Someone Racked Up $2,000 on AI in 24 Hours. Here's What They Did Wrong.

Pops and Steele dig into a problem hiding in plain sight: uncontrolled AI subscription spend. As teams experiment freely with AI tools — often on personal corporate cards with zero IT or finance visibility — costs are quietly compounding into what could become a governance crisis. The two draw parallels to the early days of cloud computing and Shadow IT, arguing that AI subscriptions, tokens, and usage-based billing need to be tracked like any other IT asset, ideally landing in a CMDB. They walk through real-world cautionary tales (a $2K/day AI bill, a Meta token-spend anecdote), debate who actually owns AI risk within an organization, and lay out a practical cadence for reviewing AI spend — monthly until you understand it, then scaling back. The episode closes with concrete advice: read the fine print on your AI tool's usage limits, bring a real cost forecast to finance early, and don't wait for the invoice to start the conversation.
Key Takeaways

Shadow AI hides wherever there's no intake and governance process — not in one department, but across every team running its own point solutions.

Usage-based billing breaks traditional software asset tracking. Unlike flat licensing, token/consumption-based spend is jagged and hard to forecast without active monitoring.

Treat AI subscriptions like governed IT assets — track what models, datasets, and prompts are in use, ideally inside a CMDB, the same way you'd track any other asset with blast-radius risk.

Review cadence should match maturity, not comfort: monthly (or even daily/weekly for new capabilities) until the org actually understands its usage pattern — then it can stretch to quarterly.

Ownership of AI risk is shared, but accountability isn't. The team that brings a tool in without going through proper process still owns the consequences.

Bring a number to finance before they ask for one. Proactive cost forecasting protects the relationship — and the budget.

"Ferrari to the grocery store" problem: using frontier/premium models for simple tasks is where a lot of runaway spend comes from — match the model to the job.

AI spend management, Shadow AI, Shadow IT, AI asset management, CMDB, IT asset management, ITAM, AI governance, token-based billing, usage-based billing, AI budget, finance and IT alignment, AI subscription tracking, consumption-based licensing, AI cost governance, enterprise AI adoption, CAB governance, AI risk management
Suggested CTAs

Is AI spend already a line item at your shop — or are you still finding out about it from the invoice? Drop a comment and let us know.

If you're wrestling with AI governance at your org, hit subscribe — we're covering this space every week.

Tag someone in IT or Finance who needs to hear this before the next invoice lands.

Transcript

SPEAKER_02 0:00

So there's a statement when we did research for this topic, and I'm gonna read it off the sheet because I just don't bel I I can't believe this is really happening a lot. I think it might be happening in small shops, maybe where it's not a lot of governance, but I can't see me doing this. I mean, my my boss would be all over me if I did this. But here's a statement AI spend is the new shadow IT. Individual employees and teams subscribing on corporate cards or expense reports with zero visibility until the invoice lands. I can't see that happening in my my workplace.

SPEAKER_00 0:35

I can absolutely see that happening. I believe it. I promise you that is. Really? Yeah, I have no doubt on that.

SPEAKER_02 1:35

Somewhere in your company right now, three different teams are paying for three different AI tools on three different corporate cards, and nobody in IT or finance knows any of them exists. That's not a hypothetical. That's just a regular Tuesday. So let's pick up a wrench, pop open a cold one, and let's open and dive in.

SPEAKER_00 1:52

Or two, since it's Tuesday. That's right. The business wants speed, IT wants solutions to meet the desire for that speed, and finance wants control over the spending. And whoever ignores the budget conversation loses the seat at the table. Today we're talking about why your teams need to start treating AI subscription like assets before finance does it for you with a blunt instrument.

SPEAKER_02 2:15

Yeah. So what does shadow AI actually look like inside an organization right now? Where do you think it's hiding?

SPEAKER_00 2:22

I think it's hiding anywhere you have an AI client and a human being. I think the theme that I've seen and the space that we're in today and how we got here is there was a lot of experimentation, a lot of encouragement. I don't know if you remember the term token maxing. That is something that is slowly going away. And in building out these solutions and experimenting, all these ideas came about on how we can create speed and how we can be faster in delivery, all the new insights that we can gather from the data that we was previously siloed. And folks are building a lot of point solutions that are solving a very specific problem, sometimes the same problem, in their own way across whatever AI it is that they choose. So when you ask me where is it hiding, I would say everywhere you don't have an intake and governance process for it. Why is AI spend harder to track than traditional software spend?

SPEAKER_02 3:23

I think you can track, I don't know, the generic invoice, annual invoice, monthly invoice, quarterly invoice, wherever that comes. You may be able to track the assist tokens, whatever that is, and maybe even towards the account of who's using it, but how they're applying it, where to differentiate where Brian and Steele might be working on the same project and not know it, and you're actually losing, you know, money, money, and spend, right?

SPEAKER_01 3:53

Yeah.

SPEAKER_02 3:53

They have projects, but what I term a project and what you term project could still be loosely, and my projects don't link with your projects, at least from what I understand, what I see.

SPEAKER_00 4:03

I think there is an arrears tracking in that you probably have a pack of tokens or a pack of AI usage, and you type in a prompt and you walk away. You may have Deep Think on, you may have the most expensive frontier model on, or you're testing out the new one. And some of this licensing is turning into consumption base, or what you are building with it is more complex. You're no longer asking, hey, what's today? What's two plus two? You're putting in design specs and you're one-shotting an application, and that is consumption-based. And I think the reason why it's so difficult to track is because the lack of intake governance, traceability, the amount of usage is jagged. And if you have somebody who has auto approval on, or I think the best example I've seen is there are people that have let AI with OpenClaw completely control their computers. Well, the entire time it is running, it is generating spend. And so we have the technology, we're using AI, but how are we actually tracking that spend? And I don't know that we have a good answer for that.

SPEAKER_02 5:19

And I don't think the things are moving so fast and there's so many unknowns that we're not building a we're not building the fortress we have back in, like we do, like we end up doing with cloud eventually. You know, we've done with cloud now or hardest construction, right? Yeah. It's like we're venturing into the the Wild West. We're building log cabins, you know, we don't know what kind of elements, warfare, you know, always different things. I don't know how I'm gonna feed my family the next day. You know, you're kind of just going day to day, trying to figure it out and not die from it. And I think that's sort of where we're at. But the advantage is you have freedoms, you you have a whole wild west of things, you have new new experiences. You know, you're not limited by governance from the old way old world. So I think there's things there that is that's enticing you, but I think there's still a lot of things to figure out. And I think that's the part that's hard to track, you know, and and just by figure out where you gain.

SPEAKER_00 6:20

You can build a log cabin now with a keyboard. Maybe digitally. Just digitally. Not not yet physically.

SPEAKER_02 6:27

At least you're a 3D printer, maybe. Yeah. Has finance actually started pushing back yet, or is this still building towards a breaking point?

SPEAKER_00 6:36

I have not heard I've I've heard of some anecdotal evidence where some of that free interest period rate expense has increased and where folks have gotten some shock bills. I've heard of one person racking up two grand plus with uh, I think it was Fable Five in 24 hours. And I I think that's using the wrong model for the job. Uh was their finding.

SPEAKER_02 7:02

Those are huge outliers. These stories you hear about are things that I bet you there's other stories with that person that are like that.

SPEAKER_00 7:10

Yeah. And I think what that shows is you took a Ferrari to drive to the grocery store. You didn't take it to the track, you didn't use it for its intended purpose. Not an apples to apples analogy, but essentially it. What the takeaway was is that there was other models that were not frontier models, just as capable at planning and providing some of that result where they should have just used the engine from the Ferrari, they didn't need everything else.

SPEAKER_02 7:39

And he also took the Ferrari and drove it as if it was a pickup truck or a or a Volkswagen or used it like Google or GitHub. It isn't. It isn't the same thing. You know what I'm saying? They they they didn't do it with guidance, they didn't do it with caution, they didn't do it with education, right? Or they wouldn't have done what they did, they wouldn't have had the bill they had.

SPEAKER_00 7:59

Yeah, it it was the new thing. Let's test it out. Turns out the new thing is also really expensive.

SPEAKER_02 8:04

Yeah, yeah. Don't we know that from Azure, from you know, AWS? Didn't we learn that lesson then?

SPEAKER_00 8:10

Yeah, and it's it seems like those past lessons when cloud came onto the scene also seem to apply here. And I think there's lessons that we should carry forward in this space.

SPEAKER_02 8:21

So there's a statement when we did research for this topic, and I'm gonna read it off the sheet because I just don't believe I I can't believe this is really happening a lot. I think it might be happening in small shops, maybe where it's not a lot of governance, but I can't see me doing this. I mean, my boss would be all over me if I did this. But here's a statement AI spend is the new shadow IT. Individual employees and teams subscribing on corporate cards or expense reports with zero visibility until the invoice lands. I can't see that happening in my my workplace.

SPEAKER_00 8:57

I I can absolutely see that happening. I believe it. I promise you that is. Really? Yeah, I have no doubt on that. That's not just I mean, if you have given the authority to someone who owns a corporate card and is at a certain level, and they can swipe it, even if go back to the 99 cent app examples, you have budget, you put it on a card. That goes back to governance. That goes back to looking at well, it's it's governance.

SPEAKER_02 9:27

It's it's when you get the card, you get taught how to use the card, what could be on the card, what we're gonna re what we're gonna repay on the card, and what you're responsible for on the card.

SPEAKER_00 9:35

I've seen SaaS applications paid for on a corporate card.

SPEAKER_02 9:38

That's what I had more confidence in in our world. So okay. We we uh we like our projects, and I now know more.

SPEAKER_00 9:45

We like to learn the same lessons over and over.

SPEAKER_02 9:48

Okay. All right. So what would it take to treat an AI subscription uh on an API key like a governed IT asset? What's the minimal viable process?

SPEAKER_00 10:00

So you use a key word there, yeah, a couple of keywords, but the one that stood out for me was besides the word key? Besides the word key. I I'm lacking on my puns today. And that is the term asset. And Taco Bell did not pay for this.

SPEAKER_02 10:16

Or we'd like Taco Bell to sponsor this, but Taco Bell did not sponsor this episode.

SPEAKER_01 10:22

Or mighty taco up in New York.

SPEAKER_03 10:25

A dirty bottle, dirty bottle.

SPEAKER_01 10:27

Yeah, delicious though.

SPEAKER_00 10:29

So being able to just like CloudSpend, SaaSpend, tracking your subscriptions in AI, tracking the utilization. In order to track it, it's we consider this an asset. It's something that we need to manage the financial values of. And there are platforms out there that are capable of tracking it. And the place where I'm seeing this pop up first is in a CMDB. I say first is probably manually in a spreadsheet first, and then eventually makes its way into a CMDB configuration management database. And then you are able to look at what models are being used, what data sets are being used, what skills, what prompts, was it connected to? Discovering that, although it's that technology is still early. That is the direction that it is heading because of the blast radius being so large with AI and so quick. We talked a couple episodes ago about someone losing their entire database in nine seconds and the backup. Being able to track the blast radius, being able to track what has access to what, where, and being able to track the spend is incredibly important to the success of adoption and sustainability and keeping this going forward. And so you have to treat it like an asset. You have to treat it as something that you manage as part of solutions that you deliver. If you use it to deliver a solution to your customer, and then you have to remove that feature because of unmanaged spend, you're gonna have an upset customer. So I think MVP is knowing what you own, how how much you spend and and what so I agree with this.

SPEAKER_02 12:08

So before you lead to your question, I'm gonna try to, I'll probably end up pitching you the softball in this one, but my question is on my question, does it what's the difference between I know the difference between the software itself is different or the application itself is different, but what's the difference between that and Salesforce service now, Microsoft 365 and and and some AI subscription? Aren't they all the same? They're cloud, their subscriptions are user-based. Sometimes it's use-based within the the service.

SPEAKER_00 12:38

Yeah, there's by and large, it it's it's Dalton tokens.

SPEAKER_02 12:42

Now you might you run that sometimes with a with good with ServiceNow, you run into that without AI even.

SPEAKER_00 12:49

So it in s uh and hopefully I'm understanding the question here. In things like ServiceNow, I think it's called Assys. So it's not tokens, it's assist, and you buy that in packs. In other platforms, if you're using Cloud, if you're using uh copilots, etc., it's based off your your usage.

SPEAKER_02 13:06

But it's all kind of same thing, different terms.

SPEAKER_00 13:09

Yeah, same thing, different terms. That's accurate. And underneath it is tokens, that's the the thing that is being consumed, which in order to produce tokens, you need power, you need the compute, you need memory, everything that the data centers are providing. And anyways, all the vendors are essentially selling you that consumption of tokens at compute, which may be at a token level, it may be at a per month level, and then there's different tiers of that. And I think that changes constantly. So even what we're saying today or the landscape that we're in now could look very different in three months, six months, could completely change.

SPEAKER_02 13:50

And these would include things like Zap Year and Make and stuff like that too, right? Yes. Because we tie, we use, we use a utility of it, we tie into it, we still need to treat it as some type of asset.

SPEAKER_00 14:04

Yes. And I would say the exception to this would be anything that you do locally. However, that really struggles to scale if you don't have your own data center. Yeah. So how does usage-based billing, so tokens, API calls, et cetera, break that traditional software asset management playbook?

SPEAKER_02 14:22

So there used to be like Lexis Nexus in legal. Back in the day, you would have users and subscriptions, but you also had to pay per printed page on the Lexus Nexus printer that sat in your physical office. I know it's 1980s, 1990s when this happened. And we probably weren't speaking, but that idea was there. And so you had to watch what you printed. When no when Windows printing was a very big thing back then because Windows printing kind of circumvented that Lexus printer. But so this sort of thing, I think it's just another level of I don't know about asset, but costing, right? Budgeting.

SPEAKER_00 15:03

It's not talking on the phone. Talking on the phone and paying for your minutes are what I think of.

SPEAKER_02 15:08

Yeah, it's it's a lot like that. Yeah.

SPEAKER_00 15:10

Yeah. It's nothing new, it's just in a new area. Yes. And what it is providing and the value it is providing is different.

SPEAKER_02 15:18

It can be buildable. Well, soaking minutes, soaking minutes, long distance with in-law firms were build to the client.

SPEAKER_00 15:24

Yeah. Yeah. So and so I think it it doesn't really break the traditional playbook. I think there's answers for it that exist. I think it's that we have to develop the muscles for tracking it, and maybe the tooling looks a little different.

SPEAKER_02 15:41

It's showing up in places that we didn't have before. It's sweet potatoes, it's growing up inside your green beans.

SPEAKER_00 15:46

Uh we're gonna we're just gonna leave that one on the floor.

SPEAKER_02 15:49

How to bring that Indiana heritage out. Okay. So where does ownership sit for an AI tool, a business unit adopted on its own? IT, finance, or the business unit?

SPEAKER_00 15:59

On the CEO. It's on everybody. It's everybody's responsibility. If you think about the cost, you think about risk, it is introducing risk into the footprint of the business.

SPEAKER_02 16:11

Everyone owns it. You're right.

SPEAKER_00 16:13

Yeah, I I everybody has to own that risk. It is essentially at the end of the day, the business owns the risk. The teams that brought that into the footprint have responsibility to follow the governance, the policies, the standards to manage this thing. It is shared risk.

SPEAKER_02 16:31

Yeah, there's a two-minute hockey penalty for that team that brought it on without going through the right process.

SPEAKER_00 16:38

Yes. And I think being able to catch that and correct is a good capability to have.

SPEAKER_02 16:43

Depending on the degree of what they brought on and the history of how many times they've done this, it could be job changing, my opinion.

SPEAKER_00 16:51

That's fair, depending on how many databases it deletes in nine seconds. What's the right review cadence to that point? So let's say we've brought stuff on, we've reviewed it and we've approved it. What would be a cadence in reviewing that? And I think let's keep this simple. So let's just focus on the cost element of it. So let's say we brought AI in, we're using it, and now it's is it's time for the Piper to get its due. What is the cadence for that look?

SPEAKER_02 17:19

It's monthly until you understand it. And only when you understand it can you bring it out to quarterly, biannual, or anything different than that. But you got to look at every invoice because you don't know what scenario has unearthed itself in this invoice, right? So you have to look at every invoice with your vendor in you and the user who's doing it to understand what's at play, how did it happen, where do we put it, were we supposed to get build for that? Is it in the contract? Just to understand all those elements. So one, when I go to renewal, I also know where do I start searching for cheaper token unit pricing? You know, you need to understand every parts of that. I think it's monthly until you understand it, and then you can discuss breaking it further out.

SPEAKER_00 18:02

I'm gonna ask you a follow-up here. What about when you roll out a new capability?

SPEAKER_02 18:06

What do you mean?

SPEAKER_00 18:07

So let's say you roll out knowledge knowledge article generation. You introduce an AI search, knowledge article generation, uh, a build agent that can develop things for you. How uh how often should you check when it when it's new functionality?

SPEAKER_02 18:24

Yeah, because that can happen moment to moment. I mean, some of that stuff you can go almost really weird and go daily and weekly for a while. It's all about understanding it. If you once it's not no different than an asset. If once you understand your asset, you can get a little bit further away from it. But until you understand your asset, you have to you have to stay on it, you have to get you have to nestle up to it.

SPEAKER_00 18:46

So you have to understand your asset before you can cover your asset. That's right.

SPEAKER_02 18:50

Yeah. That's true. Yeah. Because you got an answer to it, right? Someone's got someone's gonna be coming to you and saying why. And you have to be able to explain why. And if you can't, you didn't do your job. Yeah. So I wonder these other little oh, yeah, another question. So go ahead. That's me in it. How do you build a training tracking process? Okay, this is where we work. It doesn't feel like it's bureaucracy to teams trying to move fast. I don't care. We have to understand it.

SPEAKER_00 19:15

Yeah, I it's governance, unfortunately. And governance, you know, I I'll I'll rephrase. I'll rephrase your honor. I think it is governance so that you can smooth out the process. Slow is smooth, and smooth is fast. And that growth and maturity allows you to move fast and prevent the bad things from happening.

SPEAKER_02 19:37

Well, at the servers now, rep comes to you and says, Hey, I'll give you six months free in that area to let you develop it, get used to it. Are you still gonna watch it as tight as if you're being billed for it?

SPEAKER_00 19:51

Daily, sometimes weekly. Yeah. That that's your chance.

SPEAKER_02 19:55

And at some point you gotta still understand it. Because at some point you're gonna be paying for it. Why then people get used to it, depend on it? If it becomes the the 12 million, what it was a 12 million dollar squirrel, whatever that little statement thing we read of.

SPEAKER_00 20:09

Yeah. Well, I think I heard just recently that Meta spent six used 60 trillion. 60 trillion. I think it was 60 trillion tokens at the tune of 900 million dollars. I just couldn't remember if it was in uh a month or a year. I that I can't remember. It must have been a year. There there's there's no way they did that in a month, hopefully. But we'll we'll we'll have to look that one up. But that's that you you gotta understand what you're using.

SPEAKER_02 20:36

All right. The tune-up. We're in a tune-up section. Beyond cost, what data and security risks come with untracked AI tool sprawl?

SPEAKER_00 20:45

All of them. Everything that keeps your CISO up at night comes with them. I think uh to use an example I saw recently, that somebody rolled their own MCP server. Custom, they used AI to build it, which is great. They got really fast value. It used about six to nine third-party libraries to do that, a whole bunch of Python code. And when I looked at those third-party libraries, they weren't up to date and they were vulnerable and put them in a software tool which shows the vulnerabilities at the different versions. And I mean, it was a significant amount of versions back. And uh, that's so easy to do. If you've been given, hey, this is an easy way to create an MCP server, and you can have your AI do it and generate it, then you're often working. Why would you ever think to go back and double check the what is working? And so it's that type of devil in the detail things that you you have to lean on your good process, on your security practices to be able to find that stuff. There are processes built to find it, but if you stop following your processes, that's gonna come with some sleepless nights. What's the first conversation IT should have with finance to get ahead of this and spend? Conversation instead of reacting to it.

SPEAKER_02 22:02

All right. So my first thing is I I want to know how much do I think I'm going to spend in this tool annually. You're right. When I talk about, okay, I'm going to bring this on. I need to find out for my vendor about how much we're going to spend. I need to find that out before we can do anything about it. I need to go to finance and say, okay, I got a new tool. I think it's going to spend, we're going to spend about X amount this year with probably a 5, 10, 15%, 20% increase every year. What is your appetite for this? Where are you on this? And I just want to know kind of where they're feeling. I don't want them to give me a number. I want to give them a number, but I also want to see their reaction to that number and where it fits. And I just I kind of play along with them to some extent. If they say we don't have room for this, I at least the financial people I've I've worked with, they would say, you need to get X to sign off on this. We we don't have any approvals to move that high up or move that far or to commit to this, go to this person. I've always had a decent relationship with finance because I've always been honest with them and try to do my best for them. And I think that's that's what I would do. I'd come up with a number and I try to stand behind it the best I can.

SPEAKER_00 23:11

Is there anything that you would bring collateral-wise to that conversation?

SPEAKER_02 23:16

I would bring with a diagram what we're trying to achieve. I would bring like the flow. This is what we're going to gain from it. I'm not going to promise offset of costs that they can take money to shift and pay Peter to PayPal, because that could never work usually out the way it's drawn out. Some of that does work out, but it's usually in different ways, different amounts than usual. And if anything, that's just icing on my cake, right? That if I get that, then that's great. I I delivered more. It's like I'm trying to set the bar at my worst, at my modest in a sense. That way I can do better. I just don't want to oversell. And then they say, Yeah, but you said this. And they're not going to believe me the next time I come knocking on a door. So that's at least my first conversation with them is here's the product we're looking for, here's what I'm trying to establish. And we do that. We have these little cards that it gives the pros, the cons, the risks, the timing, the vendors, the operational expense, the capital expense, the annual recurring expense. It's all on a one sheeter that we send to finance. So I think that's my first conversation. Is there a version of this that goes too far where governance kills a legitimate experimentation of this period?

SPEAKER_00 24:24

Yeah, I I think there is the absolute no AI policy whatsoever that kills it.

SPEAKER_02 24:30

I think there's yeah, go ahead. Sorry. Sorry.

SPEAKER_00 24:32

Yeah, yeah, yeah. And I'll obviously add on to this. We're in a world today where there's legitimate use cases where I excels. It is a tool that we have not had before that does certain jobs better than pre-existing tools. To not experiment, to not try it out is a disservice to yourself, your clients, and your customers that you are trying to serve. The cat is out of the bag in this space. The toothpaste is not going back in. And there are folks that are reaping just an insurmountable, insurmountable amount of value for specific use cases. So I think if you don't use AI at all, that would be going too far and you're doing yourself a disservice and your customers.

SPEAKER_02 25:15

So let me draw a line. Would you say use an AI in replacement of Google using AI? Google search.

SPEAKER_00 25:22

If you're if your use case is that you don't want to have provide your search queries to Google and you want to send it all to the specific AI, or maybe you have an agreement that they can't train on your data, whereas Google does, it would depend on the use case. That was I'm just throwing out an example use case. I don't know that I would use it, replace an external vendor search with it. If you have internal knowledge, internal tribal knowledge that is important to the company and this is your intelligence product. You shouldn't use Google either. I I think then that builds up a use case for using search in that context. Still, if you're if you're doing something in the public space, just be wary of what you're giving your data to, where your queries are going. And it it just depends on the use case. The thing I would just uh close out and then I'll pass over to you. Um if you are replacing something that does not cost with something that does cost, is it providing the value that you are paying?

SPEAKER_02 26:27

Yeah. Okay. I I get my approach is I I'm like you, I'm in that. I'm also I would try the low-hanging fruit of AI before I go splinter the back end of my business bind, trying to correct something that in an agentic AI or something very deep, you know, I would try out getting used to prompting the mechanics of it, how it's built, how it acts, what it does in a sense, before I would take bet my business on it. I think that's fair.

SPEAKER_00 27:00

If a listener wants to start tracking AI spends at their org next week, what's the first concrete step?

SPEAKER_02 27:09

I think it's a weird question because you know, there's my business has contracts with like ChatGPT and some other entities, and they can see different types of billing than what I can see as a cloud pro individual user that I pay my own money for. Um, so it's a little bit different there, maybe. I just think it at whatever level. I think if you're an individual user, you just gotta understand what you're being built for. Just like this Fables thing, this Fables 5 that was just been introduced for the next couple of weeks. I'm gonna have to use it. But it uses 50% more of my usage. It sure does. And you know what I mean? Understand that. Read read that fine print and understand what it's gonna do to you before you start having to pay out and beyond the pro $20 a month. So I think reading and understanding, you can't just do what we did with Google and everything else and just let the the license agreement just be check it and move on. You need to understand this stuff.

SPEAKER_00 28:05

I know that they have metrics underneath the utilization of all of these models. And it's at their prerogative to set the price of each one. And I don't think it's an accident that their latest model is double the price of their last frontier amazing model. I think there is a little bit of pricing strategy that is going on, and they're gonna find out where that optimum band is, and that's probably will where they will stick. That that's but I I think the advice that you had, which is use it personally and look at the spend, is phenomenal. I think that will show. I mean, I'm very much having to use it to see it type of person. That's how I grasp concepts, and that is what helped me understand AI the best. The limitations, but also the expense uh was using it personally.

SPEAKER_02 28:57

Yeah, so there's also the how they play with you. Like I'm looking at my Claude and I can see my current session. I got limits on my current session. I've also got current on my I got limits on my weekly, I got fable weekly limits, and then I got a little checkbox that says turn on usage credits to keep using Claude if you hit your limit. In other words, you want to pay additional for it, right? There's they're but they're they're playing fuzzy with you. Why is there a daily usage? And then there's also weekly usage. It's almost like, and it says in there, sometimes you may hit your daily once, only once in 30 days. They'll still let you use it because your weekly has barely been tapped. But if you've tapped your weekly and you've hit your daily, they'll so they're they're playing fuzzy math with you a little bit.

SPEAKER_01 29:47

If they are tracking it, shouldn't you?

SPEAKER_03 29:53

Yeah, yeah, yeah. I agree.

SPEAKER_02 29:55

All right. Visib see visibility restriction is a strategy. Orgs that build lightweight intake tracking models before the spin spike goodwill with the business. So orgs that only react after the invoice keep neither goodwill nor control. So I think that's a good statement.

SPEAKER_00 30:13

Yeah, whoever ignores the budget conversation loses that scene at the table. I think that's it. That's the episode. Don't ignore that utilization. So drop a comment. Is AI spend already a line item at your shop? Or are you still finding out about it from the invoice? Specific beats generic. Let us know what you think.

SPEAKER_02 30:32

Yeah, this has been the Wi Garage with Pops. I'm Pops, Nat Steel, and that's a wrap. So like, subscribe, and we'll see you on the next one. Cheers.