is that PBF or that was the Philadelphia refining. Yeah. Okay. So, they ended up [clears throat] heating I forget which unit it was. |
| 14:37 |
21st, 2019. There you go. 2019. Solid work. That was the day before my anniversary. Happy anniversary. Yeah, good stuff. But yeah, that kind of stuff really interesting. That incident at the Philadelphia refining facility: they were doing risk-based inspection on their piping circuit. They had a failure in between condition monitoring locations, CMLs, and they always took CMLs at a reg like a regular basis. |
| 15:01 |
They had a lot of information on the thinning mechanisms throughout that piping circuit, but they missed this elbow. And because they missed the elbow at a localized thin spot, ate through the wall enough to cause a loss of pressure containment. And then that eventually that hydrocarbon found the fire and ignition point and actually ended up being a really bad thing. |
| 15:26 |
But they were doing the right thing. Risk based inspection. They were trying to manage their risk effectively by being proactive about what damage mechanisms were there, but they missed it. So, to me, one thought was: it’s piping. Piping is a big issue across the industry because the majority of our failures for mechanical integrity incidents happen on piping. |
| 15:50 |
So applying a good principle like RBI which is essentially engineering-based decision-making on piping, wasn’t working for them. So how do we make programs like that actually work and address these kinds of issues without skipping over it? Because these CMLs were pretty close to each other, but they just completely missed that thin spot. |
| 16:09 |
So, yes, incidents seems like they’re getting published and pushed out quite a bit. I hope the numbers go down, but I think what we’ll see, especially with the increase in like data centers and energy needs and infrastructure, there’s going to be a lot more fast energy plants that are going to come up, maybe gas plants that are generating a lot of power. |
| 16:30 |
They may cut some corners or try to cut some corners because they’re trying to build it as quickly as they possibly can so they can supply a lot more power to the grid. And we’re going to have to ensure that we can go out and try to help them or at least the industry, not just us, but the industry is going to have to do good quality control, good management of their maintenance and integrity and reliability programs, good process safety, all that good stuff. |
| 16:53 |
But I think we might see more incidents just because more infrastructure is being maybe brought back to the states and/or being constructed to keep up with demand. I don’t know. That’s what I’m thinking. King, what’s your take? I know that maybe you don’t follow it to the same extent as Justin and myself. |
| 17:10 |
But what’s your take based on, you know, just your gut feel perhaps? Are we seeing more incidents? Are we seeing kind of slippage in the industry as far as process safety oversight goes? I will hearken back to when I first got into oil and gas. [laughter] Huzzah. |
| 17:29 |
That’s a word. When I first got into oil and gas, I was at Can I say the refinery? Yeah. I wouldn’t mention any client names or anything like that, but I was at specific refinery. They had just had an explosion and it was probably 3 years after the fact and that same unit hadn’t been touched for legal reasons. |
| 17:50 |
But I walked by it a lot and I saw the aftermath and I saw the mangled piping. It was desolate. But I was walking with someone and they said this is kind of like a graveyard. Like people died here and so that impacted me quite a bit because I’d never been in oil and gas and this was 20 years ago so I’d never been in oil and gas. |
| 18:26 |
And it kind of brought home and it kind of drove home how important what we were doing was. And it’s easy after a while to dismiss it and get complacent, which is why I’m glad we’re doing this because it kind of and stories like this and anecdotal stories that we have and things that we share with [snorts] things that we share and things that we hear hopefully that impacts people that are listening and watching. |
| 19:01 |
I hear you. Yeah, I think a lot of us, perhaps even Justin and my wife, we’ve been in the industry for a while. We’ve all been either directly involved or certainly have known someone, involved with, either a fatality event, near miss event, something significant. But one thing I wanted to do versus just saying, “Hey, we think I did take a look at some numbers.” |
| 19:23 |
And when you start looking at, and just so y’all know, there aren’t tons of publicly available databases for us to go look at and say, “Hey, are we actually seeing more incidents?” EPA has a database and there’s a group that’s called peer and basically it’s a public group that does evaluation of engineering research engineering analysis and you can actually look at some of their data and when you look at the EPA side it looks like there’s a decline across 2014 to 2023 but then when you look at the shorter term set of years like say 2023 over to 2025, it’s a significant increase. |
| 20:02 |
In fact, from 2024 to 2025, you’re seeing a 22 to 23% increase in incidents. And so, I think depending, it kind of goes back to what you said, Justin, depending on what set of data you frame your lens to, it’s going to give you perhaps maybe the response you were looking for. |
| 20:26 |
Totally. You know, one thing I always worry about and you know, I gave a paper back in April at the CCPS conference about it or GCPS conference about it is logical fallacies and cognitive biases in process safety. And I do wonder if maybe we’re not necessarily seeing more incidents but maybe we have definitely seen the I guess effect that maybe we’ve reduced the frequency but we haven’t necessarily reduced what happens when an incident occurs right like they’re still just as bad. |
| 20:59 |
And sometimes when you look at how we approach PHAs, HAZOPs, and LOPAs, we always focus on when we identify a gap, we focus on how can we reduce the frequency, right? Because it’s harder to sit there and go, well, how am I actually going to hit the consequence side of this, you know, risk equation. |
| 21:21 |
So we always go to the frequency side. And sometimes I do feel that I’m not saying we fudge the numbers or we manipulate the numbers, but we’re engineers and we work with numbers. And I think we start developing systems that drive us to build in perhaps some cognitive biases driven by what we think is logical number checking, right? |
| 21:40 |
Or number exercises. I do worry about that. Yeah, that’s interesting. What were the dates on the first one? You said 2014. So 2014 to 2023 on the EPA side was showing a decline. But then when you look at 2021 to 2025 for the CSB reports you’re seeing a 57% rise across that 2021 to 2025. |
| 22:08 |
Now again to be fair I believe there was a ruling back in the early 20s that drove more transparency from certain government agencies. So you might be seeing an uptick simply because they have to be more transparent and reporting has proliferated. Yeah, for sure. I wonder if those numbers can be tied to the amount of manufacturing was that was happening in states at the time. |
| 22:28 |
So like 2014, you know, refineries, chemical plants, producers that make widgets, all that stuff. Have we seen a decline in you know North American manufacturing over that time which may reduce the amount of people involved in incidents and all that stuff or incidents actually occurring and then in the past three years was that 23 have we seen an increase in people in manufacturing is it correlated right because yeah I think a couple things and again I don’t want to get into the paper too much but you hit on one of the things that I touched on but another thing is what happened in 2020 2021 that may have driven a lot of our memory away from the industry. |
| 23:12 |
I can barely remember. Yeah. Right. The fog of COVID, right? I had to grow a mustache. I was tired of the face coverings. [laughter] But I mean, we saw a that’s when you saw an uptick in the boomer generation saying, you know what? |
| 23:28 |
I’m finally done, right? I’m I’m going to retire and this is good time to kind of hit that retirement button and, you know, pull the parachute cord, right? And you got younger people getting out into hazardous environments and potentially don’t have all that experience and they do the wrong thing and take the wrong step or you know somebody neglects fixing a handrail, right? |
| 23:50 |
Things happen. Yeah. So not to be too abrupt in a segue here, but I think that’s actually a good point to probably transition into segment three. Is that okay? Intern Alex, we still looking okay? Sounding okay? All right. He gave us a thumbs up, folks. One of the John and I were talking earlier about what does the advent of AI mean for process safety? |
| 24:11 |
And one of the pros I see is I feel that we’ve lost a lot of memory. I feel that we’ve lost a lot of expertise. Process safety became something that wasn’t always sexy to chemical engineers. And so you saw a lot of chemical engineering expertise migrate away from that domain. One thing I do think AI can bring is perhaps a hardening in some places, but certainly under human oversight. |
| 24:34 |
But I do feel there are two camps when it comes to AI in our space. Yeah. Adopters and it’s the maybe not the Antichrist, but it’s the worst thing. Yeah. It’s like it’s [laughter] the devil, right? |
| 24:48 |
It’s like AI can’t do it. Yeah. It’s dumb. It’s wrong. So let’s have a conversation about what’s been your experience, maybe how are we seeing it? What do you how do you see it playing out over the next six to 18 months? Yeah, for sure. So, and maybe also explain what you’ve been up to, right? |
| 25:12 |
Yeah. So, I’ve been a early adopter of artificial intelligence for a while. Even when the chat GPT craze kind of happened basically two years, three years ago, right? But I’ve been dabbling in programming my own machine learning algorithms and those kinds of things. COVID really inspired me to get into that quite a bit more. |
| 25:32 |
So learning a lot of Python code and trying to apply math and all sorts of cool stuff. But in the recent months, I’d say year, really started trying to figure out how is this AI thing gonna actually work better than just helping me rephrase an email as an example. Or I really want to have this paragraph, but I want to cut it down by half. |
| 25:57 |
Have AI generate new paragraphs. So that’s your chat bots of your large language models that are kind of doing that. But over the last what at least eight months we’ve really gotten to the point where we’ve built and excelled outside of a chatbot and have built our own internal engineers on our computers. |
| 26:17 |
So a lot of people out there especially in the engineering world don’t think AI can do engineering stuff. I think they know that it can do from a LLM perspective, it can answer some questions, maybe not with great accuracy and it doesn’t do all the research that an engineering workflow that would do the research. |
| 26:35 |
So, I just want to jump in here real quick. Right. So, just because some folks listening might be going, “Ah, you’re wrong.” Yeah. Right. Sure. For context though, when you say chatbot, you’re talking about either going to chat GPT or Claude or Grok, going to that prompt box, typing in something, getting an answer back based on that model, going to the big wide world and all the context, you know, all the brain rot, all the good stuff, all the bad stuff. |
| 27:10 |
It’s trying to sift through all of that. Correct. Yeah. Sometimes it comes back and it’s not necessarily correct. Right. Yeah. And I’ve given it the given the example. So how the LLM works, it’s probabilistic model based on linear algebra, right? So you have this 3D space and you’re trying to get a vector through this 3D space and this 3D space has a whole bunch of dots, right? |
| 27:31 |
It’s all over the place and it’s got a it kind of has a coagulation of dots into this mass. So that’s your training set of data. Big words, right? Intern Alex, can you Google these? Yeah, coagulate. While he’s talking. Yeah, we’re going to have to provide [laughter] like a That was a joke. |
| 27:46 |
Intern Alex, don’t need you just Google it. Put your phone down. Put your phone down. So, anyway, are you keeping up? Keep it up. All right. So, you got all these dots in this big mass and it’s a 3D space, right? So, three different axes. So what the LLM is trying to do based on all the data and the model that was built for it is predict your next word of your sentence. |
| 28:08 |
Essentially give you an answer that matches the context of your question. Okay. So if I if I say Justin has a wicked and then leave a space and blank and put that into the chat GPT or the you know Claude, etc. The world you hope it would say mustache. They would say, “Well, based on what I know about my training set and the next highest probable word, I’m going to fill that in to say mustache, but it could be Justin has a wicked headache because he’s sitting in a room podcasting with lights on him.” |
| 28:38 |
And if you ask it again 10 minutes later, it could be a different answer. Correct. Especially with different context. So how do you get beyond that? Right? Because I think what’s occurred is some people have stopped there. Yep. You have to build your own way or your own methodology to take the power of a chatbot to then march it and walk it through a workflow or create your own model or you know tools etc. So there’s lots of different tools retrieval augmented rag models generation. |
| 29:07 |
Those rag models are like taking a certain set of engineering information, taking a workflow and using a chatbot to essentially have it use that context and that specific keywords to retrieve information that you’re really looking for. So, if you’re looking up like a specific equation, you could be like, “Hey, what’s the equation on hoop stress?” |
| 29:33 |
And because you have that training set and you have that hoop stress equation in there, it’ll go in and find it exactly where it is, give you a reference and give you the information. So that’s like a rag model. But just to be clear, right? So this is not just prompt engineering where you’re prompt engineering a Python script or an artifact. |
| 29:53 |
This is perhaps a combination of prompt engineering building agentified tools using rag methodology. Yeah. But I think what I’m also hearing is the context is very specific. Context is very specific, very important. So like in the engineering world, we don’t have a lot of tools that are AI tools or built for engineering because it’s you have to have domain expertise in the field that you’re trying to solve a problem with when it comes to AI and you have to have the combination of AI expertise on top of that. |
| 30:26 |
So the way you can take a AI powerful AI LLM or even a local one or a smaller one and make it even more powerful and get the results you really need is to break it into think of the LLM as an engineer that lives inside your laptop and it’s reading work instructions. |
| 30:46 |
It’s reading equations how to do things and it’s got a crew of 20 other engineers that work with them. And you’re saying, “Hey, I want to go and design and build this object.” And the object at the end of the day is a deliverable of some sort. So it could be like an engineering file of some sort. |
| 31:07 |
And so this head engineer is going to sit there and go, “You know what, King? You’re my first engineer. I’m going to ask you to go out and figure out what data I need to do the next piece. I’m not going to go do it, but I’m going to give you a set of instructions. I’m going to give you a tool to work with and I’m going to basically time you and make sure you’re doing everything you need to do.” So, we’re getting away from probabilistic to deterministic. Correct. But using probability to help it, right? And I think that’s what some folks are. They’re not making that leap. |
| 31:38 |
No, they aren’t. And it’s very difficult to do that because you have to understand first how the models work. Then you have to build your own determin deterministic skill set, instructions, codes, python scripts, etc. So as the deterministic LLM walks through that engineering workflow, what King would do is hey, I’m I’m grabbing data and that’s his only job. |
| 32:00 |
That’s where you get a lot of hallucination with LLMs is they try to do everything right. And what you’ll have is you’ll have context creep or context overload. You can only fill up my coffee mug with so much coffee. As you fill it up with more context, it’s just going to overflow and dilute it and do all that stuff. |
| 32:18 |
It’s no different than us. Correct. In a way, right? If I try to do everything all at once and even try to multitask and push along, you just got to jump in here, man. The floor recognizes king. [laughter] Huzzah. |
| 32:27 |
How many different AIs are there? There’s I think I know Grok, chat, and tons. Oh, there’s a lot. There’s so many. I mean, they’re all kind. I’ve never even heard of that one. Gemini, you have a lot. Is each one really good at something? They based on their model and the training set they all have their strengths. |
| 32:51 |
So what’s chat’s strength because that’s the most popular. So like ChatGPT Codex is really good at coding right it’s built on a lot of coding training sets and it knows you know the Python libraries and the different ways to go out and execute code. I wouldn’t necessarily use like Grok as the main AI that would help me figure out some coding problems, right? |
| 33:16 |
Gemini really good at images and those kinds of things. Creative type work. There’s there’s lots. I think Claude or Anthropic through its suite of tools has developed a good development environment. But then Perplexity is also known for being good for research. So perhaps more in the professional and academic circles where research is really the work product I see perplexity perhaps being used more so than say some of the other models. |
| 33:48 |
Yeah. So when you’re building like these AI tools you don’t have to use just one LLM right you don’t have to use just you know Anthropic or ChatGPT or Open AI and whatnot. What you do and can do is those 20 engineers if one of them needs to be creative, you build your tool to work with Gemini for instance and Gemini will help that engineer achieve that workflow quicker and better. |
| 34:14 |
Right? So if I had out of those 20 engineers a software developer that was doing code, I would have like Codex ChatGPT work with that engineer and help them finish their task better. So, so kind of bringing it back into the fold of our safety margin focus, you know, I think we can see the benefits to the advent of AI technology and the incorporation of it into what we do in our space. |
| 34:42 |
Not just from our consulting side, but even from the owner operator side. What are the vulnerabilities though? Right. So, let’s talk a little bit about that before we kind of wrap it up with our last fun segment. Definitely answers that are not correct, right? So, safeguards: if you just use the chatbot’s chat interface and ask it a question, you get your stuff back. |
| 35:05 |
It is designed to please you, is designed to give you an answer because nobody will use it if they get a wrong answer, right? That wrong answer is going to push them away potentially from that AI tool. And just like a lot of people and you know when you talk to them like AI is going to be a big thing in the engineering world they go well it doesn’t do a very good job at anything except for helping me with my emails right so they’ve already been turned off because the answer that came back was not correct or it was mostly correct but partially wrong so you have to but there is this ground swell that has continued to move though [clears throat] correct and we’ve been part of it where if you can get beyond that definitely there is that upper leg of the K economy and it’s just going to be more expensive to try to get back up there when the computing costs and the data centers really get moving. |
| 35:58 |
Right. So, I think the early adopters do in this case have a bit of an advantage. Yeah, huge advantage. Got it. When we because we’ve built several tools that do a lot of things for us like takes a lot of the busy work out of what we do, right? Allows us to actually do more analysis, check our work even further. |
| 36:15 |
You know, we’re all people and we only have so much time during the day to not only be effective and productive and deliver on time. We obviously want to do a good job, but we can’t spend all day doing that one thing that’s super important every single day. So what we’ve seen is we’ve been able to extract data and information quicker. |
| 36:40 |
We’ve been able to organize it better and we’ve been able to check on it and validate 100% that it’s correct and it allows our engineers instead of focusing maybe 10 20% on analysis up to changing that to maybe 80 to 90% of their time is focusing on analysis which allows them to check more work do more things and do a more thorough report. |
| 37:04 |
A report that would take us on like doing a corrosion control document may take us six, eight weeks to pull it together and it’s 55 pages, right? Going through a facility, this is exactly how this facility is going to feel and have effective damage mechanisms on it and how we want to try to manage it. |
| 37:21 |
It may be now 100 150 pages and pointing out things very important such as hey when you have a rapid depressurization brittle fracture isn’t necessarily an issue when you’re operating under normal operating conditions but hey in your startup shutdown procedures those kinds of things you can have brittle fracture on these nozzles or these pieces of the equipment because of this scenario this is something to flag so think about it put it into your procedures right and we wouldn’t have the time to do that typically. |
| 37:50 |
No, I agree. One thing I’ve heard you say and a term that I think we’ve adopted and you hear it around the office quite a bit is the golden thread. [clears throat] Because I think as folks are listening to this and as they’ve enjoyed their own journey of AI what certainly has been realized by us is you get volumes you can get volumes of just new valuable insight. |
| 38:17 |
Yeah, but then you want to make sure it’s right. So the human in the loop becomes critical, right? The competency of the human in the loop becomes very critical because the golden thread that thread that gets you from the beginning to the end the maze of you know everything AI produces and whether it’s hallucination or not that golden thread of verifiability and auditability becomes critical the human in the loop becomes critical and for us we’ve seen a business model shift around the golden thread and I think that’s what some folks might be struggling with because if your business can’t shift to preserve the integrity which is essentially the failsafe safeguard against all AI vulnerability then you don’t have the safety margin right totally safety margin look at that safety margin that was the most incredible thing I’ve ever seen [laughter] so golden thread just in case we’re you know needing to explain it is that auditability, traceability, and it also gives the engineer that’s reviewing the data the reassurance that the data that they’re looking at was pulled from the right place. |
| 39:26 |
They can see how it traces through the workflow and they can validate it because it tells you this is where it came from. This is the confidence in that data and this is its effect on the output and there’s a sensitivity analysis that’s on there too. It’s really cool stuff. It is cool and I know we could talk about this. I mean, especially, you know, Justin, myself, I think Kings right now sitting there going, “Holy crap, I really got to learn a lot more about [laughter] AI.” |
| 39:49 |
I used it to help with our dog treat business. There you go. [laughter] By the way, one of our unofficial you know, actually, I’ll say I’ll say it’s the unofficial dog treat of the Safety March podcast is Am’s Am’s Pup Keys. |
| 40:06 |
Yeah. And you can get them where? At pupkeys.com. Pupkeys. Pville Meat Market. Or the Belleville Meat Market. Yeah. Yeah. In Belleville, Texas. Is that your shout out? Does that count as your shout out? That is my We’ll do shout outs at the end. Okay. That’s That’s just a I’m not losing my shout out. |
| 40:24 |
That’s a plug. [laughter] Shameless plug. [snorts] That’s a plug. All right. So, but you know, let’s go to the fun segment, right? We’re getting to we want to make sure you guys can hear this, listen to it, maybe across a one-way commute, maybe roundtrip if you have a shorter commute. |
| 40:39 |
But we wanted to do something that is along the lines of safety margin, but perhaps a bit more fun. You know, we have some folks that really cautioned me against this next segment. But I think that’s why we’re really going to have fun with it. So, it’s actually called medieval battle safeguards as asset integrity. [laughter] But before we wrap up the segment, just so you know, I am going to tie it in to my new Blackstone griddle. |
| 41:05 |
Oh, nice. Yeah. Yeah. Yeah. Because you know I think you see where I’m going with this. I see where you’re going. So, but anyway, to kick it off, recently my wife and I were overseas on a trip, 25th anniversary celebration. Happy anniversary. Oh, thanks. Appreciate it. Big fan of marriage. |
| 41:17 |
And go marriage. Go marriage. [laughter] but I like going to museums and I like seeing things especially when it comes to armor and swords and caliber. |
| 41:30 |
And one thing I kind of internalized this trip over to you know Germany, Austria, Hungary is there was actually quite a bit of work to preserve the integrity of battle armor. Definitely. Yeah. So initial thoughts on this? All I think is that’s metal. Metal corrods. You think of any metal outside doing any of its thing in a non-airconditioned environment, it is going to not last a very long time. |
| 42:00 |
So, you have to preserve it in some way, shape, or form. So, who was doing this? Was it the blacksmith? Was it the servants, for lack of a better term? I think the blacksmith was obviously the craftsman, but you probably had the armorers. You know, there had to be people in medieval times, if you had a big castle and a whole bunch of knights and those kinds of things, somebody that was just there taking care of the battle armor, right? |
| 42:24 |
So, the armorer likely knew that, hey, if I don’t condition this on a frequent basis, then we’re going to have to go buy more armor, and the boss doesn’t like that. The knights don’t like that because it’s got to be custom fit and whatever it may be, right? Yeah. But what I as you dive into it, what you realize is you had campaigns that lasted months and years for sure. |
| 42:46 |
Away from you know hardened civilization, right? Away from a castle keep or a a city. Yeah. Right. Back in the day. So they were out there you know German forests right along the Danube the Bavarian the black forest. The black forest [laughter] like the Ottoman Empire. |
| 43:08 |
But anyway, so these armies, they had to have, I guess, swaths of maintenance people. So, is this the advent of maintenance and integrity? It could be for sure. I think Yeah. Like the movies don’t do a good job of showing all the logistics to see that, right? I would watch it. Two hour I would watch a two-hour film of somebody of somebody polishing armor. |
| 43:27 |
It’s like pressure washing. Yeah. So, I’m sure the knights that were out there that had their armor and their swords and stuff, they had some type of oil to keep their armor from going, you know, tarnished and getting rusted out and all that good stuff. So, they probably did their own. |
| 43:43 |
But there was probably a lot of those knights that were relatively famous and I mean, these were noble men sometimes. So, so what do you think the prevailing damage mechanism was? And how does that play out against a like a single breast plate versus like say chain mail? Sure. I mean, I’m more of a chain mail guy, right? |
| 44:05 |
Yeah. I mean, you’d have atmospheric corrosion, right? For like nice smooth clean metalized surface mineralized surface, it would not be good. Yeah. That’s You’d have leather in Houston. But with chain mail, [laughter] you’d get contact-point corrosion at the ringlets, plus fretting from all that surface contact. So the ringlets may not be a great idea, since there’s a lot of surface area there, and to keep it nice and clean and oiled up, you’re going to have to do a lot of oiling. What oil did they use? Linseed, I’m gonna bet. |
| 44:51 |
I’m gonna bet in Hungary. I saw a lot of sunflowers there. Sunflower. Sunflower one. Yeah. Interesting. That’s a good one to Google. Yeah. We’re going to have to We’re going have to get some golden thread on that. Okay. That is a We need to Google that one. All right. He’s giving a thumbs up like now. |
| 45:05 |
All right. So, hey guys, it’s been a lot of fun. First episode of Safety Margin wrapping up. I think we want to give a couple of shout outs. Yeah, definitely shout outs. So, I’m going to give a shout out to API Mechanical Integrity Summit, which is coming up. So, we’re going to be there doing that stuff. |
| 45:20 |
So, thanks API. Excellent. King, I want to shout out to U pass. [laughter] All right. So, I’m actually going to give a shout out to Michelle Horwitz over at AIChE. We recently onboarded into you know, CCPS, became an official member, and she made the process easy and smooth. So shout out to her. |
| 45:42 |
We’ll certainly tag her when we post this. But as for safety margin, be safe out there. Keep questioning your assumptions and, you know, we’re the safety margin. Yeah. See you next episode. Safety margin. Good stuff. Thought we’re doing rapid fire. [laughter] Okay, outtakes. |
| 45:58 |
Let’s keep it rolling. Rapid fire. Let’s go, King. All right. Outtakes. Okay. Oh boy. Some of this stuff is definitely not going to make it. No. You’re ready. Pull your computer out. [laughter] Okay. |
| 46:09 |
Paper or plastic? Plastic. Plastic. Interesting. Dark chocolate or milk chocolate? Dark chocolate? Dark chocolate. Hold on. It’s like saying an oil and gas person shouldn’t choose oil and gas, right? They like would be driving a Tesla. That would be crazy. You know, there’s a lot of plastic in a Tesla. Gotcha. Okay. Okay. |
| 46:32 |
Fair enough. Okay, let’s keep going. Sorry. Milk chocolate or dark chocolate? Dark. Dark. Text or phone call? Text. Text. I’m a phone call person. It’s crazy. I know you are. Sweet or salty? Sweet. Sweet. Beach or mountains? Mountains. Mountains. Oh, I’m I’m a beach person. Dog or cat? Dog. But you know what? |
| 46:53 |
I have a I mean, amazing daughter. She loves cats. My wife likes cats. So, I’m actually allergic to cats, but you know, we have cats. We’ve had a cat as long as I’ve been I used to have cats. I’m allergic to them. The things you do for love. Books or movies? Movies. Both. |
| 47:11 |
Cook or order out? It’s tough because with a family like it’s expensive and this one’s not going to eat that and all that. So order out. [laughter] I like I like cook. |
| 47:30 |
I wish I did cook more. Fortunately I married a great cook. So yeah, but I like cook. I like eating at the house and I like eating food that you know controlled. We cook a lot. That’s why I want to order out. Oh, [laughter] summer or winter? |
| 47:41 |
Winter. I like winter. Ah, winter. Yeah, but we live in Houston. I We get tired of the heat. I’ve lived in Canada. No. Winter. You know the scene from Gladiator. Yeah. Tries to pull the sword out. The frost sticks to the blade. Apple or Android? Apple. So, I was an Android person. I’ve switched to Apple for various reasons, but I will tell you I do miss my Android. |
| 48:08 |
There are things you can do in Android. You cannot do an Apple. I’ve had both, but Apple’s just the working across multiple devices across the family, all that stuff. It’s amazing. Yeah, fair enough. Chicken or beef? Beef. Beef. Beef or pork? Beef. I’m going to go beef. Beef. Finally. Gosh, but chicken’s good. |
| 48:28 |
I want a chicken sandwich though. Light beer or dark beer? Dark. Dark craft. Now, hold on. Sometimes on a summer day after mowing the yard, you can’t have a dark beer. I mean, you need that lager. You’re not grabbing like good Mexican lager. [clears throat] You’re not grabbing like a barrel-aged stout, 10.8% alcohol. |
| 48:55 |
No Guinness. Well, yeah, probably. You know, one last outtake shout out. We were at this monastery in Engelszell off the coast of the Danube off the banks of the Danube and they had three beers and they had a dark Gregorius beer. Gregorius. Oh my gosh, it was divine and which is incredible because that leads me to my last question. |
| 49:16 |
Heaven or [laughter] favorite Gregorian 90s chant band? What? The one that [laughter] The one that had Enigma. Yeah, Enigma. There it is. Enigma. I don’t even I should be the one, guys. You’re the You’re the band. Yeah. Gregorian Chant. It was a hit in the 90s. Everybody Gregorian chanted in the 90s. |
| 49:48 |
Late 90s, mid90s. Oh, [laughter] shout out to my wife and [snorts] my mom. Nice. There you go. Yeah. Is that a wrap? It’s a wrap. That’s a wrap. Safety margin. Thanks, Intern Alex. |
