Minus AI, US Edging Toward Recession & Unauthorized Agents Accessing Banks

0:04David Leary: An independent economic forecasting firm called Oxford Economics has done some calculations, and they believe that the American economy may have entered recession territory. We might be in a recession if you subtract the massive influx of capital that is flowing into the development of artificial intelligence.

0:27Blake Oliver: Coming to you weekly from the OnPay Recording Studio.

0:56Blake Oliver: You've been in Santa, you were from LA. You've been to Santa Monica.

0:59David Leary: Oh yeah, like the promenade there.

1:02Blake Oliver: Yeah, that outdoor shopping area. It's super high-end stores. Guess what I ran into when I was there? And I'll add it to the video or the screen here. It's a new TurboTax store. And this is like, these are expensive, expensive, probably some of the most expensive retail locations that are in California, right?

1:21David Leary: Yeah.

1:22Blake Oliver: How can you run a tax shop with this kind of rent? Like, this is all I could think about when I was there. I was like, these aren't, it's not like the ridiculous margins that Patagonia is running or the Apple store is running. Like, how is that? How is TurboTax doing this with these storefronts? Or is this just like, in the same way they sponsored the Intuit Dome, this is just basically a billboard. It's not really actually to do the taxes in there. It's just a billboard for TurboTax.

1:49David Leary: Well, and I mean, isn't the whole idea of TurboTax, you're doing your taxes yourself on your computer at home. I don't understand. What do you do when you go in?

1:59Blake Oliver: The TurboTax Live thing. And it's supposed to also, I think these stores are supposed to be like a little Apple App Store, a little that Capital One banking thing, come in and have a coffee. But my thing is they're doing in the most expensive real estate spots. It's not like, you think about Liberty Tax. It's just like in the crappy shopping plazas you're driving by, It makes sense. You could probably run a tax shop that way. But who can afford – this is probably one of the most expensive retail locations in LA. I don't know how they can do this. It's crazy. So for those of you that are listening, it is a beautiful storefront, all glass.

2:37Blake Oliver: Think the way you would have an Apple store. And it says TurboTax Santa Monica opening soon. And it says beautiful glass building. You see the red circle checkmark, the TurboTax logo you're used to seeing. but they're going to bring the live exports to this retail location. I just don't know how the math works for this. This has to be a marketing expense. It's not actually, they're not going to run each unit. They can't measure the ROI or the profit on one. It's just too expensive.

3:07David Leary: Well, David, maybe we'll get the answer to that in some point, figure it out. In the meantime, let's go ahead and thank our sponsors of this episode. And that is -

3:19Blake Oliver: Sponsors this week, we have Digits, Numeral, OnPay, and Cloud Account and Staffing.

3:24David Leary: All right, I'll thank Digits. Let's be honest. Accounting software hasn't changed much in decades. The prices keep going up, but the software still expects you to do all the work. Digits is different. Digits is the world's first AI native general ledger, with built-in AI agents trained on your firm's standards across every client in your book. They code transactions, prepare schedules, reconcile accounts, run quality checks, and even chase clients for open items. So your team moves out of prep and into reviews, advisory, and the work that actually grows the firm. And because everything runs inside one platform, your ledger, reconciliations, schedules, reporting, bill pay, client collaboration, there's no more stitching six tools together just to close one client's books.

4:09David Leary: Firms on digits are reporting a 70 % gain in workflow efficiency, shipping annual cleanups in days instead of weeks and running monthly bookkeeping in one to three hours per client. Accounting software that actually works for your firm. To see why hundreds of firms are making the switch to digits, head over to accountingpodcast.promo slash digits. That's accountingpodcast.promo forward slash D-I-G-I-T-S. Thanks, digits. What's new, David? I am going to the CCH Connections conference this week in Orlando. This is my first time at a CCH conference. I've been in accounting now for 20 years, and I've never been.

4:56David Leary: I've always gone to conferences like Intuit Connect and Zerocon and Scaling New Heights. Those are cloud-based, small firm conferences.

5:06Blake Oliver: Because you're in the CAST side. That's right. That's our background.

5:08David Leary: Yeah, so CCH reached out, Wolters Kluwer, and said, hey, do you want to do a podcast at the conference? Do you want to come see what we're doing with AI? And so I've been diving into the new AI features that are coming. I'm looking forward to learning more. And I'm going to do a live podcast episode there. It'll be at the CCH Access Theater stage on Monday at 2.50. and we're going to talk about all the new AI stuff that is coming to their products, which is really interesting because we've been waiting for years for this. These are the products that the larger firms are using and they haven't had access the way that small firms have because...

5:55Blake Oliver: So this is more on the tax side.

5:56David Leary: Tax side and audit side, yeah. Got it. And I'm going to talk to Jason Marks. He's the CEO of Wolters Kluwer Tax and Accounting and Kathy Rowe. and we're going to do a live episode there and broadcast that here on the show later. So I'm looking forward to that. So we've got a lot to talk about. It's going to be a wide-ranging episode this week, all sorts of different stuff. One of the stories that really interested me the most from your list, David, is this one about unauthorized AI agents accessing banks. And the banks are kind of freaking out about this.

6:33Blake Oliver: They're freaking out about it. And the big reason they're freaking out is they can't tell the difference between an AI agent. Let's say I use my Facebook Muse AI agent to do a banking task for me, or you try to use your cloud co-op to log into the bank and transfer money. It can't tell the difference between that, which arguably would be an authorized AI agent versus a rogue AI agent. So now the banks are panicking and they're just starting to shut everybody down.

7:01David Leary: Yeah. There's not going to let anybody access. Yeah, it makes sense, right? Because you can, I mean, I've been doing this to some extent, right? Like Claude will sometimes not log into a banking system or a bill pay system. There's like these internal safety checks that it does, but then sometimes it will. And so people are doing what I'm doing, right? Like I've got Claude sometimes going into RAMP in order to like add receipts to transactions, which it can't do via the MCP or pull statements from Relay, for instance. And it will go in and it can log in as me. I get it in there and then it will do these tasks.

7:47David Leary: But it's a big concern because it could also, as me in the website, move money around. And so I have to like keep an eye on it.

7:56Blake Oliver: Yeah, watch that second browser, what's happening.

7:58David Leary: Exactly. But, you know, we've got people all over the spectrum with AI and some people are playing fast and loose with it. So the risk of like these AI systems getting hacked via prompt injection and then going and moving money, this is going to happen. We are going to have fraud start to occur when a business owner or an individual tells the AI to go move money or to manage moving money around. And then it does something that they're not expecting. So I'm curious, tell me more about this story with these agents accessing the banks. How did this come to your attention?

8:41Blake Oliver: So this is all the big banks, JPMorgan Chase, Bank of America, Citigroup, Morgan Stanley, Wells Fargo. All of them basically are now going to restrict third-party AI agent just as a defensive measure because they don't know. They can't tell the difference. And ultimately, if they can't tell the difference, how are they ever going to be able to tell the difference? Unless there's going to have to be like – we've talked about this with GL. You almost need to have this layer, an orchestration layer in between. It's going to have to go through like a plaid, right? There's going to be one guy. All the agents have to go through plaid to talk to a bank.

9:18Blake Oliver: It's going to be, which sucks because that stifles innovation. But that's probably the only way they're going to be able to secure it. I mean, the banks don't have Wild West AI agents just connecting into accounts everywhere. And this makes me think about the whole self-driving cars. They have a goal to get you from point A to point B. And if it has to cross a double yellow line, it's going to do it. It's the same thing. Like, I'm going to invoice somebody. They owe me the money. And I'm going to tell an agent, go collect the money. And it's going to go figure out how to get into somebody's bank account and take the money.

9:49Blake Oliver: I think that will eventually happen. And so the banks have to shut this down because agents are going to go and try to do their goal, which is go get paid. Right? Go get paid by this client. And they're going to figure out how to do it.

10:03David Leary: Well, there are potential solutions here. You said to me before we started the show, you've got a story about how Stripe is launching a treasury agent. And so when you say treasury agent, I'm imagining this is like an AI agent that can move money around, I hope securely. That I don't have a lot by connecting to the banking services.

10:29Blake Oliver: But so Stripe, we have to keep watching Stripe because I feel like Stripe is on a march to become an accounting operating system of the future. So step back and think about Stripe historically. Stripe, at most startups, owns the profit and loss, the income line. The biggest line on the profit and loss, the income line, is Stripe. Now Stripe, remember a few months back, Stripe bought that token management company to manage the token spend at companies, kind of getting in that game. Well, soon, token spend, if we believe Hector Garcia's theory, token spend is going to exceed payroll expense under profit and loss.

11:06Blake Oliver: So now Stripe has the two biggest lines in the profit and loss they're involved in. Now they're getting into balance sheet workflows, paying bills, transferring balances, checking the balances. That's what they're announcing now. So they have these agents that will help people do that. And then they want to expand it into creating virtual cards, setting spending limits, that type of stuff. So Stripe is slowly pivoting into this accounting OS. and what i mean arguably the gl is only going to be a teeny piece of that like that doesn't knock that out like nothing and for for a lot think about all these you think about uh these startup banks like uh mercury right they're targeting these startups and they're like get your startup bank account and then we're going to give you all these banking services slash accounting services but who gets the startups first stripe because you've got like you set up a startup you usually they need to start charging people money.

12:01Blake Oliver: Yeah, yeah, yeah. So Stripe's further ahead in this startup accounting ecosystem. And so they're just starting to provide all the same services. They don't provide the bank service, but they connect to the banks, right? This, and then on top of that, Stripe also kind of has that, like you just kind of keep the balance in Stripe for a while. I mean, we do it. We keep a reserve balance in Stripe now. Like it's, they're becoming a bank, But they're also becoming an accounting operating system. We have to keep watching them because they just keep adding these layers of accounting functionality. And just like I said, just look at the profit and loss.

12:39Blake Oliver: They own the two biggest lines on the profit and loss now.

12:41David Leary: Going back to this question about AI agents accessing banks, right? Let's just think about why these traditional banks are having a problem. it's because they've never adapted to the modern need for multi-user access in many cases. Unless you're on the business banking side of like a Wells Fargo or a B of A, you don't get, as a business owner, often you don't get like multiple user accounts with different permissions. And that's been a problem with accounting and bookkeeping forever, especially on the cash side, like I was on. And often even like provisioning those user accounts with like different permissions is a huge pain.

13:34David Leary: And so people just don't. And the business owner will have one login that they have to share with other people like me as their accountant. And it's fundamentally insecure because we don't have the different user permissions, right? Even just getting a read-only login is still a hassle. And so this is the same problem. Thinking about it, it's kind of the same problem we've been dealing with. They're now having this problem with AI agents because business owners are giving the full login to the agent and asking it to do work. But they were also doing that before with their human accountants. Their accountants.

14:14David Leary: Yes. And so the banks, because they never built out the multi-user kind of permissions, now they're in this problem, right? They have this problem here.

14:26Blake Oliver: Yeah, because in theory, we use Relay. if we create a new Relay user, earmark agent at earmark.me, or whatever, and we create an agent and we create a login just for that agent at Relay, we could say it only gets access to the statements for statement retrieval, and that's all it could do.

14:47David Leary: Exactly, it can go in and it can pull transactions, but it can't pay any bills. Or let's say we want it to act as like an AP agent to do accounts payable. We could say it can stage bills and submit them for approval, but it can't pay them. That's the workaround. I mean, that's the solution, right? It's just to actually have multi-user access with permissions. And so an app like Relay, a bank like Relay, is actually well-positioned for this. And that's what I've done. I haven't actually done that with...

15:20Blake Oliver: Don't tell me you're using your login and letting it go into Relay. I'm going to freak out.

15:24David Leary: uh well i've i've got watch it while it does it right okay and it can't log in without me but um i've actually done this with zero where i've created an agent login so it's it's a login just for the ai agent just for claude and it has limited permissions right just like a bookkeeper so it can go in and it can code transactions for me via chrome and do reconciliation tasks but it It can't do the work that I don't want it to be able to do. It's got restricted permissions.

15:54Blake Oliver: Yeah, it can't get to the preferences and the admin features and all that stuff.

15:57David Leary: It's not the advisor level or admin level permissions. And so that's the solution, I think, is you want to give access to the bank for your AI agent to do work. It's really, we just have to think about this the same way. We have to think about the AI as just being like another human team member. And so that might mean giving it its own email address and giving it its own logins to the different systems. But of course, then you have to have systems.

16:21Blake Oliver: Yeah, because if you think about if you're at your accounting firm and you authorize, or we've had this happen here, Blake, you essentially have God mode to all our systems. So if you let AI use your login to our systems, now AI has God mode to all our systems. And yeah, you have to create, you have to separate this out. you need to have AI only have, you have to treat it as a new employee, spin it up its own credentials, give it the proper access, and then let it go to town. You just can't have it use your login, especially you, Blake. You can't have it. Whoever the highest level of permissions at your company, they should not be allowed to connect AI because that's going to give it God mode.

17:01David Leary: I want to keep talking about AI and some better use cases actually for it. I don't think that moving money around is the best use case for AI at this moment. I've got a story here from Zapier about a solo lawyer who uses AI to run his practice five times faster. And there's some great examples in this blog post that also would work well for solo accountants and CPAs. I want to get to that. But first, let's thank our next sponsor. And that is Numeral. Let me ask you something. When you recommend a sales tax solution to a client, are you confident in the support they'll get? What about the pricing?

17:45David Leary: What happens if something goes wrong? As accountants, you've all seen it. Clunky support, surprise invoices, and nobody picking up the phone when there's a problem. Numeral is built differently. It's a modern sales tax platform built for accountants, and it supports registrations, filings, nexus tracking, exemption certificates, and even VAT and GST in over 70 countries. all with great support, simple transparent pricing, and no surprise invoices. Then there's the guarantee. If your client's sales tax isn't filed on time, Numeral pays the penalties and interest. For accounting firms looking for a true partner, Numeral works alongside you to support your clients as they grow.

18:27David Leary: Whether you're implementing it yourself or making a referral, Numeral has you covered. To check them out, head over to accountingpodcast.promo.com numeral. That's accountingpodcast.promo forward slash N-U-M-E-R-A-L. All right. So we've got small accounting firms, solo accountants using Claude, ChatGPT, Cowork. What's the Microsoft one? Copilot, all in so many different ways. And I wanted to highlight a few of these examples from the Zapier blog post. I think it's a great example. So this was on the Zapier blog and the attorney is named Damian Guzman. He's the founder of Oakland-based FTE Legal and he serves small businesses and fintechs.

19:19And he has maintained a self-imposed ceiling

19:24David Leary: of roughly 30 billable hours per week. And he has rebuilt his practice around Claude. and since April of this year, Zapier MCP. And the Zapier MCP is Zapier's AI interface for all of its tools. So if you do Zapier, they have connections to hundreds and hundreds of apps. And the Zapier MCP allows you to access those connections that they've already built through AI. So you can connect Claude to Zapier's MCP, and then it gives Claude access to all of those tools and all the things that you can do with them. And so Guzman has basically built his practice around these apps. And he has estimated that the practice now produces roughly five times as much output, completing equivalent work in 20 % of the former time.

20:23David Leary: And here are the examples in this article of the workflows that are really powerful. He has an hourly scan that has Claude inspect Gmail for actionable client requests, create tasks in Clio, which is his practice management software, and then begin work such as drafting NDAs, reviewing contracts and marking up language. so what's happening is that the emails are coming in to his gmail from clients and then he is claude is automatically creating the tasks in his practice management software and it begins doing the work so it's basically processing those emails and getting started on things before he even looks at it and he's saying that by the time he opens a request roughly half the work is already complete.

21:16David Leary: He's got some other automations that will replace Google Meet Links with Zoom, create a 48-hour preparation buffer, deliver regulatory alerts, monitor dockets, and then remove party names from billing narratives before AI processes them to maintain confidentiality. And he's connecting Clio, Gmail, Google Calendar, Zoom, Docs, Drive, Sheets, Xero, and GitHub through the MCP rather than manually configured workflows. So Guzman has reduced the time it takes him to do the work by a factor of five. And so he is switching to flat rate pricing because if he kept billing hourly, he'd have an 80 % revenue decline.

22:02David Leary: It's so funny to see the legal world facing the same challenge that we faced in client accounting services you know 10 15 20 years ago so i i wanted to highlight this because i think that this is the probably the number one use case right now is and this is something that you can set up now and you can do it without like a ton of work and it's basically using claude to monitor your email inbox and to start processing uh requests right away and you know good examples of that at an accounting firms is like new client inquiries. So you get somebody interested in your services coming into your inbox and you've got Claude monitoring it.

22:50David Leary: It can automatically with say scheduled tasks, you can set up a scheduled task to monitor your inbox. And then if it finds a new client inquiry can start doing the research on that new client, can draft the email response and put it in your draft messages. And then you can review it and send it. And that is something that I've been doing. I have a scheduled task that's like an email sweep. And then it reads all of my emails and it automatically drafts replies based on the type of email it is. And I've basically trained it over time as to what I need it to do. So if it's like scheduling, right?

23:35David Leary: It can go into my Google Calendar, and it can suggest possible times and put those into the email along with a link to book. And I've taught it, I've trained it as to when those times are best, where to find the times. If it's somebody looking for a new consulting project or a new social media type of engagement or somebody interested in appearing on the show, it intelligently can respond in my voice too um and it takes time but you can do this and basically just have to be in there telling it what to do and then it learns like the scheduled task is basically like a saved prompt that it then refines over time and i've got like operating rules and all that so it's really powerful um and i think that's kind of like the best way to get started is using AI as an admin processing emails.

24:36David Leary: The only problem I'm having, David, is that as I've built out this system, the way that it works better is that it has to gather more and more context about my preferences and what I want and what I need. And so it stores all of that information in Markdown files, which are basically just rich text files. saved in my Google Drive. And the challenge I'm having is that every time it loads this scheduled task, it's basically pushing a prompt. The prompt keeps getting longer and longer. And then the operating rules files that it has to load are getting bigger and bigger and bigger. And so the more scheduled tasks I have, the more tokens I'm spending.

25:26David Leary: And so the system -

25:27Blake Oliver: Because every time it has to go and reteach itself from scratch what the rules are to reply to your emails.

25:31David Leary: Yeah, I have to read all those rules and then apply them. And so the bigger and bigger the files get, the more and more tokens I'm spending. And so it's gotten really sophisticated. It's gotten really good. I mean, I've taught, you know, Claude how to write in my style. I've taught it how to do all of these things that I was doing, but now I'm spending tokens in excess of the subscription. So I'm on the$100 a month plan, but I will regularly run out of tokens like halfway through the week and then have to spend. And I've got to figure out how to get this under control because I'm at$2 ,500 for October.

26:11David Leary: So I might spend like$3 ,000 in tokens.

26:15Blake Oliver: And many times you're spending that money to reply to a stupid email that we got from somebody else's AI agent. It's kind of crazy. We're burning cash to reply to other AI agents.

26:27David Leary: But if you think about it, it's like, I've avoided hiring somebody to do a job. The number one use case for me is all these inbound sales inquiries. So now I've got basically, and outbound, So let's say we need to sell more ads on a particular show. I've got a scheduled task, which is basically the sales outbound BDR agent that I've created. And so it's running and drafting messages as me, sending emails as me, handling the responses that I get, booking meetings. and that's allowed us to avoid hiring a person to do that job and so I have to look at that spend and say okay well is it worth it to spend three thousand dollars a month in tokens to avoid hiring a position that would cost who knows seventy eighty thousand dollars a year perhaps and have to manage that person so that's kind of the ROI or cost benefit analysis that I'm doing in my head But I guess what I'm saying is that I'm seeing personally now how token spend can just balloon out of control very easily.

27:54David Leary: And it's hard for me as an individual to manage it. I can't even imagine trying to manage token spend across an organization with hundreds or thousands of people. And so that's where all of these orchestration layers, like you said, to manage this stuff are so important.

28:13Blake Oliver: And this goes like with Ramp. Last week, we talked about how Ramp launched Glass, right? It's going to be this like company-wide AI orchestration layer to where it knows your company knowledge. Everybody uses it to make their AI requests. and it's going to use whatever AI model it needs to so that way you have some control over the spend, right? Yes. And you have control over this like shared knowledge base of the company of how you're responding instead of having it go retrain itself on your Google Doc every single time, right?

28:44David Leary: Yes.

28:45Blake Oliver: So that last week, Ramp launched that. Well, this week, Pilot, they have their Meridian product and they're launching a new product called Meridian Starter. And essentially, it's a shared AI workspace that's going to connect your Cloud Chat GPT directly to QuickBooks Online and Xero for your accounting stuff. So instead of you, everybody at your firm having their own connections with Claude, ChatGPT, into your GLs, and you can't manage that. It's not manageable. The spend's not manageable. The knowledge sharing is not manageable across your thing. So they've built this. Now, what's interesting about this, I think about this more and more, this is what Intuit and Xero should have done or should be doing.

29:25Blake Oliver: they should be that trusted ai layer in between my company's gl and all this crazy ai wild west stuff that's happening be that connection before it gets to the gl and no instead what has zero and two it been doing they've been building stupid collection letter writing agents like think about it it's so stupid it's it really is like into it oh it's gonna create a want me to send a collection letter for you? Jax, Zero Bill Jax, does the same stupid stuff. Like, this doesn't mean they're dead because of AI, right? These companies. But I do think they missed a future product opportunity here. There is a huge new product offering that could exist at companies that it's not only the company knowledge base, but it is kind of like a knowledge level, right?

30:14Blake Oliver: A trusted knowledge level or a trusted knowledge layer that orchestrates the AI, talks to your GL, and Intuit and Xero could have been that piece. They've probably missed the window now for that. It's wide open for somebody else to take, but it would have made sense. Intuit and Xero have to just buy companies that do this, but it could have made a lot of sense for them to offer this and be that trusted layer instead of trying to be the AI.

30:36David Leary: Yeah.

30:37Blake Oliver: Like I could have used any stupid thing to write a collection letter, but what I need is I need a protective layer before data goes into my GL to make sure it's done safely and correctly.

30:45David Leary: Yeah. Well, the big challenge with like creating these, orchestration layers when you're talking about doing repetitive work is that the work for every client is a little bit different. So we're trying to, the problem we're trying to solve is that we want standardization across a firm. So we want to optimize our token spend by creating a process, essentially standardized prompts for how to do a particular task, such as a bank reconciliation or such as like coding transactions or preparing a tax return. And then we want to have everyone in the firm using this system in the same way, but you're sort of like fighting, these needs are at odds, right?

31:41Because the power of the AI agents is that they can do work that is very customized.

31:50David Leary: You tell it what you need. Here's the particular client. Here's what needs to get done. And so as an individual, I can use a Claude or a Cowork or a Copilot or ChatDBT, and I can create these outputs that I couldn't do before. and the AI can understand all the different subtleties of that client. But to do that at scale in a firm...

32:16Blake Oliver: Yeah, so somebody else can handle that client when you're out. It's really hard.

32:20David Leary: It's really hard. And to control spend in that way is really hard. And so whoever can solve that is going to win just an enormous slice of the market for software for accounting firms. because then we can control all of these agents at the firm level. And that's the problem right now. And this is why we haven't seen workforce reductions in firms. It's because you can use these tools now, but they have to be guided by an individual who can be the human in the loop, review the work, oversee the work. So it helps me as a CPA working for my clients go way faster, but it cannot operate without me. There's very few tasks that are completely standard across a whole set of clients.

33:17And so that is why it's this really interesting

33:22David Leary: technological shift because it's making individuals with experience so much more productive, but it's not eliminating any jobs. And so trying to figure out how to get like return on your investment as an accounting firm is challenging because how do we make more money in accounting firms? It's, it's a, it's a, it's by reducing labor costs, right? Generally, but you can't do that. So then the only other way is, and this is the best way right now, if you want to try to like get a return or make money out of AI is to add more services, right? To do stuff for more clients that you couldn't do before, sell more.

34:09David Leary: And that's where like everyone's trying to focus now on doing advisory type services and stuff like that. But that's hard because all of our really talented people are super busy. so yeah because it's literally imagine if you're at an accounting firm you're trying to do a cast

34:27Blake Oliver: practice but nobody none of your employees interacted with each other they just were working away heads down and they never communicated cross-communicated best practices they never shared domain knowledge clients once they moved to somebody else it was like starting over from scratch that's essentially what you're doing with agents right you're creating all these little miniature employees everywhere, but they don't have any context to the bigger company, all the clients, or even that client itself. And there's a huge opportunity here for somebody to go and take this. And honestly, Xero and Intuit probably missed it.

35:02David Leary: And if you want to give them that context, you can, but then the agent has to go and read all that context, which is going to cost of fortune and tokens. Yeah. Right. So, so until we can actually train a model on all of our client data and actually have it, uh, what I would call, remember everything about like the clients and the, and the firm, it's going to be really expensive because it's basically like, imagine, imagine if every time you sat down to do a task, David, if you're an AI agent, the way it works right now, the best parallel I can make is you are starting from scratch. It's like your memory is erased and you have to go and read everything about that client and the whole history that you have to gather all that context and then do the job.

35:54David Leary: And then after you've done that job, your memory gets wiped again. And so the next time you come back, you have to read it all. And that's where the token spend is coming from.

36:02Blake Oliver: And that's where the magic of like Facebook's meta, her muse is at, right? I think Muse keeps like a virtual server spun up that I'm accessing only. And so it just keeps storing all my stuff in its own server. Right. That's why it's so delightful and helpful.

36:18David Leary: Yes, but it's going to cost Facebook a fortune to run that because the more information it gathers about you, it has to keep rereading it over and over again. The difference between these AI systems and humans is that when we learn new things, we go to sleep at night and our brains process all that information and incorporate it into our mental model so that when you wake up you remember all of this stuff right that's like the theory of how like sleep works is that it's actually our our model retraining every night but these ai systems they gather context and they store it but they don't sleep and they don't uh they don't remember.

37:02David Leary: And so that's why you can make these powerful systems, but they get more and more expensive to run. So somebody has to solve this problem, or we're just not going to be able to use these systems to do repetitive work that requires a lot of context. And so that's why going back to this story about this attorney, right, who's using AI to process his inbox, I feel like that is one of the better cost-effective time-saving applications because it doesn't need, that agent doesn't need a ton of context in order to do that function.

37:38Blake Oliver: Because in many cases, you can reply to an email based on the context clues in that email.

37:42David Leary: Yes, like it identifies here is what somebody is asking for. It's administrative, often in nature, and it can go out and it can like do the prep of the email reply without having to go gather a ton of context, read, spend a lot of tokens. And that's where we as accountants get lost is in our email inboxes. So if it can process that, like, I think that is actually the best firm sort of wide way to automate things as opposed to trying to, you know, build like this way to, well, and I guess like with a system like a Digits or these other, you know, AI based general ledgers, they can figure out how to do a lot of these.

38:29manual tasks more effectively, just like categorizing transactions, reconciling bank statements.

38:36Blake Oliver: But the reality is, what firm is 100 % on one of these new AI native ledgers across their whole client base and firm? None.

38:43David Leary: Yeah, none.

38:43Blake Oliver: So nobody really can get these benefits yet.

38:45David Leary: Yeah. But it's like, you have to think about for repetitive work, what are the things that can be done basically the same way across every client every month? But then in order, like you said, in order to get the benefit of that, you really have to have all your clients in one system or it doesn't work. So like there's this, the standardization again, it's the, it's the barrier, just like with humans. If you've got all your clients on a bunch of different GLs and they're all working in a bunch of different ways, just like with humans, you're going to spend, if you have a human team, yeah, if you have a human team, when you onboard a new team member, you're going to spend a lot of time training them how to do everything because it's different for every client, different systems, all that stuff.

39:28David Leary: The same problem occurs with the AI agents. So we can basically take the same lessons from the last 20 years and apply them to this new technology. In order to get the benefit, you have to standardize how your practice does work. And then you can automate this stuff across a bunch of clients. But if they're all different, it's going to be really, really challenging still. The tech is not magical in that respect. Not yet, anyway. And I think it's just going to be really expensive if it's not. So anyway.

40:00Blake Oliver: And it's easy to see a demo or investors, VCs, they see a demo. Look, it did this accounting. Yeah, and it's really easy for this one siloed out instance. But now scale that to 100 clients. It's almost impossible.

40:16David Leary: That's why the demos are so impressive because they're not running at scale. So, you know, and I think like just in terms of like some basic stuff, like for instance, just preparing a tax return, right? Gathering the info that you need from a client, communicating with that client via email or some sort of messaging system and gathering like source documents, creating the work paper, putting the numbers into the return, sending it to review, that's the sort of thing that actually could work really well and not be expensive from a token standpoint. And then from the audit standpoint, it's, you know, once the audit plan has been created, well, creating the audit plan could be a great application of it.

41:07David Leary: And then once it's been created is like, sort of like managing that whole process because there's a standard methodology that could be powerful like doing the sampling right all of that there's a standard methodology that you use across your clients you can you can use ai agents but it all has to be like standardized that's where the the real value is from a firm-wide perspective and then anything you have individuals using AI to do the complex work, but guided by them. So it's sort of like two different, it's almost like two different things that we're talking about. We're saying AI, right? But the firm-wide AI and the individual AI are completely different.

41:54David Leary: We have to approach them in completely different ways. So... David, before we go on, we should thank our next sponsor. Let's do the next two, why don't we? on pay is next right are you tired of the payroll headaches getting in the way of the client experience you want to deliver manual workflows creating bottlenecks compliance nightmares and endless support calls that go nowhere there's a better way for your team and your clients on pay is the payroll partner that accountants and bookkeepers actually love why because it's easy to use packed with value and backed by support that actually supports you their team gets rave reviews for being fast expert, and actually reachable when you need them.

42:37David Leary: OnPay handles the heavy lifting. You get a dedicated onboarding coordinator who sets up worker profiles and transfers year-to-date data from previous providers, all at no extra cost. Their seamless QuickBooks and Xero integrations eliminate manual journal entries, and they support any type of business you serve. Farms, restaurants, nonprofits, you name it. OnPay can handle unique requirements without adding complexity. OnPay keeps pricing simple too. Everything your clients expect from multi-state filing to off-cycle pay runs is included. No hidden fees and no surprises. To book a demo, head over to accountingpodcast.promo slash OnPay.

43:13David Leary: That's accountingpodcast.promo forward slash O-N-P-A-Y.

43:19Blake Oliver: I'll read the next one for you. Are you tired of the endless search for qualified accounting talent? You're not alone. Growing accounting firms are struggling to find available and affordable team members when they need them most. Cloud Accountant Staffing has a solution with a revolutionary Candidate portal. Unlike traditional staffing agencies that waste your time with sales calls, paperwork, and deposits, the Cloud Accountant Staffing portal gives you instant access to highly vetted, qualified accounting professionals. No waiting, no hassle, just top talent right now. What makes this different?

43:51Blake Oliver: Speed and simplicity. While other firms make you wait weeks or months, with cloud account and staffing, you could interview someone as soon as tomorrow. Their boutique support ensures you're getting the quality talent that's both available and affordable, exactly what a growing firm needs. The Candidate Portal puts you in control. You can browse live candidates, make selections on your timeline, build your offshore team without their traditional headaches. To find review and book interviews with potential team members all in less than 10 minutes, head over to accountingpodcast.promo.cast. That is accountingpodcast.promo.com.

44:26David Leary: Thanks to all of our live stream viewers who have joined us on this Sunday morning. I'm surprised how many we've got. Great to see you here. We've got Boring Accountant. We've got, who else? Our attorney, Tate, great to see you. Das Baba Femi. Das Baba Femi says, the AI is magic though. So sometimes it doesn't quite work as we expected. Yes, says, happy Sunday. Sujit Kumar, great to see you too. Thanks for joining us from India.

45:07David Leary: All right, where do we go from here in the few moments that we have left?

45:10Blake Oliver: I guess based on what we were just talking about then, do you think some of these big PE roll-ups that are happening out there that they're in for a surprise? Example, the latest. So Platform Accounting Group, they're one of those industry roll-ups. They're buying accounting firms. The bet is long-term AI. We're going to be able to scale these AI firms and flip them or whatever, right? So Platform Accounting Group just closed a massive$355 million financing round led by Travell Capital Group. They've already acquired 65 acquisitions since 2015 and are serving more than 50 ,000 clients across the US.

45:46Blake Oliver: so i'm assuming this 355 million is just to buy 100 more firms possibly right 50 firms or whatever it's going to cost to do this but do you think like this bet that they're making which is ai's going to make our firm so efficient we're going to be able to do this and scale it like is this a bet that maybe it's going to not pan out for a long time because we've just talked about this this orchestration layer to really do this doesn't exist to scale well i think the The challenge for small firms is embracing all this tech is challenging.

46:23David Leary: How do you, as a small firm owner...

46:28Blake Oliver: Like a solo firm.

46:29David Leary: Yeah, as a solo firm. And the firms that they're acquiring, I understand, like the smaller ones, right? Like a few partners or something like that. And you don't have IT expertise in-house, right? It's just you. And so a group like Platform Accounting Group, it's literally in the name. They can provide the technology platform to all of their member firms, to all of their acquired firms, and then give them that productivity boost, like that solo attorney that we talked about. five times productivity boost because he's automated a lot of the, like the routine work of like just task management, emails, like that takes up so much time.

47:16If you can build that layer and you can deploy it to those firms, then they're going to see

47:22David Leary: a productivity boost. And it might not be 5X. There might not be a, you know, 80 % reduction in labor costs, but even if it's, you know, a little bit right when when labor is you know 40 50 maybe 60 percent of your uh costs in a firm if you can you know reduce that by 10 20 percent you can see a real increase in margins right and the private equity firm only needs to get you know like a you know what 20 profit margin and most of these firms are not as profitable as they could be. So it's like, help these firms raise prices, help them standardize on more modern technology, get productivity gains, help them to fix their fees.

48:15David Leary: If you can do that -

48:16Blake Oliver: If you always just do that, but you're using cloud technology, you're going to make a victory here. Exactly, exactly. So actually, that's the thing, David,

48:25David Leary: is you don't even really need as a private equity roll-up to even be doing that much with AI because you can still just get all the, like there's a lot of firms that never even got the benefit of standardizing onto cloud. And now the cloud tools are so robust and developed after all this time that it's a lot easier to roll them out and standardize them. You know, just getting all these firms on a practice management.

48:52Blake Oliver: You don't have to invent anything new now.

48:54David Leary: Yeah, like a lot of small firms don't even have practice management software. they're just doing everything in email and like excel right and so yeah i mean and there's not a succession plan so the private equity is like the only succession option for a lot of these small firms and i guess the only risk is that they won't be able to keep the people right if the people leave if the staff leave then who's going to serve the clients but if they can like add in the ai stuff and avoid adding headcount or just replace the people that are leaving over time with more automation, I think it's a good bet.

49:35David Leary: And obviously, investors agree and are willing to fund these massive roll-ups. I think the thing that's interesting is the roll-ups where... I think this one is like that, where the firms keep their local branding.

49:50Blake Oliver: That's my understanding too. Yeah, they get to keep your local relationships with your clients. Yeah, you keep your firm brand.

49:58David Leary: You're not becoming, you're not rolling up into like a big firm, like with one brand. And the problem with that is like, as you do that, then you move up market and you have to shed your clients. But these types of private equity roll-ups are really about just combining a bunch of small firms, standardizing them on the same technology stack, but they keep their branding. and they keep their local presence and they don't try to push up market with larger clients. And that's really fascinating. We'll see how that plays out.

50:32Blake Oliver: Yeah. And then another raise that happened, I don't know if you saw Gradient, they just came out of stealth. They raised$4 million of perceived funding to run an AI native accounting firm. So focusing on US startups. So this is where actually these types of startups, they're really at the risk of somebody like a Stripe, right? Like Stripe's already owns, they're solving most of the profit and loss statement now, right? For a startup. I don't know what they're going to, you know, this is a software slash service play. It sounds, feels very bench-like. I don't know if they, we'll see another$4 million at, you know, AIGL accounting as a service startup.

51:13Blake Oliver: We'll see what happens.

51:15David Leary: So I have to talk about the story because it's in the title of this episode. And I totally missed it, totally forgot earlier. So let's hit it. This is about the US economy and how we're doing because it's sort of a tale of, I don't know if this is the right analogy, a tale of two cities. We have the AI boom going on and we also seem to be, like it just feels from other indicators seem to point to a slowdown.

51:53Blake Oliver: The stock market's kind of great over here, but yeah, my bank account maybe isn't as great.

51:59David Leary: Yes. And so I lost the story here. Let me pull it right back up. So I spotted this in CFO.com And the headline is, Minus AI Investments, the U.S. Edges Toward a Recession. An independent economic forecasting firm called Oxford Economics has done some calculations, and they believe that the American economy may have entered recession territory. We might be in a recession if you subtract the massive influx of capital that is flowing into the development of artificial intelligence. Michael Pierce, the US chief economist at Oxford Economics said, all the growth in the past couple of years has been in AI.

52:47David Leary: Everything else has been falling two to three years. So basically, if you subtract AI, if you don't consider the AI category, then we're essentially in a recession. He delivered his remarks at the CFO Leadership Council's fall conference in Colorado last week, where ironically, vendors are hawking a range of AI-enabled products. so that's the the strange thing about this economy right we've got the stock market booming we've got all this investment going into ai but at the same time we've had like a slowdown in other sectors and so what is going to happen when the ai bubble if it is a bubble pops is then we'll be in a real recession is the concern so oxford economics calculated that the US economy expanded about 2 % to 2.5 % over the first half of the year and is likely to expand by something north of 3 % in the third quarter.

53:46David Leary: But you subtract all that, and then we really, you subtract all the AI stuff, and then we haven't really grown at all. And in fact, maybe I've shrunk.

54:00David Leary: So that's basically the crazy situation we're in, right? is that if the AI investment doesn't pay off, if we don't start to see ROI, what is going to happen? And the thing is that it's an interesting situation because you would think all the doom and gloom about AI has been about replacing jobs. And if it was doing that, I think we actually would be going more into a recession, right? Because we're a consumer-driven economy. If we start to get all these layoffs, people are going to not be able to spend as much.

54:34Blake Oliver: People stop getting mortgages, they stop... Home building stops, construction stops, and then it all dies. Yes.

54:42David Leary: But also... And so that hasn't happened, so that's good. But then in order for all of this AI investment in the economy to pay off, there has to be some sort of productivity boost. American productivity has to go up because of these tools. But how is that actually shaping out? like we haven't yet seen it happen in just our sector right in accounting has have we actually started to make more money for our firms as a result of ai yet yeah no companies like all my employees are saving four hours a week because they have less email to deal with now and it's

55:26Blake Oliver: turned into this output at my company we haven't seen that yeah you're not seeing that i mean i

55:30David Leary: think like if the thing I've noticed is like personally is I am working I'm able to work less because I've got the AI doing the manual tasks but now I'm spending a lot more on tokens yeah um have I personally have we at earmark actually seen our like revenue increase as a result of our AI investments maybe a little bit right on like the sales outreach side right we created this

55:59Blake Oliver: you know uh yeah agent yeah might have done one piece of reach out or got a lead yeah so you know

56:06David Leary: and and so like it's increased our sponsorships of the podcast but then if i put that in context of like the overall business i mean it's maybe delivered like a few percent right of increased directly even without it right could have happened anyway so like there's a there's a lot that needs to happen in terms of productivity gains before we can say that ai is like a good investment and so if if we don't see those productivity gains soon you know all these companies spending money on ai might cut back and they cut back and then the market sees that and then uh the these ai companies, like their stock drops, right?

56:52David Leary: The big technology companies. And then they start pulling back on investment. And that is what's been driving the economy, building all the data centers, all that sort of thing.

57:04Blake Oliver: We talked about that, the boot makers.

57:06David Leary: Right. So we have to, in this economy, we have to see actual productivity gains, actual ROI on AI meaningfully to justify these billions or trillions of dollars of investment. And the question is, will we see that soon enough?

57:25Blake Oliver: I'm going to close out on a story that is just crazy. And you can think about it when you're flying on your flight to the conference or people can chew on this. So the Chicago mayor is proposing for the second time now a corporate head tax. So essentially, he's proposing a$25 per month tax on each employee at companies with more than 125 workers, with some exceptions for businesses in poor South and West Side neighborhoods. It's part of the fair share budget he unveiled on Wednesday. So the projected$25 per month per worker is expected to raise$101 million. To me, this is like the most craziest thing.

58:05Blake Oliver: If you're going to hire the most employees in Chicago, shouldn't you be rewarded or recognized? It's like the opposite. like I'm going to hire a bunch of people in Chicago and I'm going to pay more taxes and I'm going to be punished for hiring people. It just doesn't make sense. Like, this is really, like, it's one of the most backwards taxes I could think of. You want to encourage companies to hire people.

58:31David Leary: Wasn't it Chicago or Illinois that was also talking about taxing like AI agents as well?

58:38Blake Oliver: Yeah, because this is a disincentive to hire people.

58:41David Leary: Yeah.

58:42Blake Oliver: Yeah. Yeah.

58:43David Leary: Well, David, David, that's all the time we have for this week. Thanks, everyone who joined us live. Check us out on YouTube if you haven't seen us. We are the Accounting Podcast. Subscribe, hit that notification bell icon, and you will get notified when we are live and you can chat with us. Don't forget, you can earn free continuing professional education for listening to this episode, our back catalog, and many other fine accounting and tax podcasts. Go to earmark.app in your web browser or get the free Earmark app on the App Store. Create your free account. earn one free CPE per week. Subscribe for unlimited.

59:16David Leary: You get unlimited CPE and access to subscriber-only courses. It's just$200 a year. What a deal. David, I'll see you around when I'm back from CCH Connections. I'll talk to you in a couple of days.

59:32Blake Oliver: Enjoy. See you soon.

59:34David Leary: Bye, everyone.

59:42Blake Oliver: Time for the classifieds.

1:00:11Blake Oliver: Try it at tryprofitroot.com.

1:00:17Blake Oliver: Want to get the word out about your newsletter, webinar, party, Facebook group, podcast, ebook, job posting, or that fancy Excel macro you just created? Why not let the listeners of the Cloud Accounting Podcast know by running a classified ad? Hit the show notes for the link to get more info.

1:00:41Thank you.