The cost of intelligence: AI FinOps and the future of AI spending

[Henri] (0:02 - 0:08)

But you have no idea, did it just use, let's say, a hundred bucks to generate that answer or like a one cent?

[Pinja] (0:12 - 1:23)

Welcome to the DevOps Sauna, the podcast where we deep dive into the world of DevOps, platform engineering, security, and more as we explore the future of development. Join us as we dive into the heart of DevOps, one story at a time. Whether you're a seasoned practitioner or only starting your DevOps journey, we're happy to welcome you into the DevOps Sauna.

Hello and welcome back to the DevOps Sauna. This is a topic I want to first tie into cloud and how cloud bills and cloud costs were actually organized and managed throughout the years. Because for years, many organizations were worried about their cloud bills, and there was a learning curve on how to manage those infrastructure costs and optimize the workloads and build the FinOps practices around Cloud.

And we're pretty used to those traditional software licensing prices. But the cost of AI can change this dramatically from one month to the next. And today I want to talk to you about how these costs should be managed.

So AI FinOps is the topic of today. I'm not here talking about this by myself, but I'm joined by our principal consultant, Henri Terho. Hey, Henri, welcome.

[Henri] (1:23 - 1:26)

Hello, hello, and great to have me and talking about money and AI.

[Pinja] (1:26 - 1:30)

And what do we mean when we say AI FinOps? How would you define it?

[Henri] (1:30 - 1:33)

I think it's important to talk about AI FinOps here.

[Pinja] (1:33 - 1:33)

Exactly.

[Henri] (1:34 - 2:16)

We are probably not finance gurus or something like that, but more talking about the costs and how to basically manage the operating expenses of your AI, looking into the tokens, looking into where your spend is, what are you buying from where, and not just getting the bills, but actually having some kind of way to understand what are you even buying, how much are you buying and how are you budgeting it and how you're controlling that. So it's not just about getting the bills and paying them upfront. As I said, AI is going to be very unpredictable.

And I think everybody's been following trends now that the prices are only going to go up. So the questions will be, how do we manage this? And what do we even do with AI?

Do I do all the funny pictures with AI or do I do just the mission critical stuff or how do I do this? So it ties into many areas of your company now.

[Pinja] (2:17 - 2:38)

It does. And just thinking about a regular organization, you have your business users, you have your technical users, and they have very different needs. And AI at the moment is very easy to consume.

And to be honest, in the past couple of years, I don't think we've been thinking about this too much. So teams are actually having the access to very powerful models, but do we always need those?

[Henri] (2:38 - 3:46)

Yeah. And that's a good question. What models do you use and where do you use them from?

And I think even in our organization, the differences between the heavy users and the light users are many orders of magnitude now. And I think the funny stuff is it's not the programmers who are using it the most. If you look at the tokens, I think we have one guy from basically for sales.

We have one guy from technical and one from somewhere else who are the top leaderboard people who are using the most tokens. So it's not just like building code, but it's actually the prediction of AI spend is much to the individual of what's going on. And it's not just that the engineers will use it the most or what.

But now with agentic workflows, and I think it's something that maybe we want to define that if you do kind of like chat-based AI usage, you typically don't get the agentic. So the agents do something in the background. So you just get the answer from a model, you get the chat and all that.

So that's quite predictable on how much time and effort you actually use and how many tokens you use. But when you do agentic, like telling the agent, okay, hey, find out everything you can about, I don't know, physics for me and make me write a paper on that in the background. And then it churns for, I don't know, half a day and produces you a 30-page paper.

Is that sensible? No. Is it predictable?

No.

[Pinja] (3:47 - 4:12)

Absolutely not. And that's why the resulting cost can be very difficult to predict and allocate and then optimize. And we are still in many organizations where we're thinking about it per token, per API call.

And of course, we see the bill at the end of the month, how much have we spent on it. But now the things and rules are kind of changing. There is a big difference to how this cost is actually being broken down.

And the token cost is one of the most obvious ones, right?

[Henri] (4:12 - 5:26)

Yeah. And I think the token is the best scam the AI industry has put on the whole market, if you think about it, it's like you ask them, okay, what's a token? And they say, well, we don't know.

It's a unit of information. Okay, can you show me examples? Well, not really, because that's kind of proprietary.

We don't want to show our tokenization strategy. Okay, so I'm paying you based on the unit that you invent, that I have no control over, and you're just going to send me a bill. Yes, that's what we're going to do.

So that's good. Where do I sign? Because that's basically what's happening with the tokens now, that everybody is going to a token-based billing.

Well, of course, the reason is that we don't really have – that's a very easily controllable proxy for just compute. Basically, you're just using up compute time and getting billed on that time. And they've just made a proxy measure of a token in between that they can also control so that they can adjust the pricing and all that.

If it would just be, I don't know, I think NVIDIA and the other low-end guys are now promoting how many tokens per watt their machines are capable of producing. And that's very important for them to measure that. What's the energy efficiency of that?

And, of course, we've heard a lot about that in other contexts. But it's also like, well, that is the basis of the expenses for those tokens, that how much energy do we need? We have some kind of infrastructure expenses and all of that around it.

But mostly, it's just churning out that electricity to do that compute.

[Pinja] (5:27 - 5:31)

And if that is the measurement that we have, that is what we're going to measure. That's as simple as that.

[Henri] (5:31 - 5:32)

Yep, that's what we're going to do.

[Pinja] (5:32 - 5:41)

But we also need to, of course, take into account that in addition to the usual subscription costs that we have for tokens, I think we still need to pay the humans that are using the AI.

[Henri] (5:42 - 6:43)

Yeah, for sure. I like that. Even though everybody's now, okay, we can replace everybody with AI.

We can do this and that. Well, somebody still has to do their job, but this job is still being done for humans. So kind of like, yes, we can make humans more efficient.

We can automate a lot of stuff. And maybe what I feel with AI is that the value of busy work or long slide decks or something like that, shows that we've spent a lot of time on this. That's gone down a lot.

What I've been seeing now is that the very focused one email where I have like three bullet points to a client now carries the kind of same weight and value that previously we might have had to do, like a 50-page deck to say to a big client that we are actually thinking about this. But now because the cost of making that deck is so small, it doesn't really show you that we used it all the time. And that's where you still need the people.

That's where the people's expertise comes from. That's kind of like, do you trust us to use these AI tools in a way that we can help you with your problems? And basically, hey, you can use the tools.

So the value still is the people that you're adding when you're doing any kind of business consulting or selling value. Because the tools are available to everybody.

[Pinja] (6:44 - 7:05)

Yeah. And this is something we discussed in this podcast and with our customers so many times that even if we give, let's take five organizations or five companies and we give them the same set of tools, it is not going to end up in the same result with those same set of tools. So it's still, that is not the value that you're bringing.

That is not where the value is actually being created with just the tool itself.

[Henri] (7:06 - 7:26)

Yeah. Yeah. And if you think about, for example, we do a lot, we train people.

So we still see the people, we are not training AI tools. So the value is still the people and we want those people to have the skills to use tools to make that value. So even looking at, of course, that might make us biased because we want to have the people in the business, but I think that's still what's happening anyway.

[Pinja] (7:26 - 7:42)

We do. So we're talking about token costs, fine, the humans are a cost, but then there's the AI infrastructure and still just running your basic DevOps because we cannot forget that this is still DevOps that we're doing, even though the AI is bringing its own flavor to it.

[Henri] (7:42 - 8:36)

Yeah. And pretty much, I think for DevOps companies, the AI time has been really good because the AI is basically just a new automation tool. And it's a lot better than any of the previous automation tools, but of course it's much more expensive to run.

So if you look at test automation, if you look at any of these, you're going to have the AI do all of those testing. And I think this goes back to what we were talking about earlier, that looking at the costs, it doesn't always make sense to run, I don't know, the government registered Table 5 to run my unit test for my exercise app. I want military grade testing for my exercise app.

That doesn't make any sense. But kind of like building, using AI to build just normal test automation, build those normal test automation cases using AI. Then running the test is cheap because it's just automation, but actually building the tests is going to be effective because that's AI.

And again, kind of like, I think this falls under AI FinOps, and it's not just about AI, it's about the tools that the AI is actually going to use to achieve your goals.

[Pinja] (8:37 - 9:04)

Let's go back to the comparison. So we talked about the Cloud FinOps, and somewhere somebody actually compared this to say that now this is actually AI FinOps is Cloud FinOps 2.0, but there are so many similarities, so many differences at the same time. So we first went from servers to cloud.

That meant that the cloud bills actually exploded. So that meant that we needed FinOps practices to control those. But the difference now is that this is even more uncontrollable, isn't it?

[Henri] (9:04 - 10:06)

Yeah. And kind of like we've seen, of course, humans are very good at throwing comparisons and saying, this is the same as this, and this is the same as this. And of course, we see these cycles in the industry quite a lot.

Now we are switching to something that we don't understand as well. I think that happened with Cloud as well, that immediately in the beginning we were like, hey, we cannot control this. I don't understand where my data is going.

It's just somebody else's machine. What am I paying for? How does this happen?

And then kind of like slowly we started getting tools like infrastructures, code, all of these bits and pieces that really tied it together and actually made it manageable. Well, even still, it's barely, I'd say, manageable on the FinOps side and predictability side, if you go and try to understand what your consumption is in, let's say, AWS. You're still going to get a 10-page PDF of all the routers, all the IP stuff, storage, and all of this.

There's a lot of stuff going on, but it's manageable. But it wasn't in the beginning. It was really unpredictable.

You still hear these cases where, hey, oops, one engineer made a configuration mistake, and we ended up paying 50K for a machine that we accidentally deployed for two days or something like that.

[Pinja] (10:06 - 10:38)

Yeah, that's an oops. But if we think of how predictable these current bills are that come from AI, there are so many rules that are changing. Number one, obviously, is that there is the big change in GitHub Copilot.

Not that long ago, the token pricing changed, and we have some geopolitical reasons, of course, not to go too deep into this, but the US is changing the rules and which models the others around the globe are having the access to. So we need to actually take that into consideration. But the agentic systems make the forecast really difficult.

[Henri] (10:39 - 12:30)

Yeah, and that also comes with the fact that we just don't want the AI to run itself. We have the token thing where we just generate all of those tokens. But if we want the AI to actually deliver value, we want it to have access to tools.

We've already heard of Meta's head of security giving agents access to her inbox, actually, to help her handle emails. And the helpful agent decided, hey, I just deleted all of your old emails because you weren't really reading them. That's where the actual FinOps expense as well comes.

So we still have the same problems with the cloud. Because now if you tie in your very great DevOps AI agent, and you will want to do this. You will want to do this at one point.

I want to have systems which are this smart. And they're now a little bit scary because of all the stuff that they can do and not kind of like, we need more controls, we need more structures for that. But I think that's also a big part of FinOps. When you give AI powers to deploy something to AWS, are those then AI expenses or are they just AWS expenses?

If you spin up a machine, you spin up automation, you do deployments with AI. So again, like the AI FinOps is just spreading to all parts of your company. And it's not just the token pricing, it's about the tooling that the AI is deploying or soonish, probably the tools that the AI will buy for you, for example, or the services that the AI will buy for you.

And I think this explodes also the like predictability of the whole of the AI FinOps in a huge way, because it might change from, hey, I'm running a very small, not that great model 24-7, and I'm getting the same expenses from that. Then I run, well, Fable 5 for like five minutes. And on top of that, that model can do purchases for my company.

So there's multiple levels to this, what can happen and are now unpredictable. And we do not yet have the best tools to contain and control those expenses on what's happening there. But what we are getting those and how do we want to track that is going to be a big thing.

[Pinja] (12:31 - 12:57)

Yeah. And the big difference between AI FinOps and traditional FinOps is now the cost variability. With AI, we, of course, get the GPU scarcity, which is not exactly the same case with cloud, but there is also the agent spikes.

But if we think about it, now we talk about the costs, where is it coming from and how is it different? So how do we manage the AI FinOps? So I think as we've been talking about here, we need to create visibility.

Where do we start creating the visibility?

[Henri] (12:58 - 14:07)

Yeah, that's actually a good point because we don't see a lot of that. Like if you think about even the UIs now, they use chats, GPT, or Claude, the chat windows, but all the processing is happening in the background. You have like that working icon just rolling there and that's your visibility into what's actually now going on.

I like tracking those expenses. You really don't have anything else. And then it gives you like a two sentence answer after that.

But you have no idea. Did it just use, let's say a hundred bucks to generate that answer or like a one cent. But then when you embed it into tools, you take it out of the chat window and use it through an API, then you lose even that small amount of the tracking.

But you really have to build that and you really have to track your teams. You have to build tooling to help you identify who's using those subscriptions because the variability is still there. One, of course, to help you train people to use it and to track where we are using it?

Are we getting the value or are we using it to generate cat pictures and mass or something like that? So getting visibility into which teams are using AI, getting visibility into which models are used because this is something that we've made it. Well, of course, it's partly marketing and of course it's the performance, but we are still in this mindset.

Hey, the best is still the best. So I want only the best. Of course, I want all the best.

[Pinja] (14:08 - 14:13)

Yes, I want my cat pictures to be created by the best model. Why else would I be doing those cat pictures?

[Henri] (14:13 - 14:27)

Yeah, exactly. I want the fur to be perfect. That's what I want.

But you don't need that Einstein level knowledge to build you an Excel from test automation stuff. You can run a lot lighter models to do that. So that's a very, very thing.

[Pinja] (14:28 - 15:05)

No, I don't need the best model to organize my calendar for me. That's not the best use. And to create that visibility is one thing to have, for example, a dashboard to actually make it more visible.

Where is the money going and where is it spent? And with a dashboard, actually tracking the cost per use case, for example, instead of the token. Token is the currency right now, but we need to make that more outcome-based.

For example, is it a support ticket handled? How much are we using on a support ticket being handled? For example, a user story being created, a code review, or let's say even a sales proposal if we're going towards the business users of AI.

[Henri] (15:06 - 16:03)

Yeah, and this is just, when you think about it, this is just normal operations, but just for AI, how much effort did we spend on building this proposal for a client? Are we already in the negatives because we spent like 10 of our best engineers to build an offer to a 10K case or something like that? I'm like, this is just normal.

This is what we do every day. I'm not going to bother the CTO and take his valuable time to build this 10K offer, for example, but we are going to do it with the team of juniors, or we're going to do it with some other people. And we are already like, even though we are always saying that we want the best model, but we are actually even doing the same kind of thinking in many cases.

And I think that's also like, it's easy to talk about dashboards, getting visibility, but how do we actually get the data there is, of course, we have to understand the underlying systems and instrument those to get those tracking out of it. And that means also, I think the most difficult thing about AI has been for many companies that this requires you to understand your business fully, like an instrument for your whole business. And that's been an eye opener for many people as well.

[Pinja] (16:04 - 16:41)

And we still see so many organizations where they're trying to raise their AI maturity. They just see that, hey, our competitors are doing this. We need to go to the same level as well.

They're thinking that the ROI is going to be the same, but obviously they need to do the ground work first. They need to get their documentation in place. They need to get their DevOps practice, just very simple DevOps practices in place.

We still see so many organizations who are not doing testing early enough. They don't automate their tests and then wishing that AI will fix this and then getting, like building a dashboard, fine, but actually digging down to where the problem is. Why is our spending like what it is right now?

[Henri] (16:41 - 18:18)

Yeah, exactly. This is the kind of thing that, now you are going to the territory of my PhD. So I love this territory anyway, but because you really have to now think on the business side of building software.

How do you validate what you're actually doing? And AI actually makes it super easy to do all the best practices of software building now. So it's changing the whole software development thing and how you do DevOps, how you do development, how you do everything.

Because now you have the option of making everything gold-plated on the process side of it. You can have great testing because making tests is easy with AI. You can have good code quality because building good quality code with AI is if you have a lot of other tools in place is actually quite easy.

Doing code review, doing all of the steps that previously I've been, I don't really have time and I couldn't really be bothered to do the testing and do all of this. Now it's super easy to set up agents, automations and all of that to really get your DevOps pipelines in. And again, like being the shit here that I am, it is exactly that the AI is pretty much a Trojan horse for all of the DevOps and automation and all that.

And it's the FinOps as well. Because if you have a standardized process, you don't want the expensive AI agent to run that. You want to bake that into just normal plain old automation, which is very efficient, very cheap to run.

And like making these decisions on where to use AI. And you cannot really make those decisions before you understand many parts of your business on what makes sense, what are the different costs everywhere. And I think this will lead to a lot of optimizations on every single business.

I think this is what's made this time great also on operating the AI. And of course, then you have a lot of these external factors that might be happening. Some big change happens.

Hey, this model is now closed for anybody but the US and we have no idea what those repercussions are.

[Pinja] (18:19 - 18:32)

No, this is going to be a very interesting thing to see. Have you seen a lot of these tools emerge right now that would actually provide the dashboard for organizations to track these costs and the models being used and what is the cost per use case, for example?

[Henri] (18:33 - 20:05)

Yeah, we are now getting a lot of these kinds of toolings we built. We are building them from open source software, building them from all the pieces as well, piecing it together, and getting that enterprise level tracking for a lot of businesses because we need it. Like tracking your expenses and where are you actually investing in AI? And actually this is not a segue but when we talk about AI FinOps, you talk a lot about the cost per token pricing, but there's actually another way also to think about it is buy your own hardware, run your own models.

And then you are not talking about cost per token anymore. You're talking about, hey, let's invest 100K in an NVIDIA machine where we can run these big models by ourselves. You have that one-time investment of, let's say, 150K total of getting software and getting the hardware running.

And then you basically just depreciate that token engine over time. So now you are basically swapping the equation the other way around if you are running locally that you want to get the maximum usage out of that machine that you invested into. Then it becomes kind of like an optimization question of how much token spend do I need to have before it makes sense to buy my own machines.

And then if we have gaps or we need a very spiky load because this is AI again, is that the load of using tokens is typically very spiky. When people use them, what happens is that you might have huge spikes in token usage or AI usage and then you might have quiet times. But if we can spread that out again or use your own hardware, you can actually bring the cost down in a large way.

And also it transfers that investment from OPEX to CAPEX side of the equation. So that might also be a really critical thing for your finance side of it as to running it.

[Pinja] (20:05 - 20:49)

Yeah. And I wonder, there is somebody who needs to be in an organization, I guess, to track those costs. So it can be, for example, similar to where FinOps is actually being tracked.

So this is now an additional thing to add. You mentioned the locally run models and we already talked about which model to use for cat pictures to get the best quality for fur. And model governance is one thing that has been obviously in discussions lately and model routing and actually having some kind of locally run, I want to say a block, or basically doing an API call considering what is the best model to use out of the ones that we're paying for for this individual call and task that you're asking for.

[Henri] (20:49 - 21:46)

Yeah, exactly. That model routing is something that I think many of the organizations have to build by themselves to get that visibility because one thing that we haven't talked, we kind of like the optimal situations, of course, that you're buying your AI from one source, you're buying it from AWS, Google, Azure, ChatGPT, but that's not the true case. Typically what we see in organs, you have at least like six, seven, eight different subscriptions for different packages that we have and from different providers that you're doing.

So the truth isn't that we have just one operator. So you need, and now if we bring in local models into the mix and even bring out technologies getting better, you can run some of the smaller models locally on your own machines. How do you track that?

How do you kind of say that, hey, I'm running this model locally. I'm saving the company this and this much token expenditure by running it again on my already purchased hardware. You need those routers.

You need those like locally installed software pieces and tracking portals or model routing to do that for you so that you can get that data.

[Pinja] (21:47 - 22:11)

Yeah. Now, of course, coming with the costs, we need somebody in the organization who's tracking the budget and setting the guardrails so that there is somebody monitoring the spend, but also this is one of the hardest things, actually, I would say, is the measuring the value because it's easier to say that we've used this much, this many tokens, but what is the actual value being created? What was actually doing the analysis?

What was eliminated by doing this?

[Henri] (22:11 - 22:56)

Yeah. And how do you assign this to, I don't know, a ticket or how do you track it into actually some deliverable that we were speaking earlier about as well as having the, I don't know, support ticket. Hey, we can see that we've now spent five euros of agentic spend on this ticket or for some reason we haven't used 500 on this.

What's going on in this? And for like, we can then optimize the whole flow that we have around understanding that and where can we track it? Because that's what we need.

We need to have that value-based tracking or like for this commit or for this thing or for this team. And you really need those kinds of pieces of software because it's going to be from multiple sources and not just trusting your one dashboard from Anthropic, for example.

[Pinja] (22:57 - 23:31)

This is where many tooling companies are now building their own software in a sense where they built the MCPs so that they can actually collect this information from multiple sources. So it's not just one piece of information for your ticket. It's not just one piece of information for your PR.

So everything should be in one dashboard because that's where we would actually see the bigger whole picture of what is actually, what is being spent and where did we actually win money back? For example, how many tickets did we use AI on? And what was the throughput time for those?

[Henri] (23:32 - 24:35)

Yeah, and of course it always brings these comparisons of, hey, using a service agent or service agent person, actually a human person to do this would have cost us this much if you can solve a lot of these easy cases with AI, it costs us this much. But that's just the way that this AI business will be run. You are looking at those and seeing where you can get the value out of AI.

And then you have all the tasks where you can clearly see that ideation, R&D and all of these, it's not just about replacing those costs with AI costs, but about augmenting it a lot. And I think there's like two tracks to this that will happen simultaneously in your company. You will have the optimization track where you do exactly this, that, hey, we can eliminate like 70% of the easy customer service agent calls with AI and put that expense to be really small.

But then you will also, okay, on the other side, we want to invent new business. We want to invent new ways of working. We want to invent new products, which will then be kind of like how we use AI to augment the work that we're already doing on that one side as well.

And these will probably be a little bit of a different kind of thing to scope and price out in the same way as, I don't know, any company R&D and other will be on, how do you do pricing?

[Pinja] (24:36 - 25:02)

And if we can now wrap this up by thinking about what might happen next. This is of course just us doing some predictions, but since the costs are very unpredictable right now, we know that the geopolitical situation might cause us some potential new restrictions. Somebody I talked to compared this to electricity pricing.

So could we end up in a situation where the pricing is more dynamic based on demand or peak hours, for example?

[Henri] (25:02 - 26:10)

Yeah, I think for sure that's going to happen because as I said, the whole of AI token spend is pretty much directly tied into electricity costs. And I think that's now the number one driver for new investments in nuclear and a lot of these... And even somebody, I just saw yesterday that they are now opening up a new fund to fund fusion energy research finally because AI is using so much power.

So might it be that fusion will not be 30 years away because of AI? We'll see. But like a lot of this kind of stuff is now happening.

So it can be that the next step of token pricing will be dynamic and that's already happening. You have like the two-tier token payment system for Anthropic that you have the on-demand tokens now, or you have the 24-hour, or we will run this at the slowest time within the 24-hours pool. So we will see the same kind of resource optimization happen.

And like how we've built those optimization systems, for example, is that we have built systems where you run the nightly builds, code reviews, all of that on the cheaper token pool because you don't need those results immediately. And that already optimizes your costs. But then, of course, doing the development and keeping that loop as fast as we can, then we are using on-demand.

Like splitting that stuff in there is something that I see as being very worthwhile.

[Pinja] (26:10 - 26:17)

And one prediction is that we might see some reserved capacity for critical operations as well, in the same way as electricity.

[Henri] (26:17 - 27:07)

Same way, yeah. Somebody's reserving three cores to already have some baseline and then adopt that by buying from the market. Again, like, of course, I'll probably get a better price for reserving that capacity for a long time because I'm fully paying for that, even though I'm probably not going to use it 100%, but I have that reserve.

And it's going to be the same kind of discussions about availability and all of that as well on what are the company's critical processes and what do we actually need to run? What has to be run? What is nice to have?

And how do we do the FinOps optimization and all of this as well? And then also, like, one thing that's probably going to be also interesting is where do we run it? What kind of risks are we willing to accept?

Is it now cheaper to run this in the US? Is it okay for us to run this in the US? Is it now cheaper to run this in a Chinese data center?

Is it now cheaper to run in Europe? Are we doing that optimization? Are we jumping from provider to provider?

There's so many optimization layers to this game as well that we can do as well.

[Pinja] (27:08 - 27:26)

But I think it's fair to say that this is something that organizations need to start paying more attention to to actually not end up in a situation where many, many organizations did with cloud in the beginning where the cloud bill just spiked up. But Henri, thank you so much for joining me today. I think this is all the time we have for today, but it's been very insightful and fun to talk to you about this.

[Henri] (27:26 - 27:34)

It is always, always fun. It's always fun to talk about all of these different layers and what's AI going to change. So thank you for having me and listening to a rant about money and AI.

[Pinja] (27:34 - 27:45)

That's always fun. And thank you everybody for tuning in and we'll see you in the sauna next time. We'll now tell you a little bit about who we are.

[Henri] (27:46 - 28:04)

Hi, I'm Henri Terho. I'm the Principal AI consultant for Eficode. I've been building AI since high school and whatever.

I've been involved in a lot of projects from AMD to Silo and now at Eficode is building the engineering future of it, how to integrate AI into whatever you do daily, not just the science side of it.

[Pinja] (28:04 - 28:11)

I'm Pinja Kujala. I specialize in agile and portfolio management topics at Eficode. Thanks for tuning in.

We'll catch you next time.