Revenue vs. AI Reality

Angela, good to see you. Another week has gone by and we had a few short share forms to catch up on and we were talking about how you had some really good questions back in the share forum I think a month ago month or two ago and I would love to hear what the top questions were and we can discuss through them recapping what people thought and our own insights on it. >> Sounds good. We are going to go back to the May share forum and right now we're in July. So we're going to hit some technical accounting issues that come out of the weird AI era and then we'll maybe bridge to some costs. So in May we

talked a lot about some of the new issues coming out of the fact that everybody's trying to figure out how to sell AI right now, right? So if all the product and strategy teams are coming up with new selling models, that means companies are releasing features that may or may not be priced separately all the time. And so one of the questions that came up is we have all these existing customers and of all these existing customers, we decided that they all got access to this new AI feature. And this new AI feature, we're going to

also sell maybe consumption based to our to our self-serve or product led customers. So it's going to be a separately priced product for those customers. But for our existing enterprise customers, we're just going to give it to them for free so that they can try it out and see if it sticks. Totally makes sense, right? Because nobody really knows how to sell this yet. Some people are selling by agent, which I think is interesting given me and you have both built and we're like, that's so weird. Aren't Doesn't the model decide how many agents there are?

>> Yeah. >> And some people are just trying to do it consumption based or some people are trying to fit it into their straight line normal SaaS plain vanilla selling. So the question kind of comes up in a couple different ways, but I'll stop there. Are you seeing the same thing where people are just like throwing stuff against the wall? They're treating their enterprise and their self-serve customers a little different and that means that there's a little bit of accounting chaos under the rumble, right? Because then people are trying to figure out what does that look like in

accounting? What's the performance obligation? Did it change the performance obligation? Is it a modification when you give that new thing? I think there's all these fun little things that that pop up because of that question. >> Yeah, I'm saying the same thing, no matter what you say. I think we saw a lot of the revenue leaders that we talked to in your latest event and some of the other, you know, in the share form as well are facing this as well. It's like It reminds me, and I think that's what we talked about previously, of like I just the speed of product is as product life cycles are just evolving so fast or product to launch and

probably life cycles, too, the decay of products as well, so fast where we used to be in a world where it might might have taken a year to launch a CD or on-premise software or develop updates. And that it seems like ages ago when I think I mentioned it and someone was like, "Well, how old are you?" when I talked about that at at the conference, but that was like 10 years ago, like 10-15 years ago, wasn't that long ago. And then we moved into the SaaS world where you'd launch continuous updates, but still like the updates would kind of be batched or if it was an update it happened once every couple of months, and maybe minor updates more frequently.

To now, we're seeing like the big AI companies change models almost weekly or to make some kind of refinement or updates weekly. And then you get talked to Anthropic about that, you know, like them out. And everyone's mirroring. Where almost every company I'm talking to, the product team's like, "We can launch something this week." And it changes the problem like to your to your point. Accounting or accounting function has never had to deal with this before. Normally, you have time to assess a product as a combined with your current like we assess it as a performance obligation, the what performance

obligations are embedded. What promises are they? Are they combined or not? Under six, but now we're in this world where, well, if you have this ever-changing road map of features changing within the products, you're selling them differently, how do you actually account for them? What are your performance obligations? And I think it's tripping a lot of people up. So, it's pretty interesting. Did I Did I get that right? Am I missing anything there? >> Yeah, I think that's exactly right. I think that anyone that was selling like a fixed amount of stuff for a year, every software company or SaaS company,

or whatever you want to call it, but an AI company, AI native everything, right? >> Yeah. Yeah. >> That they're now selling everything on the menu, or they're being expected to sell everything on the menu, or to continuously turn out features at a rate that is much faster because if you can't keep up with the changes, and the changes are quite literally, you know, weekly, daily, then what does that mean with your accounting conclusion that you made at the beginning that you said I had, you know, three performance obligations. Let's just do kind of a standard if you're a SaaS company

transitioning over to a SaaS plus AI company, right? So, maybe you had one platform fee that was ratable straight line. You had maybe some consumption component, but probably pretty new like X amount of storage, however many summaries you did, which is a very potentially AI token-intensive process, right? If you're pulling in tons of context and spitting something out. Um and then another piece that was professional services. Those are kind of like the standard three things that you may or may not sell. Now, all of a

sudden, that middle one, that that ratable platform piece is like, "Oh, no. Now, instead of just selling the storage bucket, we're going to sell like 15 uses of our little agent, right?" And maybe that's every month. And it's like, "No, no, no, not just that agent. We have the summarization agent, but then that thing's going to code you a mini app, which is not that crazy right now." And that mini app is going to be housed on this weird little extra part of our platform, and we're going to maintain this thing and keep it, I'm going to say socks compliant, but we all know that hasn't been figured out yet, right? So, like, okay, now you started with three

things. You now have five things. Is that a contract modification? What is that, right? >> And especially if you put on your contract, "Hey, by the way, we're going to launch new things every single week." Like, if you have a road map and your road map's constantly changing. It's not feasible at some point to continue to reassess all of your performance obligations, but you can also get very funky, right? Not to jump the gun there, but it It's like you can't just assess, well, the individual launch anymore. You have to assess, well, contractually, are you always giving all of like who are you giving all of these new offerings

to? Are you giving them to your existing customers? Because if they are, is that more of a stand-ready obligation where you have a platform and you're just changing everything, changing adding things to it and subtracting things and making it new, and is it bigger than the sum of its parts? Or are you really going to the business of like Amazon or shoes where it's like, "Hey, we're actually selling a bunch of different types of shoes, and each thing you buy is a different pair of shoes." I mean, it's debatable. I don't think anyone really knows where we're going because we're figuring this this out as we go, but I think there's some things that make sense under the like accounting

standards. Or and then go away from the even FASB accounting standards of like the basic the basic topics, like cut off, accuracy, completeness, timeliness. I think we got to got to go back to fundamentals in a lot of ways and make sure we're purporting revenue correctly in spirit and consistently, but >> Yeah. Um you bring up a great point, and it was that legal piece. And I'm going to translate that to legal enforceability, which is part of our step one of ASC 606, which is basically do you even have a contract, right? And

so we know like I I think it's fair to say that every company is trying to release new things all the time to keep up with what is expected of them in this AI world. Then there is a split between who has actually contractually changed that and who is just trying to release, right? Now, if you have that split of who has contractually controlled this thing, there's a couple different ways to contractually control. One of them is to say what you said, which is you're always going to get what we have, which sounds a lot like we have an open menu

and you have access to our open menu, which feels a lot like straight line. >> Yep. >> And then there's the people that what they do in practice is still that, but they haven't contractually caught up with it yet. >> Yep. Yep. >> So, if they haven't contractually caught up with it yet, and then you do another split, it's like do they get to keep that feature forever, or have you set an expectation that they get to keep that feature forever? That is super super gray, but if that you do, then you might actually might be in a contract modification. >> Mhm. Mhm.

>> And we have so many trees to split, but I'm just going to pick that one. Now, let's say that everybody is expected to be able to keep that feature, which I think is a little crazy, because I don't expect to keep any crazy AI feature that I get popping up in any, you know, software that I use right now, cuz I'm like, "Oh, that's cute." And I'm like, "Is it going to stick?" I don't know. Then I play with it a little bit, maybe. But, if it's a contract modification, accounting world goes in chaos, right? For those of people that don't know it, you have to end a contract and begin a new one, typically. And then if you do that, anything where real life was off from accounting life, meaning if you have like deferred revenue, or you have

a contract asset where revenue is ahead, you have to take those weird differences and then roll them into the next contract, which is not the same as what you're billing for. And you have to track this weird little difference forever until the next modification happens. So, I'm going to stop there, cuz I think that that's like the crazy thing that happens in accounting that might be trackable painfully on a spreadsheet for many existing companies, but really painful if you're doing it for all of your customers suddenly, because you pushed out a new AI feature and didn't paperwork it correctly, or

paperwork it safely for your accounting operations. >> Yeah, yeah. And I think even more, we're saying accountants and finance leaders need to take go into the front end of the business and work on product launches and I think a lot more or I think we were talking about it when you're in the last year form about how we're seeing a lot of like a lot of revenue and go-to-market roles with accountants in them become DP plus roles, because we need these leaders in this space to make sure that product

launches can happen compliantly and follow best practices and it tends to be the CPA in the room or the accountant in the room is the person who can bridge all of those gaps. Like the legal gap like hey that there's a legal issue here we need to document the contract, there's a finance issue, there's an accounting issue and there's a product launch issue and an economic issue. We need it turns out the CPA in the room is the one who can bridge those uniquely which is which is a cool thing about our profession but a problem for businesses. Like you said it's like there isn't I think there are some best practices and

maybe we can talk about those here but they aren't widely distributed and most companies don't know about them. And like you said every there's disparate practices through disparate ways or people are approaching these businesses and there needs to be a more uniform there doesn't need to be but it it if they should take a more uniform approach to this in order to make their lives a little easier. Because like you said in a real world if the the goal is we're just going to start launching a bunch of products all the time that are going to be add-on to our existing product. Well yeah we're probably more of an in a stand ready kind of role then that's a little easier to account for. Maybe that's going to change your earnings and

your revenue timing and whatnot. But like better to do that consciously and to understand it and then figure out how the pricing plays with that. Because the price pricing on a consumption based but your product's more of a stand ready type that's going to cause some heartburn with your auditors and you're going to need to get your story straight pretty well. If the opposite if your strategy is I'm really going to nickel and dime and try to launch a new product and sell thousands of products and become a new Amazon or Nike of the world but in SaaS land then then your contracts and everything else should

follow. >> Yeah you you said something about you know the dream that everyone had before was to get straight line revenue which is interesting right? But that is a disconnect from the reality of I think how people are thinking about pricing now which is consumption based which isn't just because it's like fun and like it's like the customer gets the value that they use, which is true, but it's also because of the cost side. So, before I go deep into the cost side, which I so badly want to, >> Yeah, I yeah. I see it. I see it in you. Yeah.

>> Let's quickly close out that that straight line piece. The straight line piece is you wanted it cuz you valued well a certain way, right? You had predictability, less volatility. You were able to say we were going to do X amount more next year because you had some renewal uplift that you can relatively control and forecast for. I think there is a debate out there, and I'm not actually sure but I'm curious to get your thoughts. Like, what is the investor world thinking, right? It's like, what's better now? Is straight line better? Is consumption better? Do

we not really know? Are you just trying to get close enough because if you accidentally get straight line but you're actually consumption and you flip, that's kind of dangerous, too, right? I'm not actually sure if straight line is as cool as it used to be. >> Yeah, and I agree I I think there's a big challenge with AI where the many of the investor community is forecasting based on consumption-based basis. If you mention AI or tokens or anything anything usage-based pricing, a lot of them it's lost on them if you're recognizing items ratably, but you might

be metering things on a usage-based way, which is an okay convention or okay path in a in accounting CPA land. But I've seen companies that disclose we're recognizing revenue ratably, and then investors talk about consumption-based revenue recognition. And so, there is this disconnect on do they actually understand how the numbers are being reported? And there's two benefits, right? Like, so if I'm a CFO and I'm trying to play with the numbers, which I don't recommend to anybody, but if Yeah, don't do it. But, if you wanted to, there's two way Like, hey, if you pull consumption-based, well, that tells

you a better earning story in the long run if you have back-ended growth because your consumption-based revenue will be exponential on like, let's say you have a growing customer base. Because if you start low and your products ramp over time, you'll have more revenue down the line. So, it tells a better growth percentage growth story. Rattable is easier to operationalize and is a cleaner story. It's easier to predict. Your forecasts are going to be a lot cleaner. Your accounting operations are going to be a lot cleaner. It's just it's more comparable. There's some real benefits to how easy and accurate your forecasts are going to

be. But, you're going to pull ahead some of the revenue and you're going to have explanations on like, well, where's all this deferred revenue and how you're going to perform on it. So, there's there's kind of pros and cons to each and I've seen CFOs and CAOs who like are really adamant about one versus the other. And it's strange as an accountant in this world because generally we like to like want to apply things consistently. Like, what what are the facts that make fit into rattable versus consumption base? And And the nuances here can really trigger a swing one way

or the other. I have a strong opinion that I think the world would be a lot easier if we all just said rattable and did non-GAAP for consumption just so we stop this this conversation and stop people playing games and figuring this all out, but and we get back to like consistency and cost-benefit theory for like accounting, get back to like, okay, let's focus on the real business. But, while we're in this world, it's fun to play around with it. So. >> It's a it's a very hotly debated take. And And I don't I honestly have not come up with my take

yet of what I think is better. I used to always think rattable was better cuz operationally it was easier. But, then if everybody's doing non-GAAP on consumption anyways, you have to operationally do both anyways. Maybe not to the level but you certainly have to understand it. And so, I'm not actually sure if the break from reality is amazing. My hope and dream, if I was to have one as an accountant, would be to bring back this this little old piece of guidance called the contingent revenue cap. >> Yeah, that'd be great. >> I think it was an SEC ruling wasn't

actually technically in the guidance when we were under 605, which was the old revenue standard. And it basically had this rule that said, if you can't bill for it, you can't recognize it as revenue. >> Yeah. >> Which in some ways solves a couple issues. It's not all the way to consumption and it's not all the way to ratable. It's just saying, I'm going to bring you slightly closer to the real world, which I think right now in the accounting world we are so disconnected from the real world that is insanely painful cuz it just means you need to two operating full systems of record, which is incredibly expensive and

painful. But >> I'm with you. So I would I love that you said that and I would I would pay money for a system to go back to that world. We can petition someone to bring it back to hey, and what we're saying to everyone listening, I'm sure mostly it's revenue accountants is hey, like at the end of the day take revenue as invoiced on on this if you're going to be extreme and at least we all know what we're talking about, right? And maybe maybe have some disclosures around well, how much of this invoiced revenue is actually earned or like how much like remaining obligations do you have related to it? How much is it prepaid,

etc. But man, that would make life a lot easier rather than currently, which is obscene. Like imagine like you said this new company comes in, now you have 20 new products and now under 606 you have to allocate your transaction price to all these products that you've never sold before and then recognize a portion of your total cut which portion of your transaction price, which could be your minimum commitment or it could be your order form or however you structure it has to be allocated to it and all your systems have to tie to that and you might have contract modifications like you said every time you launch a new product to reallocate

everything. >> you will have them. >> anyone Yeah, how and and how There's no way we're doing this consistently. There's no way every like you and I know we can't you could you can't you have at the end of the day you have to be able to make a deterministic logic to the way the formulas run and if Excel breaks when you do it for like a small company, there's no way these major companies are doing this correctly, right? So so I'm with you. I love that you said that and if we can bring I if we should get some shirts bring back contingent revenue cap or something like that. >> It's just a small link to reality,

right? It's a small anchor to this piece that we all understand, which is I sent you an invoice and I got an invoice from you. And that thing we hope, because we sent it to each other, is a real artifact, right? It is a small link to reality. Instead of me doing all these random calculations in my operating system for my either ratable or consumption, totally diverse, everybody decides it differently, sadly. And then we slowly map it back to consumption. And then somewhere in there is like deferred revenue with its own definition

with billing. >> Yeah, that whole contract asset topic of like where we started recognizing revenue ahead of billings got really got really funky really fast. >> And and conceptually it does make sense theoretically, right? Sometimes I do more work than I've been due. But then you get to like, okay, you do more work than you could bill for. That's like an accrual, accrued revenue, some people call it that. But but we're actually talking about something much more conceptual, which is I think that I'm going to So, you have a

consumption-based billed thing for a year. And you say, I think that this consumption-based billed thing is actually more like me keeping the lights on, so I'm going to take it ratable. But I'm also going to have to estimate how much I think I'm going to make over the amount that I put in the contract. So, not only is it this fake ratable, which doesn't It's not fake ratable, but like obviously it's not perfectly ratable. Sometimes it gets close and that's an argument. But if I have to do this ratable plus this fake amount that I think I'm going to get, called variable consideration, there's like two extra

disconnects from reality. I haven't even talked about whether it's allocated or not, which you brought up. So, you might have then three disconnects from reality, right? Then you throw in a modification, which I'm going to stop saying the word modification for a second and just say upsell, cuz everybody that's ever listening to this that's in sales, they're like, "Yeah, we Everyone wants the upsell." I.E., the goal of having this contract is to upsell your customer. If the goal of your If this the goal of this is to upsell, you likely will have a lot of them, hopefully, which means that this system is made to do this. And let's say we just said three to four disconnects

from reality. Disconnects from reality, which is a computational like just conceptual mathematical calculation for what this thing is. And every step that you remove from reality, you get further away from that number, i.e. further away from the billing, which is the true artifact that left my inbox and went into yours. >> Yeah. >> I I think that what makes it reality is that two people now agree on it, i.e. it went into your inbox. >> Yeah. >> Yeah, exactly. >> And >> the fact that we are getting more and more loops away from that reality, I think you have to take that step back and say like, why? Why are we doing any

of this? Is rolling forward that random contract asset that now is four or five steps or six or by the time you do two upsells, what, 16 steps away from reality? Is it worth it? >> Yeah. Yeah, I think it's a good question. When our job as accountants at the end of the day, and I I believe in accountants as an honorable like honorable profession term, I'm not saying like just bean counters, but like looking out for the books, we're like there to to figure out what the truth is. Like that's the whole point is like how can we accurately and materially represent this information in a way that everyone can understand for very complex businesses.

So, the further we get away from the truth, we're missing missing a bit of the core substance of our role, right? And so, I'm with you. Bring back contingent revenue cap. We got to print some shirts and get it start selling them at your GAAP savvy events or something because I think I think that's I completely agree. And like to go back to the people's questions of like how do you account for this, I think the answer is you account for it like yes. I've seen I've seen both practices and you kind of choose a lane and your auditor is going to be contingent, but your legal form is going to ultimately change how you account for

these and I recommend having a good person in the room to advocate. It almost reminds me of a sell-side and buy-side diligence. Like we originally like back in the day, back in the day, and not too long ago, if I had a business, I'd sell it to somebody or my best employee takes over my business. But then we all started selling businesses to PE firms and VC firms and people started doing shady stuff. So, you have a sell-side and buy-side representative to like, "Okay, I'm going to help you represent your business accurately." And someone's going to be like, "I'm going to make sure someone's not doing shady stuff." And I feel like we've come to this same world in

accounting where it's like, "Okay, yeah, we're have to launch these products, but we're really not sure what the product is now and we're trying to sell it, and we haven't figured out the legal form, but we're going to bill for it anyways." And so, we need someone to help represent this to our investor and the investment community, which happens to be your CPA, just like diligence. And so, teams need that. You need a CPA in the room, a qualified technical accounting expert in the room when you're launching your products to at least work with your lawyers to figure out a defensible position with your

auditors that's consistently applied, and you have to choose a lane for now, and hopefully in the long term, and not hopefully in the middle term, we can update the guidance to be a little more consistently or easily applied for this new world of living. Cuz it to be fair to have everyone who wrote 606, it's been 10 years now or whatever. I don't think anyone anticipated this type of change in the market, so. >> Yeah, I don't think anyone did. And as you're talking through that, I feel like all I hear is the word like really understanding your why, and I can't help

but loop it back to our other favorite topic that we like to talk about, which is AI, right? How to use it and how our world is changing in it, which is like right now we don't we prompt very differently than we did last year or the year before. Because right now we actually set goals, right? You set goals, like right now it's July, so we've both been playing around a lot with Babel, right? And Babel's a very different animal than the other models, right? You have to use its little expert big brain differently, or else your costs will skyrocket, right? It's a very

expensive and intense model. And it's really, really good at edge cases and judgment. >> Yeah, very impressive. >> Today will be very different than 1 month, and something else will blow our minds. But, if you think about that idea of like the way you prompt or solve a problem right now is you figure out the why really, really strongly, because that drives like how you get there can be many, many different ways. But, the why and the goal, which is like you said, is what is economically happening in this contract?

Like, what are we trying to do as an accountant? Like, what is the underlying economic substance? And if that's the goal, if the goal of our job is to record what is happening, and we get stuck in all of these weird little words of contingent very like yes, we have to navigate it, but I'm almost tempted to just put in like this weird complex thing that we're dealing with as a society in the accounting world, and like see what it comes up with. Like, maybe think of like tell it to ultra think, which also just came out, right? Like, what would happen if we ultra >> ultra super think level 10,000, you

know? >> all these weird little code words. But, like I'm tempted after this like what if we tried to ultra think this problem? Because I'm not sure like I think we both know the accounting guidance hasn't caught up. I don't know how it could. We we could barely keep up, but I I would I would I would have to be humble and say I'm probably not up to date, cuz how could you possibly be? But, like I don't I don't think we're doing what we what we think we want to be doing, at least in the accounting guidance. So, maybe that's the like contingent revenue cap is the best way I

can think about it right now of bringing us back to a link of reality. But, like really framing like what is the goal? What is the goal of the guidance? And if the goal of the guidance is to reach the economic substance, which I know is also how we got to this rule for the contingent asset or the asset contract asset. But, like when you take that step back and you look at all the systems and the processes that are going into place to create fake realities within a company that get in then gets like redigested by the investor community back to consumption, which barely anyone can understand anyways. Like, right? Like,

is this serving the goal? >> Yeah, well said. I think I wanted to go I think it bridges really nicely to your next topic that you want to talk through, which is cost. Because at the end of the day, what are companies trying to do? They're trying to solve a problem for profit. How do you make a profit? Well, you have to make margin on whatever your costs are. And so, every business right now is trying to figure out how much AI will cost them, how they can charge on top of that with all of their other employee business costs, et cetera, and and ultimately drive a drive positive operating cash flow and drive

net income and drive positive margin. And to do that, their pricing has to match how they're being priced. And we all know Anthropic is a consumption-based business and all these other businesses are very I mean, they have their subscription offerings, but any business is going to be using specific consumption. And like you said, which I think is the best connector, is like these models are getting expensive. Like very, very, very expensive. Like when I had the max, $200 to be playing with this. And as soon as I used to never run out with Opus. I used Fable for like 5 minutes and it ran

out. Like my 2-hour window ran out like in a second. I was like, "Whoa." And I companies I've represented I've seen their spend going from hundreds of thousands of dollars a month to a million in a week with Fable. So, yeah, Anthropic's trying to make its money and we'll see we'll see how long people are going to pay for that, too. So, that's kind of a question in the market, too. How much are people willing to pay for this? Or companies are looking for cost benefits. So, if people stop being like, "Whoa, I'm not going to pay that much." People need to shut off Fable or whatever it is, that's going to be a different story. But, if from the matching principle of revenue is supposed to recognize the costs that are

related to the revenue that's being generated, then like should we all be moving to a consumption-based world, which then it goes back to your like well, easiest way to do this is to at least mirror invoicing and some to some degree. So, kind of land back in the same spot. But, but that is an interesting piece, right? Like the the cost basis, which we never had in SaaS used to be profitable because well, you after you built the products and people bought it, then it it just pretty like just you have to worry about the maintenance. But now we are really in a world where yeah, these AI products are seem to be getting more

expensive not less, which I don't think a lot of people were anticipating. >> Yeah, I think we are at an inflection point in July of 2026 with what happened with the Fable model because it is such a different animal. I think we we were messaging back and forth about I was like, "Oh, you you got to use Fable as the big brain and Sonnet as the like as like the manpower to implement, right?" So like Fable's the advisor, Sonnet actually does is the workhorse. And so Sonnet will come up with the plan, you do all the planning or the architecture,

it does the the adversarial attacks and the edge cases, right? But who implements it? Like it's the Sonnet agents that actually go out and do it. So I say that all to say that even the way we use the models are constantly changing. I think the surprise bills this month, I think a lot of people will be angry just to be totally honest and I'm not sure how it's going to be handled, but I'm sure that the reconciliation back to reality is going to happen through billing. >> Yeah, yeah. Well, here's the thing that people like people tend to miss especially with the AI in businesses,

everything's token-based right now, right? And so like based on how many tokens you use, you're going to get charged. That's from from these AI companies like Anthropic, Gemini, etc. for APIs and their connections to it. But everyone's using tokens differently. Like I hear I have some companies I represent that are like, "We're selling tokens." And then they have a completely different definition or credits that are related to tokens. They're all using similar nomenclature, but there's a problem with that pricing theory. And the problem is you have no way of anticipating how people how much tokens

like let's say you have used Fable versus Opus. And if Fable costs two times as much as Opus, that's what they say. It's going to cost two much times as much per token equivalently. But Fable is also just going to run a ton more tokens cuz it's trying to think a bunch more. So for the same use case, it might actually cost a thousand times more. It's going to do it better, but it's going to run a lot more tokens to solve that same problem. So, it's like you can't We're not even working tip-for-tat. It's like if I sold you a pair of shoes and I like, here's your Prada shoes or I don't know, nice shoes. I don't buy nice shoes cuz I'm an accountant. So, whatever, your nice pair

of shoes and then here's your Walmart brand and the Prada ones are 10 times more expensive. At least you know you're getting a pair of shoes. But here, you might be like, I'm choosing Fable and you think you from a shoe to shoe, you don't just spend 10 times more, but you also somehow show up with a thousand extra shoes on your doorstep. You know, like you you're act can't anticipate how much they're actually ordering in these situations. So, like you said, I think it'll be interesting to see how that resolves. >> Yeah. >> Over the next couple months. >> And when you were bringing up that we're probably raising more questions than there are answers, but I I think about the fact that like, yeah, you can even

talk about tokens, but the the the price of a token is totally made up. So, the concept of a token and a price of a token. So, like picture again, two two variables. Like as at the end of the day, we're human beings and all of all of our ideas are based on Actually, let me take a step back. This is something I've always found fascinating. Our concept of currency is completely made up. Meaning, my belief in what a dollar costs is only only works if you also believe that dollar is worth that much. >> Right. >> The concept of currency is actually fully based on a trust-based belief

system of value. And so right now, we're actually working out that trust-based belief of value. So, what is a token worth in size, right, and in terms of dollar amount. And so like, Anthropic and OpenAI have completely different ways of strategy of dealing with this. Like I I remember I saw this like Codex ad pop up that was like, when when when when like cloud runs out, go ahead and, you know, pick it back up with Codex. And that's actually true. Like that is like, you get so much like you never I've never run out. I've never run out

on ChatGPT or Codex. Like I've never run out. Whether that strategy has to do with me getting addicted to using it, questionable. But like, that just means that we're playing a little bit of like we don't know how this cost is going to roll out cuz that cost could be absorbed with you know their current investors, right? It could just be this weird amount on their books. It could be some weird liability. It could be caught in like partnership like we don't actually know how much this costs yet. Like I'm saying me and you don't know but maybe someone deep inside that company knows.

>> Well no it well it's it's strange because just like the models, right? Like you're not getting deterministic outputs. It's non-deterministic. So it's a little weird. It reminds me of and I'm going to shout out to any gamers watching or any people like there's certain games that are RNG based like random number generation based that are there like for people who are trying to do things that they enjoy the experience of gambling but they're not going to go gamble and there's a lot of games that kind of scratch that itch. I'd say the sports betting itch and the prediction market and it's a little bit like that.

I mean I'm curious in your opinion though and like I kind of feel that way whenever I'm running something. I'm like I wonder how much it's going to cost me this time. I'm not sure. I wonder if it's going to eat up all my usage or I wonder if it's going to just take me and maybe get a little better at it as you like work out specific workflows but I can only imagine at scale there's a little bit of this RNG random number generation of like what is what is this use case actually going to cost me to solve? >> Yeah. >> It's like it's not clear. It's not transparent because we don't there's no way someone can know oh yeah I'm going

to do a reconciliation on this on my month end. I know that's going to cost me 27,000 tokens. Like from you and me I like we know what a dollar is but a token is pretty abstract. So >> Yeah, well I think that I think we are just hitting that phase. Like two months ago I wrote this up stack article about you know like everyone's trying to figure out the ROI but we were actually just trying to get people to use the technology, right? Like how do we get the accounting industry comfortable with like playing around cuz the only way that you can learn is to play. >> Yeah. >> And the spout seemed to be open. You could you seemed to be able to run and

maybe you'd hit that that barrier but then you're like oh I'll switch to the next one. But like, what we were trying to do was just adoption actually at that point. I feel like now we're hitting this point where we're like, oh my gosh, this bill's kind of high. But like, why is the bill high? Is the Is the bill high because now we're getting charged by it we weren't before? I don't know. >> Yeah. >> Right? Like I actually have no way to audit the system. Or is it because the the rate went up? Like I don't like PG&E, for example, and a lot of people say that it's moving towards this utility-based thing. It's like suddenly like one month my my bill is like double or triple sometimes. And it's like, oh,

it's just cuz we decided that the rates were going to be higher for San Francisco. I was like I I have no control over them doing that. And and I have to use electricity, right? >> Yeah. Yeah, exactly. >> And and and because the world is moving like it is kind of like that random number generator, but like there is more control at certain levels and less control. We probably have the least control. You know, two months later from writing that people were like token maxing, right? Now it's like we got to pay attention cuz I just ran through my budget. Like I can't just build random

apps. Like I have to have some destination, which honestly kind of sucks for those that hadn't been playing around before. >> Yeah. >> Right? Cuz I don't know how you learn without being able to waste some. A lot, actually. >> Mhm. >> And so like, it's weird cuz the education bar just got a little higher actually cuz now it's more expensive to learn in a way. Unless like if you're on Claude, like go nuts on Haiku, right? So on, it's pretty pretty good to like you can like probably do whatever you want on those instant models and chat you like whatever. But those expensive ones you probably like there should be like

little warning signs or like there should be meters on them. You just say like, oh, and and they did do that. Like I'm actually very thankful for Anthropic that does say, you know, you're you're running close to your limit. Cuz then I like this honestly this week only I found my little usage meter in Anthropic cuz I also have this max plan. >> Yeah. >> And so then I started tracking it by my like usage window limit and then like weekly resets. But like I never had to play that game before until now. And I I think it's because the ground is moving underneath us. >> Yeah, I agree. I agree.

And then yeah, if you have the question I think there's so many different ways it can go and we can't predict the future, right? But it's like will be a utility. And does this become more like utilities in the future? If it does, is that like Anthropic the next PG&E, you know, >> [laughter] >> like what or is or is it really a technology? And then if it's a technology, businesses are going to need some transparency and understanding of their pricing. And that's where I kind of go with like in a revenue model world, are we spying and selling shoes? Are like like just like a business like Nike, well I have to know

how much each shoe is going to cost. Am I going to do product cost costing or am I selling access to a service? And I think the costs are eventually going to have to mirror that or at least become stable enough to understand. Like we can't live I'm going to say we can't we have to say yeah, but it eventually companies have to have predictable cost structures in order to survive. You can't have unpredictable cost structures just to survive. So you can build your business on top of it because those costs get passed down to consu- passed down to consumers and place a margin. And eventually

people like I we're fortunate enough, right Angeleta, to like pay $200 a month for the plan. And there's a lot of people in our bubble that are fortunate enough. But you go down a few rungs of middle class like lower middle class or different even in different states, that's a lot of money to come out of a someone's budget to use a product. I mean it's a lot of money for me. It's like 200 bucks it was like it wasn't like yeah, let's go. It was like man, this is going to be that's a lot of money. It's more expensive than Riverside or any of these other companies that are like 20 30 bucks a month. Right? It's it's pretty premium cost. But in other countries, that's like the

unforeseen like a lot of countries that's not affordable for most people. And then if the cost continues to climb, who's going to be excluded from it? And then what businesses will be excluded from it just from a cost perspective? Yeah, I mean it becomes an interesting like how to what's the long-term business model? And then I'll put the counter to it. Well, most of the investors in these companies are also their largest consumers like Google has a large stake in Anthropic, Microsoft has a large stake in OpenAI. And so I like to say out loud like, well, this will got to sort itself out. Well, not if like the pony show keeps, you know,

continuing or it's like, yeah, well, rev number goes up for OpenAI, the number goes up for Microsoft. Well, I'm sure they'd love to see that continue as long as possible. And so, so that might move forward despite consumer challenges. But like I'm sure there's a number, right? Like if the next plan became a $2,000 a month or $10,000 a month. There's a number where it's just like, okay, this doesn't make sense anymore for our business. And so, yeah, I don't know. I don't know. Rambling a bit now, but you get what I'm saying, hopefully.

>> I was wondering if I should wrap it back to accounting because I know we can go down this rabbit hole quite deeply. >> Yeah. >> But a a lot of companies, so this also came up in that May share forum is like, do you capitalize the cost these AI costs, right? To build these things. And how do you do that? And one of the thoughts that came together was, you know, that a lot of these companies are moving towards doing a per employee per employee budget, right? So, as an employee, you get a budget of 200, maybe

like, you know, whatever plan or however many tokens cuz it's an extension of your ability to create, essentially. >> Yeah. Yeah. >> And so, um the discussion we had was, is it more like the way that you do a percentage of an employee costing with like their equity and things like that that was like a portfolio approach of how you determine how much you'd capitalize or what piece they're working on. But it's an interesting concept cuz in general, it is this thought process of like, it's an extension of the human being. Like, who gets to have the extension of themselves? Like, you can afford it, I can afford it, like these

companies can afford it. Cool, these companies can 10x their build speed, right? And release these things at this cost. And I bring back the at this cost because, like you said, software used to be like 90% margin. You went into SaaS, it probably got a little lower cuz cloud costs. And now we're squarely in a different world, which is like you got your cloud cost plus you got your AI costs, AI token costs. And now that is being more equated to a human being, like a employee plus 200, employee plus 2,000, right?

>> Yeah. >> Per month, right? And so now we're choosing, okay, you're going to be the enabled employee. This is how much this is going to cost. Like we're actually getting back a little bit to manufacturing, right? >> Yeah, that's Yeah, that's where my mind goes. That's why I keep saying Nike and whatnot, right? >> Now like lifeo, FIFO costing, right? It's like actually your margins are only, you know, in manufacturing what, a 30% margin was really good. Maybe that's where we're going and we haven't caught up with it. But that would be quite interesting because like there's been all these predictions that like, you know, like AI will take your job, but I know that me and you are busier than

ever. Doesn't feel like AI is going to take my job. >> It feels like the opposite, yep. >> It just seems like you've more crap to sift through, right? But like I I think that that's an interesting piece. Like are we moving closer to the manufacturing world, which it it already kind of felt like for certain infrastructure-related, cloud cost-intensive things? And now it's like, oh, it's a per employee plus count. Interesting. >> And yeah, if we go in that world, the manufacturing world, then my gut says like why are we capitalize it? Like this is all short-term. And like it kind of

like if we keep changing and updating products and everyone can make new products all the time and we're going to update and sell, does it stand ready or do we just have these little discrete products? Are we selling shoes? And if we're selling shoes, well, the new pair is going to be we're going to sell another pair next month, so do we need to capitalize for all the new all these developments because like the obsolescence is so fast. Yeah. >> Absolutely, right? Like I I In In some calculations, right, it's like how much of the code base do you rewrite and reuse, right? And it's like, I mean, me and you are new developers, if we could

even say that. >> Yeah. >> Like there's tons of stuff >> New pseudo developers. >> I have no idea what I'm doing most of the time. So like, you know, like it's it's an it's an interesting place. Like, maybe we are getting closer to selling little widgets, but at some point, there is going to be this code base that works really well. My gosh, okay, sorry, one more tangent. I've been thinking about this one a lot. Maybe this Maybe at next time one. I keep thinking about this very specific idea Like, in accounting, we have a, you know, term license concept, where the

term license is worth X amount, i.e., how much is the code base worth up front? If you did perpetual license that wasn't a cloud, and how much is maintenance, right? It was this weird thing that happened when we shifted to SaaS software. And so, you no longer had the weird blip of revenue up front and a small maintenance line. And as accountants, you always love to call this thing, like, every big four firm loves the 80/20 split. 80% is the license value up front, and 20% is the maintenance. Again, when we went into the SaaS world, that was like, "Oh, screw that. We're just going to do 100% over time." And that was nice for AR. We

loved it. Now, we're in the AI world, and and I really think about that 20% a lot, and I think about how much it costs to build the first thing versus how insanely painful Like, anyone that's built software knows this how insanely painful it is to maintain that thing, and make sure that thing works. Is that 20% way too low? And does that then like 20% the upfront, and then the 80? That's funny. I never thought of that. That's great. Which actually points more towards straight line, which is your kind of first argument, which is interesting. Anyways, that I like wrote this on my board. I was like, "What is the cost of maintenance?" I was like, "I

don't think it's 20%." Yeah, cuz I tried to like what I tried for my last class, I made an online module for all of my multiple choice questions, which no professor I I've never met a professor that's like, "Yeah, just made an online website with all your multiple choice questions using AI." It was great. It took me like a weekend to do that, which was amazing. Like, I I had hundreds of multiple choice for a finance class. The students loved it. But then a student came up to me and was like, "Hey, one of these multiple choice questions, could you change the answer?" Cuz I think it's a little off. I tried to go do that, to

figure it out in the, you know in the code. It was going to take me longer than the freaking build, right? Like to go how how what's the one question and I like try to go work with Claude on it and Claude's like, "Well, we have to re-go through the whole code base." And it's just like, "Oh my gosh, which where did I put this and how do I access the code and which Yeah, like so exactly. We're going to have to go through that world really fast. >> What is the cost of maintenance? What is the real cost of maintenance? Because I think that the upfront build is so easy, the maintenance is going to be horrific. >> Yeah. Yeah. Yeah, exactly. So then you

get people like us, the pseudo developers, right? The pseudo developer like, "Yeah, we made this amazing thing." Yeah, but it I wanted to do this little thing. Can you fix it? Oh my gosh, no, we can't fix it. It's going to take us a year. I want to I want to bring one other thing. I didn't do Have you heard of Jevons paradox because I think it's a cool way to close this. >> No, tell me. >> Okay, that was my thing I learned about this week where there's a traditional theory that if we the more something becomes available, the cheaper it becomes. Right? In in economics. Like supply goes up, costs go down.

Jevons paradox said the there's that's actually false in a lot of industries. And he's the coal This guy Jevons used the coal industry as an example. As like as something becomes more available, a lot of times the cost actually increases. Especially in the short to mid-term because consumption increases. And there's an interesting example of water heaters and coal. Which so like people 100 years ago, I don't know when the water heater was invented, so take that plus or minus whatever amount of

years, didn't have water heaters. So it was like a pain to heat water for your bath. People took showers once a week and just like threw cologne on or whatever. And you had to like boil water over a gas stove or and people thought that like, "Well, when we there's only such demand for hot water. Like you need to bathe once a week." But as soon as we made water heaters, it turns out there's this exponential demand for quite a while. People like, "Yeah, I like hot showers every single day." So water heaters went from being small to pretty substantial and it seemed like the demand was infinite. So, like the access

to more hot water energy, like the ability to heat water in water heaters or the access to coal is a good example. Like as soon as a steam engine was made, the train was invented which consumed a lot more coal than was ever needed before. It actually increased the price of coal, increased the price of gas and hot water in the short to medium term until you actually saturate the market with the overall need. And so, I think there's like this question of when will the market are we here now? Like will the costs of AI actually be exponential because and energy and all of this until

we actually figure out what our saturated the saturation point is which could be longer than we expect. Like with water heaters and coal it was like multiple decades, right? >> Does >> Yeah, I think we're at the beginning. I really do cuz it's it's it's intelligence or knowledge on tap, right? Oh, I I think people call it intelligence on tap, right? And like we've always wanted that like people I do think it's like that paradox where more people will want it. I do not think the cost is actually known or real yet at least to us.

>> Yep. Yep. >> And I think we're literally making up new ways to increase capacity, right? The whole SpaceX piece, right? >> Yeah, it's data centers in space. >> Right? Creating new capacity. And so, I think it's interesting what you mentioned Jevons paradox is like if we're talking multiple decades, are we at the beginning? And I feel like that is, but every day just moves faster than the last one. >> Yeah. And well, and and so I I guess to that point, it's easy and I was in this camp like a year ago. I think you and I have had discussions like, "Well, the AI

models are getting faster and better so costs should go down." I think everyone was kind of like, "Oh, that was the trend." Even when I was at Google, that was a very common thing is like, "Oh, this mod this pricing strategy seems very expensive now, but the price will go down as the models improve." And they are improving exponentially, but our demand is exceeding the improvement rate which is extremely interesting, which puts that Jevons paradox. So, well, like, okay, we're using Fable, like, Opus was great, and now we're using Fable, and it's like we can't get enough of Fable. So, if there's like 100 more iterations of this that are continuing to be more and more expensive, when will

we be satisfied? Which I don't know. It's like you said, I think we're at the beginning, which is kind of crazy. And what does that mean for accounting? >> We always wonder how >> by it. Means stand ready, contingent revenue cap. Let's go. >> Yeah, but we we always go off path, but we always come back to the accounting because the why behind accounting is to capture the economics. And if the economics are the at the beginning under Jevons paradox, then we should be humble that we actually don't know where this is going to go because both lines are moving. The top line with the revenue,

the cost line is moving, the technology is moving. It's not to say do nothing. It's just to say stay humble and stick to the pieces of the reality that we know, which might be the contingent revenue cap in that bill. >> Yeah. >> And then >> your contracts. Like, at least work with a get a CPA in the room with your lawyers to figure this stuff out, and don't pretend like you know it. Yeah. >> And you know, [laughter] >> we're figuring this out as we go. So, I hope that answered your I hope that answered your audience's question. I don't think they meant that for us to go on this tangent when they asked about road maps, but I hope it satisfies them.

>> I'm sure it didn't, but I'm I'm excited about where it went. >> Okay, good. I'm glad. Well, this is fun. I think this is a good spot to pause, Angela, and we'll continue the conversation next week. >> Sounds good. >> Appreciate it.