Sign in or Join the community to continue
How a Logistics Giant Keeps AI Data Locked Down
Posted Oct 05, 2026 | Views 6
Share
Speakers
Jason Ward
FinOps Manager @ Slingshot Aerospace
Jason Ward, MBA is a FinOps Manager at Slingshot Aerospace with 15+ years of experience across cloud infrastructure, architecture, and platform engineering. He specializes in cloud cost optimization, AI and Kubernetes workloads, and building financial accountability into engineering practices.
+ Read More
Alex Salkever
VP - Content and Research @ The Linux Foundation
Alex Salkever is a leading expert in exploring the intersection of technology, business, and society, with over two decades of experience covering cutting-edge advancements in a wide assortment of fields such as AI (and ChatGPT), green energy, genetic engineering, cloud computing, virtual reality, and self-driving cars. As a former editor of BusinessWeek and an award-winning author, Alex has a unique perspective on the ways in which technology impacts our lives and well-being. Based in the heart of Silicon Valley, Alex has firsthand access to emerging technologies at the forefront of development and adoption. He regularly engages with researchers and innovators working on over-the-horizon ideas that will shape the future.
+ Read More
SUMMARY
Picture a startup with ten engineers hammering away on AI and one person lying awake over a $20,000 bill that hasn't arrived yet. That's the scenario we put to Jason Ward, who handles FinOps for AI at C.H. Robinson and recently joined the FinOps Foundation's AI working group.
Jason's first answer is not glamorous: tag every AI resource so you know who owns it. The rest of the episode is what that makes possible. He walks through how C.H. Robinson runs AI across order entry, quoting, booking, and tracking, and why the team routes Anthropic models through Vertex AI to keep data locked down.
Then come the metrics. Jason uses AI to dig through his own observability platform for signals he didn't know were there. One of them is how chatty a model is. That signal turned a prompt bloat alert into a bug in the code that kept retrying and burning tokens. Its opposite, context starvation, burns tokens too: a model with too little context keeps failing and trying again.
We also cover why agentic and conversational workloads need separate baselines. Jason explains why cost per order is the easy win, and why most of the real work doesn't fit into neat discrete tasks. That's where his experimental cost per thought metric comes in, with reasoning ratio and cache hit rate alongside it. His advice is simple: your AI is the best tool you have for understanding your AI.
+ Read More
CONTENT & TRANSCRIPT
Alex Salkever: [00:00:00] Imagine that you're talking to a friend. He's worried 'cause he has 10 engineers hammering away on AI. He's worried that he's gonna get a $20,000 bill or something like that. How do you, in your role, try to measure, observe, analyze, and make sense out of AI usage? If you were gonna give startups advice, how would you build that program?
Jason Ward: First, I would suggest
Alex Salkever: I'm here with Jason Ward of C.H. Robinson. Jason, tell me, uh, your title and what you do and what C.H. Robinson is, please.
Jason Ward: Uh, my name's Jason Ward, of course. Uh, I work with C.H. Robinson. It's a third-party, uh, logistics company. Just recently I've joined the AI working group within the FinOps Foundation. So C.H.
Jason Ward: Robinson's one of the, I would say, leading, uh, third-party logistics companies fully utilizing AI in a lot of what they do. Not everything, of [00:01:00] course, but a lot of what they do. So I'm trying to get ahead of the conversations that could be coming. So the ROI questions, the value questions about what services are bringing in what value, and so I've been really trying to get ahead of that and, uh Just trying to get the right people the right information so they can use that for what they will.
Alex Salkever: Very cool. So I'm curious, logistics companies, not something that everybody thinks about all the time even though they're critical.
Jason Ward: Right.
Alex Salkever: What are some of the AI use cases that you see inside of C.H. Robinson that have been pretty successful or areas you're exploring for AI?
Jason Ward: So a lot of the areas that we use for AI currently is in from like order entry, email classifications.
Jason Ward: Um, just recently they announced, uh, that we use it for our full end-to-end logistics, so pricing, quoting, um, booking the loads, tracking the loads. If you can think of anything logistics, we're using it to help that, that platform out.
Alex Salkever: And w- what platforms are you using, uh, for AI right now? Um- I think you mentioned Azure and a bunch of things around that.
Jason Ward: Yeah. So right now we're mainly an Azure [00:02:00] shop using Azure OpenAI. We are doing some testing in GCP with, uh, their Vertex AI, also using their Anthropic models because we can use GCP as the broker, and we know that the data will stay with us as opposed to sometimes using the other open A- or the APIs where you have to share your data with a lot of these other companies.
Jason Ward: We can lock that down.
Alex Salkever: Tell me why that matters.
Jason Ward: You know, for obvious like security reasons, stuff like that. We don't want a lot of data being shared or spread. There's stuff you gotta think about, right? And it's almost to the point where some of the stuff, if you were in even like a government space, you'd want to obviously to have an on-prem data center GPU to really have your models on there and your data never leave.
Jason Ward: So it's kind of around the same thing. You just don't wanna exactly have all that information out there to other people or even if it doesn't get shared, there's a possibility.
Alex Salkever: Very cool. Okay, interesting. Yeah, I m- I may come back to that a little later. I'm interested in hearing how the developers specifically are using AI and how you're observing that and trying to understand the way that they're [00:03:00] really leveraging it, uh, effectively or ineffectively.
Alex Salkever: Like, I mean, what, what are, what are you doing to get at that problem?
Jason Ward: So a lot of what, like the, from the developer side, I know they're using AI in a lot of the services. Obviously, a lot of companies come down with it too, as well as ours, that, you know, they, you can fit AI into your workload, try to find a way to automate that process.
Jason Ward: Um, and another way developers are using it is obviously with GitHub Copilot, um, using it that way.
Alex Salkever: So for the code themselves. For the
Jason Ward: code itself, yeah- Okay ... to help kind of, you know, move it along, make the mundane not as mundane so they can kind of speed up the process. Not full vibe code stuff, but obviously just use it to help read the code repositories, find the errors, find where the changes were made and, you know, and, and suggest fixes for those codes.
Alex Salkever: What's your favorite AI hack that you've seen the developers do to, uh, automate the ba- uh, the obnoxious parts of their lives?
Jason Ward: Honestly, I have one more for myself than the developers.
Alex Salkever: Fair.
Jason Ward: And so I, I kind of see where I use AI to hack AI, and I use our current, uh, uh, LLM observability platform. It has this LL agent that will go [00:04:00] in there, and I use it to help find metrics I didn't know existed to help me show more value around our AI spend.
Jason Ward: So it can... I- it hunts itself in a way and finds the metrics that are in our platform that may or may not be known to myself, and then I'm able to compare that with a cost and be able to get better, uh, dashboarding around that.
Alex Salkever: Can you give me a for instance around that? That sounds really interesting.
Jason Ward: Yeah. So, like, one, I'm able to find... It's kind of hard, there's a lot of math involved in it, but I can look at the, what they call, like, the verbosity of the, of an API call or the AI call. So it checks how chatty it is, so if it's doing way too much stuff, so it's, uh, kind of goes along with, like, prompt bloat, um, right?
Jason Ward: Or, and so it, it looks at the context and figures out is it talking too much to get the answer when it could say things less and, and be more efficient.
Alex Salkever: W- I mean, how does it do that? Is it looking for just, obviously so, like, prompt exhaust, but is it just average prompt si- or, or, or, like, average-
Jason Ward: Yeah,
Alex Salkever: like for me-
Alex Salkever: conversation size, token consumption or?
Jason Ward: So many calls per trace-
Alex Salkever: Okay ...
Jason Ward: and, and, and uses that way, [00:05:00] and then it comes back with, like, and then it was able to come up with the math. I'll be honest, I used AI a lot to kind of help- No, no ... hack it. Yeah. So as, uh, my brain just couldn't wrap around some of the way these tokens and stuff work and the full traces and how all that stuff works, so it was able to do that, figure it out, put the math, and then come up with, like, a scorecard kind of around that.
Alex Salkever: So this, and this is actually, like, you have that now-
Jason Ward: I do ...
Alex Salkever: as part of your arsenal. I do,
Jason Ward: yeah.
Alex Salkever: That's really cool. Yeah. So essentially AI hack AI to give you metrics that you didn't even know existed.
Jason Ward: Yes, exactly. I mean, it helped to have them already there to begin with because if you don't have the observability around your traces, your spans, and all that, you're, you're missing half of the data that you could use to track your AI value, your spend, uh, how well it's working, right?
Jason Ward: And if you don't have that in place, then you're missing out on a lot of the information you need.
Alex Salkever: So if, imagine that, uh, you're talking to a friend.
Jason Ward: Mm-hmm.
Alex Salkever: You know, he's maybe the AI engineer or the sort of chief bottle washer at a startup company- Mm-hmm ... A, B round or something. He's worried 'cause he has 10 engineers hammering away on [00:06:00] AI, you know, and he's worried that he's gonna get a $20,000 bill or something like that.
Alex Salkever: How do you, uh, in your role- Try to measure, observe, analyze, and make sense out of AI usage. So essentially what, like, people used to do with FinOps, but now with AI.
Jason Ward: Right.
Alex Salkever: Uh, but I believe you have, 'cause we talked about this before the, before the interview, uh, some pretty granular models and some thoughts, methodologies on how to make this work.
Alex Salkever: So, like, advising, if you were gonna give startups advice- Yeah ... how would you build that program?
Jason Ward: First, I would suggest, uh, finding the best way that you can to tag the resource, because that way you can track the ownership of who owns it. Uh, second, uh, I would look at your token usage.
Alex Salkever: Um- Resource, you mean...
Alex Salkever: Can you tell me what you mean by resource? Yeah. Is it the model, or is it the, the-
Jason Ward: So I would say if you wanna take it to the model, go to the model or your workload, so whatever your AI app is or your agentic workflow resource that way.
Alex Salkever: Okay, so it could be, like, a coding agent. It could
Jason Ward: be- Uh, yeah, [00:07:00] exactly.
Jason Ward: Okay. Whatever that particular AI workload is, try to find a way to tag that to show your ownership of it.
Alex Salkever: Okay. Yeah, I ask partly 'cause, like, I know Copilot, you could run, like, 10 model... I mean, there's-
Jason Ward: Yeah ...
Alex Salkever: almost unlimited number of models that you, you could switch back and
Jason Ward: forth. Right, and Copilot's the one area that's hard, 'cause I'll be honest, I myself don't even have access to the billing on that.
Jason Ward: Oh, okay. I can just... So I'm very- Yeah ... so even though I'm acting as a FinOps role, I haven't been giving- Yeah ... been given every- the full reign of everything. So eventually when I am, I know how I kinda wanna do it, and I, uh, I'm looking more towards value around it than the actual spend. Right, right. So I plan on taking, like, a survey, finding out how many hours are saved, average that out, and then do cost per hour saved- Right
Jason Ward: to kinda show a value. Yeah. And that's what I would kinda tell somebody else, right? Because spend anymore that I've seen in the AI space, it's just not the entire story. It's, it's only part of it, and AI spend can fluctuate. You could have AI spend go up and a model be way more efficient, and therefore you're getting more done compared to a model that, say, is cheaper but is taking way [00:08:00] longer, and therefore- Sure
Jason Ward: still costs more in the long run. So to me, efficiency and being able to show the value or the ROI of a service has been way more fruitful than, than just spend alone.
Alex Salkever: So dialing back to where we were, tag the resource is step one. Yeah. What's step two?
Jason Ward: Uh, step two would be find a way to show it. Um, get your...
Jason Ward: Once you have your ownership, break it down por- per, um, uh, AI app. So show by, I would do cost by app. Um, another way if you want just more genericized would be you could even break it down to cost by model. I do that. Um, I track spend and efficiency by model, so devs can look and see what's the best model per their workload so they can pick that.
Jason Ward: Um, and then make sure you're talking to the right people, because some devs aren't necessarily gonna care, but the leads may, or the leads may not care, but the devs do. And if you can start helping the developers at least see what they can do to kind of reign that in a little bit or, [00:09:00] you know, they're not given enough context or too much context in their prompts, then maybe then they can tackle that onto other people and it kinda, you know, trickles out to others.
Alex Salkever: Tell me a little bit about how one of those instances worked for you with. So you have a lead dev who's super token- Mm-hmm ... conscious, wants to, wants to be efficient. Uh, you know, what, what was something you called out for, for them and that they, they said, "Huh, I, I'm surprised"?
Jason Ward: Uh, one of the services, I, I, I do tr- uh, prompt bloat, and one of them came back and it was extremely, like, high.
Jason Ward: And so I looked at it and real- and I reached out to the owner who owned the service, and come to find out there was a error in the code so it kept causing it to loop, and as it was looping it retries, and when it retries it's reusing all those tokens again. Yeah. So the prompt bloat led to the retry on that and all the token usage, and they were able to catch it, change it, and then it was off the list.
Alex Salkever: Huh. So- Super interesting.
Jason Ward: Yeah.
Alex Salkever: So, so I mean, basically you saw prompt exhaust or some signs that something was not quite right, and that was a sign that there was a problem in the [00:10:00] code.
Jason Ward: Yeah, yeah.
Alex Salkever: So essentially almost like your QA for the coding team.
Jason Ward: Almost. It, it can almost act like that, yeah. Same with, uh, I have one for context starvation, so people who don't provide enough context in their prompts.
Jason Ward: Oh,
Alex Salkever: really? Tell me about that.
Jason Ward: So what that does is it'll s- read the context and it, it's a little s- I'm just basically working with it right now so it's not all the way accurate, but what it does is if you say you want it to do something, it returns and it errors out because it doesn't have enough context to support what the job is supposed to be.
Jason Ward: Then again, it goes back into the retry loop until it gets it right. So therefore you can go to the dev and be like, "Hey, this isn't, this is failing because you don't even have enough context in here." Well, then they can make a change to the code again. So it kinda equals the same end, but from different areas, I guess, of-
Alex Salkever: Interesting
Alex Salkever: yeah. So it's almost like you're starving s- you're starving your co- your code.
Jason Ward: Yeah, yeah. You're starving
Alex Salkever: your model, so therefore you waste because it keeps coming back 'cause it- It- ... like, "Feed me. Feed
Jason Ward: me" Yeah, exactly.
Alex Salkever: Huh. So interesting.
Jason Ward: Yeah.
Alex Salkever: Uh, w- w- w- where do you tend to see those types of situations pop up the most?
Alex Salkever: I mean, what, what kind of [00:11:00] applications or what kind of use cases? Um- Or is it all over the
Jason Ward: place? It's kind of just all over the place right now. Okay, fair enough. Yeah. Yeah. There's, there's not really anything straightforward about it. I mean, everything changes so fast, it, it's hard to really keep in, and find that trend still, so yeah.
Alex Salkever: So you, you'd mentioned, so to tag the resource, uh, identify the, the s- sort of where the use is.
Jason Ward: Mm-hmm.
Alex Salkever: Right? Is that correct was the second one? Yeah,
Jason Ward: yeah.
Alex Salkever: And then, um, know who's who in the zoo, basically. Right. So who you can go to.
Jason Ward: Yes.
Alex Salkever: Uh, w- when you are reporting all of this stuff- Mm-hmm ... out to, you know, to your boss, what are the metrics that you're reporting on or what are some of the newer metrics, like you talked about vibracity.
Alex Salkever: I'm curious if, like, sort of what does your reporting arsenal look like?
Jason Ward: So it, it ki- right now, honestly, like, for our NBRs or our showbacks It's mostly just total AI spend, and then it's broken down. We use, um, more ROAs, ROI based, so cost- Right ... like a value. So for like our order entry, we use cost per order.
Jason Ward: So how, right, and then that's, we know that's a total- So that's an
Alex Salkever: easy one
Jason Ward: to measure. Yeah, that's an easy one. Then the other one, we try to break out, [00:12:00] um, cost per automated task. So-
Alex Salkever: Oh, interesting ...
Jason Ward: so the automate- the task that's fully automated, the AI is working on, we break that out. I talk to the analytics team, 'cause they're the ones that handle tho- that side of it, and I'm able to get the information, and then we can do cost per automated task.
Alex Salkever: So they can actually trace, like how many tokens are required per automated task? Correct.
Jason Ward: Yeah.
Alex Salkever: So this works really well for dis- sort of discrete activities- Yeah ... or discrete tasks.
Jason Ward: Yeah, yeah.
Alex Salkever: How much of your workloads are, are those kinds of tasks? And-
Jason Ward: Honestly, not a lot. Um-
Alex Salkever: Okay. So this is nice, but it's low-hanging fruit.
Jason Ward: It is very low-hanging fruit, yeah. So like the other thing's more around efficiency than anything, and we don't use them in the NBRs yet. It's still go, go, go, but there's, there hasn't been a c- need to come down for true value or efficiency around it, just on a couple apps here and there, but nothing really, um, bulk- nothing in bulk.
Alex Salkever: So, so are, are your developers then, it's not like- They have gas town run. I mean, it's not like they have these enormous agentic loops or they're like- No ... rob- robining across 20 different agents that are running at the same time.
Jason Ward: No, no,
Alex Salkever: nothing like that.
Jason Ward: Okay. Mm-mm.
Alex Salkever: So you had mentioned that you actually [00:13:00] break down between agentic and conversational.
Alex Salkever: Trying to. Tell me what that means, yeah, and sort of how you do that and why you do that.
Jason Ward: So again, it goes back to the prompt bloat. I ran into an issue where the, my top five apps that were coming back with prompt bloat, I would reach out, and it turns out they weren't, like, single-shot prompts. They were m- agentic workflows.
Jason Ward: So it's where they run multi prompts. Well, the way I had it set up, it was just picking those multiple prompts up as one and saying the context was just bloated. So it was giving me a false, false positive, I guess, or a false negative, whatever you wanna call it. And so I was trying to find out a way by just doing research and looking to see what could be best, and it seems that until you divide that workload type or what it's actually doing, then you can get more, then you can get more precise on that, that, um, metric.
Jason Ward: So by having, like, an agentic workflow versus a conversational workflow where it's just a single prompt back and forth conversation, right? Where the agentic workflow is just constantly going. So if you can divide that out, then [00:14:00] it's easier to separate and show true bloat versus what's, what's not.
Alex Salkever: Is that like...
Alex Salkever: Or is there a way to apply, like, a sub tag to the span or something like that?
Jason Ward: Very well could be. I, I'm very much early on in that. Oh, okay, cool. So I'm still trying to figure that out myself, yeah. Yeah.
Alex Salkever: So, so back to the, the, the, you know, helping your friend. Mm-hmm. Uh, resource, uh, coding infrastructure, basically.
Alex Salkever: Mm-hmm. Talk to devs. It sounds like also, uh, understand very clear- like, b- look for these additional metrics that, uh, you know, might, might be interesting or useful- Yeah ... for your particular use case. Yeah. Uh, also, it sounds like making sure you understand the actual context of, of the a- of, of the conversations 'cause that's- Yeah
Alex Salkever: it's not just looking at the token flow enough, isn't, uh, all by itself isn't enough. You have to understand a bit more about what they're trying to do.
Jason Ward: Yeah. The more you know, the better, honestly. I mean, the more... Ideally, from, I would say, the, the sooner you can be involved in what they're doing and figure out how their workflow works, the better you can report on it.
Jason Ward: If you don't know enough about it, you can't report on it. [00:15:00] So having that knowledge base is, is huge.
Alex Salkever: And your CTO, where does he, over time, hope that this all goes? I mean, like, in your perfect world, what does this look like for you, uh, you know, being able to package up true intelligence and o- of course, following on that, uh, optimization and control?
Jason Ward: That's a tough question. I wish I could answer it all the way. I don't know. I would hope he would look at it and see a way that, uh- We could have everything, not everything 'cause it's just impossible, have the bulk of it monitored and be able to show more value for everything that we have. Um, not just a few apps here and there, but be able to really say, using AI created this much value and it, it helped us with X or whatever the...
Jason Ward: whatever he wants to use it for. Um, it's kind of a hard question.
Alex Salkever: Yeah. No, I understand- Yeah ... 'cause it's like on the one hand, like measuring how much, how many tokens it takes to fill out a form, that's easy. Right. Whereas like measuring a much harder task, uh, you know, like if it's sales and [00:16:00] marketing or if it's, uh, some other type of forecasting or something- Yeah
Alex Salkever: you know, where the savings might not be as obvious because- Yeah ... it's not that discreet. Yeah. Um, how, how are you though? 'Cause I mean, obviously if the bulk of your AI consumption is not with these discreet tasks, uh, how, how are you trying to measure efficiency and outcomes?
Jason Ward: One of the things I'm using is c- I, I call it cost per thought.
Jason Ward: That's an efficiency metric I'm using. Yeah. Tell me about
Alex Salkever: that.
Jason Ward: And so what that does is broad, so it's your total cost divided by the, the trace. But it looks at like your RAGs, your, your, uh, your tokens, your conversation. It talk- It takes the entire trace, and then it, it adds it back. It does some math on the backside of it, and then it's able to come up with what they call a thought, or you could call like cost per I'm trying to think of another way to word it
Alex Salkever: How do you know when to divide?
Alex Salkever: Like, what, what are the dividing lines of a thought?
Jason Ward: So it, that's the hard part. I'm still, I'm working on it. Okay So it's a work in progress. Like,
Alex Salkever: where does it end and where does it go?
Jason Ward: And like- Yeah, yeah. So you basically have to go from [00:17:00] completion. So input from the time it takes the input token to the output token, and then it traces that entire trace, and then it takes the output, and then it figures how efficient it is based on that.
Alex Salkever: So this is... When you say thought, you mean an AI, an AI's- An AI, yeah ... thought as opposed to the human's- Compared to the human ... conversational flow?
Jason Ward: Yeah, yeah. Sorry, yes.
Alex Salkever: Interesting. Okay. Yeah. So basically, like, when you s- s- s- essentially from, like, first input to last output-
Jason Ward: Cr- yeah, yeah ...
Alex Salkever: how much, how much, how much is it putting out?
Jason Ward: Exactly.
Alex Salkever: And, and how do you sort of benchmark that or, like, compare it, uh, say, you know, from prompt to prompt or from, you know-
Jason Ward: That, that's the hard part. So I... The idea is there, but the execution is, is definitely a work in pr- And that's an area that I'm honestly trying to look at and edit as I go, because I need data to back that up.
Jason Ward: So once I get a couple months worth of data... I mean, this is still pretty new. I've been thinking about it and trying to implement it for a while, but the actual practice of it, I'm still trying to get all the metrics together to myself to confirm what I'm doing is, is working.
Alex Salkever: Very cool.
Jason Ward: I [00:18:00] would say another metric to pick would be, um, either reasoning ratio, so how m- the ratio of how many tokens- Oh
Jason Ward: are actually reasoning versus not, and cache, uh, hit rate. So I track cache hit rate percentage, so the prompts that are hitting cache prompts. So if the service is hitting a cache prompt, obviously it's gonna- Yeah ... save. And doing that, you can see 60 to 70% savings, you know, just in that alone.
Alex Salkever: What's a reasoning rate?
Jason Ward: The reasoning rate is, like, your toke- your reasoning tokens versus non-reasoning tokens.
Alex Salkever: Got it. Okay. So when the, whether when you're hitting a, a thinking model or- Correct. Yeah, yeah ... like, a chain of thought model or something. Yeah, yeah. Is there... I mean, is that just because it flips in the model and it's easy to tag 'cause it says, "Thinking, thinking," or-
Jason Ward: Yeah
Alex Salkever: how, how do you figure it out?
Jason Ward: Um, so I... Right now it's just based off of a reasoning model versus a non-reasoning model.
Alex Salkever: Okay. So- So whether they're using one or the
Jason Ward: other. Other, yeah.
Alex Salkever: So that's a little sort of a crude measurement of it.
Jason Ward: Right, yeah, 'cause f- right now we don't do any model switching with a router to where it goes through, finds the correct model, and uses it.
Alex Salkever: Okay.
Jason Ward: They deploy based on a model that they pick.
Alex Salkever: So the, the session is a reasoning session or a non-reasoning session.
Jason Ward: Yes, yes. Got
Alex Salkever: it,
Jason Ward: okay. Yep. And then we [00:19:00] have our LLM traces that we can... or our observability platform that we can use that measures all of that. Sure. And then I pull it from that platform.
Alex Salkever: You said you haven't started exploring yet with, uh, um, dynamic ra- I mean, dynamic routing is very new to
Jason Ward: people and- It's something that I've been looking at and researching and trying to bring up here and there, but it's something that I would love to see used or try to be used if it could. 'Cause I think it would, it would help in the long run for sure.
Alex Salkever: And what was the second metric you mentioned, um-
Jason Ward: The cache per-
Alex Salkever: Yeah. Oh, that's... Which is an obvious one, but like- Yeah ... how... I mean, are you... Is there, like, some kind of automated function where you essentially look at, um, prompts that are common? 'Cause I'm assuming that's what it is, or responses that are common.
Jason Ward: Yeah.
Alex Salkever: And I'm su- obviously they're not gonna be exactly the same- Right ... but they're, like, 95% the same so that you can cache them. Is that- Correct. I mean, how do you approach the, like, loading a cache and, and, and using it the right way?
Jason Ward: So again, it's all handled through the observability. I use my ob- I rely really heavily on the observability platform- Okay
Jason Ward: because it tracks all that through its, its, all the metrics and stuff that it pulls. And then I, in turn, just, I, again, I use the [00:20:00] AI part to go through, hunt for those, find the correct metrics, and then bring it back.
Alex Salkever: So the best tool for a, a, a smart FinOps AI practitioner is use your, use your AI to, to improve your AI.
Jason Ward: Honestly, yeah. And then, uh, you just- It's good to- As you research it and you know what you want, and you see what other people are doing or talk about, use the AI to kinda help figure out how to work in your environment, because every environment's different. So what you can find work better in your environment, the better off you'll be.
Alex Salkever: I think that's a good note to end on.
Jason Ward: Okay.
Alex Salkever: Thanks, Jason.
Jason Ward: Thanks
Alex Salkever: for having
Jason Ward: me.
Alex Salkever: Appreciate it.
+ Read More
Watch More
How To Move From Barely Doing BI to Doing AI - Building A Solid Data Foundation
Posted Dec 16, 2020 | Views 1.3K
# Ternarydata.com
How Data Platforms Affect ML & AI
Posted Jan 26, 2024 | Views 538
# Data Platforms
# AI
# Machine Learning
# The Oakland Group
Best Practices from a Live European AI Agent in Logistics // Vanessa Escande // Agent Hour
Posted Jan 09, 2025 | Views 296
# Logistics
# Europe
# AI Agents