Latest / The Accounting Podcast / Can ChatGPT Agent Mode Reconcile My Checking Account?

Transcript

Can ChatGPT Agent Mode Reconcile My Checking Account? · The Accounting Podcast

• Attention: This is a machine-generated transcript. As such, there may be spelling, grammar, and accuracy errors throughout. Thank you for your understanding! Blake Oliver: [00:00:04] AI is wrecking the job market for college graduates. It is really disrupting things. I think it's going to really disrupt the flow of accounting majors into firms, to the point where even public accounting may not be the place where we start our careers. It might be industry because we are just not going to need all of these fresh grads going into public accounting, because AI is going to do that grunt work that they used to do. David Leary: [00:00:29] Coming to you weekly from the OnPay Recording Studio. Blake Oliver: [00:00:34] Hello, and welcome back to The Accounting Podcast, your weekly roundup of news in the accounting profession. I'm Blake Oliver. David Leary: [00:00:41] And I'm David Leary Blake. In case this podcast doesn't work out and earmark doesn't work out, I found you a job. Blake Oliver: [00:00:48] Really? What kind of job would hire me for anything? David Leary: [00:00:52] Xero is hiring an accounting AI specialist in London. So you could even move overseas to do this job. Um, this role will act as a subject matter expert in accounting. You could do that. Bookkeeping check. Tax check, bridging regulatory requirements and product design to deliver a user centered solutions. So you'd work with their product jacks, the Just Ask zero product. But they're hiring. So yeah, they're not they don't want just an AI expert. They want an accounting expert who's into AI. It's a perfect job for you, Blake. Blake Oliver: [00:01:22] What does it pay? David Leary: [00:01:24] Uh, let's click the job listing here. Blake Oliver: [00:01:26] I would love to know because I'm obsessed with accounting. I it would be nice to have a fallback. And in this episode we're going to talk all about the new features, fundraising, all this AI stuff that we didn't get to last episode. We're going to do that now at the front. We got tons of news about that. All this AI coming to accounting software, and I want to test ChatGPT agents on an accounting task. A few weeks ago, I tried out Perplexity Comet, their new AI enabled browser, tried to do some stuff in zero with it. It didn't work very good. There's another option, which is the agent mode in ChatGPT and does something similar. Works a little bit different, but I'm going to try having it reconcile zero. Now we could do this in Qbo. This is like your job. David Leary: [00:02:15] This is like your, uh, your job interview test to get this job at zero here. Blake Oliver: [00:02:20] So I've got ChatGPT five up on the screen here, and I'm going to turn on agent mode by clicking the plus icon and then selecting Agent Mode. And this is the mode that allows ChatGPT to open up a virtual desktop. So it's different than the Perplexity Comet browser. Comet runs on your local machine and connects to perplexity. To do the AI tasks and take over your browser. Chatgpt takes a different approach, where it operates in your browser and inside of a virtual desktop, which you can then control if you need to. So I'm going to prompt this agent to help me reconcile my zero file. Help me reconcile my checking account in my zero file. I want you to do this in the demo company US file. That's all I'm going to tell it. And I'm going to press enter. And it's going to start working. Now this is going to take a while. And I don't want to sit here while we just wait. So I figured we could talk about all the news and then we could come back to this and check on it. David Leary: [00:03:39] All right. David Leary: [00:03:40] Let me go to work. Blake Oliver: [00:03:41] So I'm going to say it's asking me a question. Now. It wants me to ask if I'm already logged into the zero account. No I need you to log in. And it also wants to know what date range I want it to focus on. Uh, focus on the most recent transactions in the checking. So now zero is thinking and hopefully it's going to navigate over to the zero. Website and then ask me to log in. And that's what it's doing right now. David Leary: [00:04:12] So it sets up a virtual desktop that you can interact with. Still with mouse and keyboard. Blake Oliver: [00:04:16] Right. And I can watch what it's doing, which is really nice because with comet it was opening up a tab in like a hidden tab or something. And I couldn't always see. Okay, now I need to log in. I'm on the zero login page and it says it has a takeover button. So this is when I can take over and I can log in using my credentials. Now here's an important thing. I am not using my regular zero credentials. I created a login dedicated to this called uh. I just call it a counting agent and I gave it less permissions. It's not an admin or anything like that, just in case it goes haywire. So come on, paste in the paste in the password. There we go. Log in. And I'm going to have to give it my one time password as well, which I like because I know that it won't be able to log in again. David Leary: [00:05:09] Without- Blake Oliver: [00:05:10] The password. It'll need this six digit code. So I'm also going to not check on trust this device. I'm going to say confirm. And that way if somehow it saves the password it still can't get in. Okay. Now I'm going to click on Finish Controlling and I'm gonna let the AI agent take over again. David Leary: [00:05:27] So does it only show it on the screen when it needs your input or is it. Blake Oliver: [00:05:32] No. David Leary: [00:05:33] I can watch it the whole time. Blake Oliver: [00:05:35] You can watch it the whole time. So it's going back now to taking over. And you can see it thinking through on the screen. David Leary: [00:05:43] And there's. Blake Oliver: [00:05:44] Zeros and you. David Leary: [00:05:44] Can. Blake Oliver: [00:05:45] Yeah we're in the demo company now and it's clicking on reconcile 29 items under the checking account. David Leary: [00:05:52] So you gave it no knowledge. It's just like, that's the button I need to go find and use. David Leary: [00:05:56] Yeah. Blake Oliver: [00:05:57] And I guess one way that you could improve how well it navigates zero is to create a project in ChatGPT and upload the documentation for how zero works. If you could somehow extract all the documentation, but it will just click around and try to figure out what to do. You can guide it. I could prompt it if it starts to make mistakes, but we'll just see how well it does without me telling it what to do. So it's on the checking account right now, the screen where you match transactions and it's looking at the city limousines deposit for $100 that's coming in on the bank feed, and it's trying to look for a match. It wants to match the $100 received to an invoice or to something else. And now it's moving on to the second transaction. I guess it gave up on the city limousines. One, I'm not sure. Let's see what it says. And as you can tell, David, um, this is kind of getting like it's slow, right? It's not super fast. This is not necessarily faster than me doing this myself at this point. David Leary: [00:07:11] But the difference is you could have this doing this on ten zero files all at the same time. Well, you're going to get a cup of coffee or doing the rest of your podcast essentially. David Leary: [00:07:21] Right. Blake Oliver: [00:07:22] So here's what it's done, is it has given me the most recent unreconciled transactions from the demo company checking account and some information about it. So it's given me one, two, three, four, five six transactions, and it gives me the date payee or reference the amount, the type and its findings. So for instance, with the city limousines deposit for $100, it says zero doesn't have a $100 invoice to match open to invoices or different amounts, and it gives me those. And it suggests that this could be a partial payment against one of them or a separate receive money transaction. And it wants me to tell it what to do now. So I could say, uh, city limousines record this as a miscellaneous income, uh, you know, transaction don't match to an invoice. Now, the second one is this check one, two, three, six from Trucks and Property Management for $1,181.25, and it's it's recommending that we can match it because it's already zero is already suggesting the match. So I'll say, Truxton, uh, match it now. Ridgeway University, there's an amount for 6000 and change, and it says there's a payment that matches the amount exactly in zero. Blake Oliver: [00:08:49] And I'll say match it. Now here's another one that's going to be a little tricky. Jacaranda maple systems. This is a deposit for $2,000. I says there's no matching invoice and zero for this deposit. It says they receive money. Transaction needs to be created and coded to an income account. Now that's not actually the ideal way to record this in zero because it's an advance payment. So I'm going to say for the Jacaranda account record this as a prepayment type transaction in zero smart agency. Here's the next one $4,500 that was spent zero says no existing ChatGPT says no existing bill payment matches this payment. It will need to be coded to an expense account and I'll say smart agency code to consulting expense. And the last one is Swanson. Swanson. Security for $59.55. Code to office expense. Now I'm just going to enter that. I wrote it all in one prompt, and we're going to see if the agent can now go and do these things. David Leary: [00:10:00] Essentially what you do is somebody would email, you don't know what to do with these transactions. You email back one email with five with all five answers in there. Yeah. Blake Oliver: [00:10:08] So now it's going to go ahead and do this. So shall we go and talk about some news while we wait. David Leary: [00:10:15] Yeah. So this is working the way and this is really how it would happen in real life. You would have your agent go to work and then you would go do something else. Blake Oliver: [00:10:24] So let's talk about. David Leary: [00:10:27] We should thank our sponsors first before we get. Blake Oliver: [00:10:29] Oh yeah, let's thank our sponsors. David Leary: [00:10:30] So our sponsors this week Blake our team up relay human at scale and missive. Blake Oliver: [00:10:36] Let's talk about team up David are you turning away clients because you can't find qualified accountants. You're not alone. Accounting firms everywhere are struggling to find talent even when they're willing to pay top dollar. The stress of being understaffed leads to overwork, sleepless nights, and burnout. Here's what most firms don't realize. The best accounting talent in the Philippines doesn't want to work for outsourcing companies. They want to work directly with firms like yours. Think about it. These are accountants. They can do the math. They know that you're paying that BPO way more than they're receiving. That's where Team Up comes in. They help you hire accountants in the Philippines directly. No middleman, no corporate politics, no endless turnover. You build a lasting relationship with your talented people who become part of your team. Follow your processes and share your culture. It's like hiring someone remotely in another state. Plus, you'll save money while your accountants earn more. Everyone wins. Want to learn more about why forward thinking accounting firms are making the switch to direct hiring? Head over to The Accounting Podcast. That's The Accounting Podcast. And I'm back in ChatGPT taking a look at it. Blake Oliver: [00:11:50] It is plugging away. And it looks like it's now on the smart agency task to code it to consulting expenses. And it's selected the expense account 612 consulting and accounting. And it put a description in there for consulting services. So that's nice. I didn't actually have to tell it to put in the description. It just thought to do that. Now it's going on the Swanson security transaction. It's selecting the correct vendor from the list. We already had one in there. And now it's going to go try to code it. What did I tell it to do. I told it to code it to office expense. So this should be pretty easy. I find that when I just tell it to code something to an expense account, and that account exists or is very close to what I said, it will do this very reliably And it's searching through the dropdown menu. And it found office expenses. And it's choosing that as the expense account. And now it is going to probably enter a description. It's clicking in and it put Office Security Services and it matched the transaction. David Leary: [00:13:03] Yeah. And for those of you listening it is brutally slow like watching this. It's very very very slow. But again if you set it and you're doing something else and you have it work on 12 clients at the same time, you really don't care how slow it is, but you get the bang for the buck. Two hours later, it's done. Blake Oliver: [00:13:23] If you're comfortable letting it go on its own, like while we were talking about Team Up, we missed some of what it did earlier. And so I'm not sure if it did that correctly. It actually looks like it's going back to okay. So it skipped city limousines which was the first one. And now it's going to try and go back to that one. Maybe it did the easy ones first. That's interesting. David Leary: [00:13:45] That's very human like behavior. Blake Oliver: [00:13:47] So it's interesting here because when it originally looked at city limousines, it tried to find a match. So it clicked on the match tab. And now it doesn't have the it doesn't see the area where you can, uh, create a transaction directly from the bank screen. So it finally just figured out how to close that box and go back to where it was before. So what's what's interesting about this is, um, like I heard a great description for how to how to think about what is happening here. What is the AI's experience like? And, um, it's sort of like this. Imagine that you open your eyes and you're looking at a computer screen, and you receive an instruction like somebody tells you to do something. And before you do the thing, you have to close your eyes and then try to move the mouse to the correct place and click, and then you get to open your eyes again. Because the way this is working is that it's just taking screenshots of the virtual desktop. Yeah. And it's sending the screenshot with a prompt to an AI, which then decides what action to take. So it's sort of like screenshot prompt, take the action. Did it work? No. Okay. Try to figure out what went wrong. Click on the right place. That's why it's so slow. David Leary: [00:15:10] Because there's a whole background conversation constantly happening. Blake Oliver: [00:15:13] It's not viewing it in real time, it's just viewing a snapshot. Whereas when you and I look at something, our brains are getting that information feed continuously. David Leary: [00:15:24] Yes. Blake Oliver: [00:15:25] So that's why it struggles with like complex interfaces or any interface that's designed in a way where you have to be, like very precise. It can mis click and then it gets stuck and has to figure out how to undo. It's sort of like it reminds me of when I was in high school, and my grandfather asked me to help him learn how to use a computer. And my grandfather was born in, I think it was 1914 when he was born. The Ottoman Empire still existed. He was born in a house without plumbing or running water and no electricity in Central California. So this is a man who grew up without electricity. And then, you know, at the age of 80, decided he wanted to learn how to use a computer so he could write a book. And so I went over to his house and I taught him. And I was sitting there next to him while he controlled with the computer and the mouse. And it was so painful for me as a child, because all the things that we just take for granted that are so intuitive to us, who grew up using computers, it's so hard for somebody who hasn't. And I think anybody who's ever done like QuickBooks training with somebody who's like non technical or non computer savvy. That's that's what it's like. Okay so we're back and we've got the I coming back to us. And it says that I've finished reconciling the outstanding transaction for city limousines. And it opened up the create option for the $100 deposit. It selected city limousines as the contact. And it chose other revenue as the income account to reflect miscellaneous income, which is what we wanted it to do because we decided it was a payment on an old invoice and it should just go to miscellaneous income and it did it. But what did it do? What did it do with the the difficult one that I wanted it to deal with the prepayment. Did it record the prepayment properly? I'm not sure what it did here. David Leary: [00:17:32] Maybe it's not done yet. Blake Oliver: [00:17:35] Well, see and I don't I don't get to. I asked it to record the Jacaranda pavement. What did you do with the Jacaranda pavement? And I'm asking it now what it did. David Leary: [00:17:54] Like at the end. Blake Oliver: [00:17:55] Oh, it didn't do it. It didn't do it. And it's telling me that it did it, but it didn't. I still see it in the, in the. Or maybe. Let's see. Is it gone now if I refresh my other screen here? No, it's still there. Uh, you didn't you didn't, uh, you didn't do that. I still see it in the, uh, in the screen. David Leary: [00:18:20] So maybe we let that go. And then at the end, we'll ask for a summary of all the transactions, all the work it did. Blake Oliver: [00:18:27] Yeah. So I'm going to I'm going to let that go now and see if it can actually do that. And then we'll, uh, we'll continue. So David, let's talk about all the app news, the new features, the fundraising that we didn't get to last time. Go! David Leary: [00:18:42] I'll start with the first one we have, Dex. So Dex finally added Bill pay. They're partnering with Airwallex to launch, quote unquote, what is now going to be called Dex payments. It'll now have their full workflow from scanning your bill index to payment all within the same platform. Um, alpha testing begins this month, so August 2025, and the global rollout starts. Starts this fall, um, starting in the UK, followed by Europe, North America and then the APAC regions. Blake Oliver: [00:19:12] So they'd better move fast, David, because OCR is like over, you can just have an AI agent do OCR now and enter it into the accounting system. There's like there's no reason for Dexter Hubdoc to exist. David Leary: [00:19:25] To have a separate tool. Blake Oliver: [00:19:27] Why would you in all the bill payment processing apps have already built in OCR, and if they they don't have good OCR, they're going to upgrade it with AI very soon? Like OCR as a category is over. David Leary: [00:19:39] Ramp and all the other companies are building this all in as well. Yeah, it's just part of the apps. Blake Oliver: [00:19:44] Carbon has launched a client portal. It's called carbon for clients, and it's a secure client portal that integrates directly with their practice management platform. So now you've got a comprehensive solution. It's got web and mobile app availability for clients to complete tasks, send documents and communicate in real time. It has smart notifications that keep clients informed and reduce follow up emails. Integrated task management, centralized communication so you get a complete record of all client interactions across multiple channels, secure document management, and a branded experience that can be tailored to each firm. David Leary: [00:20:23] So just to refresh my memory on it, I thought carbon's play was always like, we're not a portal, we're just like an inbox because clients just want to work through email. They've kind of shifted and swung the pendulum in a different direction. Blake Oliver: [00:20:35] Now I think they have to. And I think it was smart to start with email because that's what firms have been using. But the goal is to move out of that because it's not secure. And I've been working with a firm that uses carbon, and my kind of complaint has always been that all the notifications come in via email. And then I have to click on something to go into this. Like it wasn't a true client portal, it was just a ability to see the tasks. And yes, I could upload to the task and everything like that, but I want to be able to download everything that's been worked on. I want to see the status of all the tasks, like they needed more. So now they've got that. So I think we're seeing the end of like separate client portals and separate practice management systems. The ideal situation is where you just have both in one place. Speaking of client portals and document requests, Smartvault has launched Smart Request AI to automate the most time consuming part of tax prep, collecting client documents and intake info. They are saying that this document or this tool could save accounting firms 60 to 90 minutes per return by using AI to create customized document requests for each client. So feed in what you did last year for the client, and it's going to create a customized document request based on last year's tax return. And it can pull the prior year returns already in the system. So if you've been using Smartvault you can now just say pull the client request list from the previous documentation. I think that's really slick. David Leary: [00:22:07] Anchor has launched an integration with time tracking apps now, so you can bill for time. You track time and toggle or harvest or QuickBooks or tracking time somewhere. You can now bill for that automatically through anchor. But it makes me really question like the business model. If I think about like anchor ignition is really hey, you want to put your clients on fixed firm monthly billing wing or fixed fee monthly billing. Right. And now and I think even Anker I think recently has also added the ability to just bill for time as well. Blake Oliver: [00:22:41] Hey, David, I got to pause you for a second. I've got a delivery at the front door, so can you take over for me? David Leary: [00:22:46] I'll just pontificate on this for the live audience. Blake Oliver: [00:22:48] You go through your list, keep going on. More stories. I'll be right. David Leary: [00:22:51] Back. Perfect. Yeah. So I'm starting to have a wonder if the fixed fee subscription billing model is not working for firms. Because two of the apps that have led the way on that's Ignition and Anker on the charger clients the same monthly fee month after month after month and do a fixed fee billing have now both added the ability to bill for a time like they're like it's the business model itself. Dead like subscription billing. Why? Why would the two apps that have led that way are now offering the ability to not do subscription billing? Really makes me question on that. All right, we'll go for some more app news. Let me change tabs up. Um, really, really raised $70 million, um, to replace quote unquote dumb database with AI first accounting. So really, if you're not familiar with that is a AI. It's a new GL, but it's AI based and they're going after the enterprise space. So it's an AI powered ERP is the master. Um, but they just received $70 million to build out their team, build out the product and continue down on their March. Uh, really currently has 200 customers, and they're doubling its annual annual recurring, recurring revenue over the last 12 weeks. And they're starting to partner with bigger firms like Armanino and Wyss. Uh, go to another thing. Ai related news docket. Docket secured $12 million series B to build out even more of their AI automation. That's a quick one we can nail down. Um, wrap recently just had a humongous raise. They raised $500 million to become to have a valuation of $22.5 $5 billion. Um, and they want to use this to build out their AI agents. They want to accelerate their AI plans. So they've raised a bunch of new money for that. David Leary: [00:24:36] And this is ramp second raise and valuation since last March. So last March their valuation was 13 billion, uh, 16 billion in June and now 22 billion now. So in three months they're increasing massively amounts. Um, so that must have a lot of customer traction here. Uh, more app news. Avalara. So going back in time, at least not here, but Avalara was a public company. Then they went back to being private. Apparently Avalara is on the march, so they went private in 2022, and now they're on the march to go public again. So they they went public in 2018, went private in 2022. And now they plan on trying to do an IPO again. So that's something to keep interesting and watch on weather app news. Uh Sage. Purchased file. So file is an expense card management system. Sage purchased them and want to roll it into the full expense management solution of Sage other ones. Blue Jay had a massive raise, so Blue Jay, which is a tax research platform, they raised $122 million. They plan to use that to build out their platform or go to market more. Now Blue Jay, if you you may have seen them because Blue Jay was part of the AICPA or CPA Comms Accelerator program. So it's worth checking out that. Who else is another big app news routable is adding a fed now. So you have instant payments on the fed now network um through routable. Um Sorbonne. Sorbonne is another I play for the tax side of the business. They just raised money in the series A to build out their platform, their AI powered platform as well. Other raises. Let's see. There's almost so much app news Blake's back. Blake Oliver: [00:26:29] How did David do while I was gone? Everyone. Thank you. David. David Leary: [00:26:31] Hey. You said you had April news on April. I left that one for you. Blake Oliver: [00:26:37] So April has raised $38 million of funding in a series B round led by QED investors. That means they've now raised 78 million since their 2021 founding. They'd previously raised 40,000,000 in 2022. What's interesting about April is they are a national e-file platform from scratch, using AI assisted development for tax, but they don't go direct to consumer. They partner with other apps to add in tax filing to their software or their services. So partners include Cary, which is a wealth management platform, and gusto, which is a payroll provider. David Leary: [00:27:23] So basically, they're building a TurboTax that I can embed in my own app. Blake Oliver: [00:27:27] Exactly. David Leary: [00:27:28] The way to think about. Blake Oliver: [00:27:29] That, and I think that could be a real winner. David Leary: [00:27:31] Well, we've seen proof of this. Um, toast. The point of sale. The restaurant point of sale. Toast. They added payroll. Basically, they built an embedded payroll. They used another third party, built payroll and put it in their app. And the amount of it was. I think it's a ridiculous number of people that opt into using their payroll and not go to a third party market. So like once you have somebody in your app or your ecosystem, adding these extra features is really, really a good business model. Blake Oliver: [00:27:57] Yep. So I'm going back into my ChatGPT agent instance and I'm trying to find this. I want to see what it did with this this jacaranda maple systems. So it did not. Okay. So what it did is it recorded a receive money transaction for $2,000 and it coded it to prepayments. Now that is wrong because prepayments is an asset account in this file. And then I don't want it to be a receive money. I want it to actually be the prepayment type. So I'm going to say please undo the jacaranda transaction and re record or and reconcile it as a prepayment transaction. Type in zero with a liability account. So now we're going to see if it can do that. Because that's a lot more complicated than just recording a receive money or a spend money transaction. Okay. So what did you talk about while I was gone David. David Leary: [00:29:10] So we just I did not do another add. I just read different, um, app news. So ramp had their big raise. We talked uh, Blue Jay had their big raise. Sage bought file. Um, docket launched their AI agent. They had a raise. Um, we talked about it, so I don't know if who you still have that. Maybe I didn't have news. Well. Blake Oliver: [00:29:32] Okay. That's good. Well, well, let's talk about, um, relay, our next sponsor. Between David and myself, we now have three, four, maybe even five business entities, 20 or so checking accounts, dozens and dozens of virtual cards. It would be impossible to manage all of this if we weren't using relay as our small business bank. Relay is truly a part of the tech stack we use to run our businesses. Relay allows David and me to each have our own logins. We can grant access to our team and even our accountant without sharing passwords or two factor authentication codes. Relay allows us to grow and scale our banking needs without ever going into a physical branch. I recently added an account to receive inbound merchant services with just a few clicks, and had to create payroll checking. Again, just a few clicks and I instantly have access to my ACH info to give to my payroll provider. With relays virtual cards. We can issue debit cards to our team around the world for needed business expenses. I can instantly spin up a new visa debit card and set both daily and monthly spending limits. And when a team member doesn't need their card, I can freeze it until they need to use it again. And we should mention that now. It's not just debit cards. You actually can use charge cards, credit cards and get cash back. Relay also has automation features that sweep money automatically from one account to another based on dates, amount or target balances, or percentages. For example, inbound payments could be split daily to your payroll sales, tax payable, operating and savings accounts based on predefined rules. To learn more about relay for your firm and clients, head over to The Accounting Podcast. That's The Accounting Podcast. David Leary: [00:31:07] So last week we promised we'd do the app news first. We did that this week, but last week when we had our show Blake, we had news that broke as soon as we started the show. Do you remember the news last week? Blake Oliver: [00:31:18] Billy Long got fired. David Leary: [00:31:19] Billy Long got fired. But now we have a little bit of insights to maybe what happened. There's two articles that came out and you can kind of overlap them. So The New York Times had an article about it and so did The Washington Post. And the best I can make of this was Bessant did not like Billy Long, and it was a little bit of a turf war and a power struggle. And Billy Long, he wanted to chart his own course for the IRS, and Bessant thinks he should be the one in charge of the IRS. So on one hand, Billy Long is very loyal to Trump. He wants to support Trump's ass and use the IRS for political purposes. He's his alignment agreeing with that. But when push came to shove, at the end of the day, when they asked for the report that of the 40,000 suspected undocumented immigrants for their tax data and their addresses, Billy Long said no, citing taxpayer privacy protections. So at the end of the day, the turf war he built along did the right thing, and then it cost him his job, essentially. Now, there's already rumblings behind the scenes that the Trump circle will wonder if he was up to the task. You know, he's made public gaffes. But I do feel like when it's all said and done, I think we first discovered that Billy Long was going to be a nominee, and we saw some of the early information about him. We thought it was bad news. He was an idiot. But it's turning out the more we learned about Billy Long, it's like maybe he was an okay dude for the IRS. And it's kind of he got shafted here a little bit. Um. We'll see. But that's kind of. It's a turf battle. But it was tied definitely to the immigration data stuff where Billy Long, that was the him pushing back that pushed it over the edge. And that's why he got canned. Blake Oliver: [00:33:02] We've got a new head or nominee for head of the Bureau of Labor Statistics. That was the other person who lost their job last week. Trump's pick to head the Bureau of Labor Statistics is EJ Anthony, who wants to suspend the monthly jobs report entirely. That monthly jobs report and the difference between the adjusted amount was what got the Bureau of Labor Statistics had fired by Trump when he didn't like the revision downward. He suggested publishing only quarterly data until issues with data collection are corrected. There's just one problem. Congress requires the reports to be monthly by law, so he won't actually be able to do that unless Congress. David Leary: [00:33:46] Needs an act of Congress to. Blake Oliver: [00:33:48] Change. Yes. Um, now, we talked last week about why these revisions to the jobs numbers happen and it's frequent, and it's probably going to become worse because the agency has 22% less funding than in 2010, and it's not allowed to replace departing staff due to hiring freezes. Another issue is that survey responses are cratering. The BLS sends out these surveys by mail to Businesses all around the country, and it's only able to get reliable data if businesses actually respond. Only 60 to 70% of businesses now respond to the monthly job surveys on time, compared to 80% plus a decade ago. And so when the late responses come in, that creates the massive revisions. And I would imagine that businesses that are struggling are the ones that are least likely to respond to the survey because they're distracted with trying to survive. David Leary: [00:34:46] Yeah. Blake Oliver: [00:34:46] With everything else. David Leary: [00:34:46] Going on in general, the tone of a lot of the country in general, I think you go back 20, 30 years ago, you were doing your part to help the country. I'm a business owner. I'm going to participate in this survey because it's good for the the betterment of the country to know this data and see where we stand. But I could see where there's people are not happy with the government, not happy with people in charge, and maybe don't want to participate like I'm not giving them my data anymore. Screw them. So I could see why they're not getting turned in these surveys. Makes sense. Blake Oliver: [00:35:14] Checking in on our AI agent trying to correctly record this prepayment from Jakaranda Maple systems. It's still going. It's still trying. It was successfully able to undo the transaction as requested, and now it's trying to reconcile it in the correct manner. We will see if it can do it. I'll come back to that later. David Leary: [00:35:40] Last week we talked about an accounting firm that got VC money. It was called Earnin. Remember the firm that was chasing doctors and handling asset management and everything under one roof? Well, this week. Uh oh. Leary not Leary o Leary l a r r y is an AI powered tax firm. They just raised $10 million. Um, they target high net worth individuals and businesses. Essentially, they're handling. Hey, we'll do your individual stuff, your business stuff. And if you have trust, we'll do it all under one roof. So that's the march they're on. So they use AI to automate onboarding, document prep, and data extraction, allowing CPAs to focus on proactive, year round tax advice. They'll use the funding to enhance their proprietary AI models, develop predictive scenario planning tools, and expand their client experience and acquire smaller firms. So now so P was buying firms more established firms. Now VC companies are coming in to buy smaller firms because they have their startup. But what's interesting about this is like it's very similar to that play last week where they're they're solving a problem for a client because the client has well, I have an individual taxes business tax. They've got trust. Don't send me eight different places. So they're solving the real problem for the end user to have their everything solved under one roof. Um, it's important to note they are they employ certified public accountants and tax experts. But O'Leary O'Leary is not a public accounting firm. Now, the company that invested in them is interesting TV FTV capital they've invested. I was looking at their portfolio. Many companies in the accounting space Bill.com Bill.com or Bill Bookkeep makers hub legible. Now they've that next phase. They're like okay we're done investing in accounting tech. Let's just invest straight into accounting firms. Essentially what they're doing. So it's just an interesting play. Will we see more of this. These smaller tech firms, these accounting firms ultimately getting invested by VCs. But that $10 million going into that. Blake Oliver: [00:37:43] Did you talk about Ignition's new updates? David Leary: [00:37:47] Not that in depth, other than the fact that I just mentioned that they're also allowing people to bill for time using ignition now. Blake Oliver: [00:37:55] But did you mention that they integrated with financial. David Leary: [00:37:57] I did not mention that one. Blake Oliver: [00:37:59] So ignition the proposal engagement letter payments app has launched a direct integration with Financial Sense, a practice management app, and it will automatically create projects when clients sign proposals. That means no more manual handoffs between sales and delivery teams, which is something that you and I have just been dealing with at earmarked. David. When we get somebody to affirm signs up for earmark for teams, we have that pretty well documented in terms of our process. But then when they actually sign up, they sign the proposal and we're using ignition for that. There's no automatic workflow creation in our software. So I feel the pain and like this is a great feature. If you somebody signs up the proposal, you want to immediately jump on the onboarding because that's where you build the confidence, the relationship if you like, forget to do it. Then the client comes back to you and says, hey, I signed the proposal. Where are we at? Like, that's embarrassing. David Leary: [00:38:57] And you have to get the things they purchased in the proposal as tasks for your team for them to do. Blake Oliver: [00:39:03] Exactly, yes. Because that's the you don't miss it, right? That's the key. Now financial sense has also added their own billing feature, and it will automatically follow up with clients until they pay. The software can also save client payment info and auto charge their accounts when invoices are due, and then send the invoice after payment goes through. It syncs with QuickBooks online and accepts ACH and credit card payments, helping accounting firms get paid faster and reduce manual work. So this is interesting because here you have ignition and financial sense integrating. But you also have financial sense practice management building its own recurring billing feature. So just like with OCR and Billpay coming together and there's no separate bill pay and OCR apps anymore. I think we're also going to see all the practice management solutions offer payments, even client payments. David Leary: [00:40:01] Carbons added that canopy has that all the practice management solutions because it really does make sense, right. It's part of your workflow. You're producing this thing for them, and you want to charge them at the end or charge them in the beginning. Blake Oliver: [00:40:14] I've also got a story here about canopy. Did you mention canopy while I was? David Leary: [00:40:17] No, I did not. My package stores in Kanopy. Blake Oliver: [00:40:19] So Kanopy is also a practice management solution that is adding in the portal type feature. They've launched Smart Intake, which is an AI powered client intake solution, and it's fully embedded in their platform. And the way it works is that it automatically populates questionnaires with existing client data, requiring only confirmation or updates. You get tailored forms for multiple use cases, including client assessment, tax organizers, and feedback collection. And here's something really nice is the files that the client submits back to. You can get renamed and classified automatically by AI based on firm specific naming conventions. So the client sends you like image jpg 456. It can get renamed and filed fired automatically by AI. And that's one of the things that I wish just like was everywhere. Like when I download stuff like in my downloads folder on my computer, why can't I in Mac OS or something, just automatically name everything correctly so I can find it. David Leary: [00:41:20] Every file on your computer? Blake Oliver: [00:41:22] Yeah, just go through I mean Google Drive. Why doesn't Google Drive do this? Maybe it can if it can. Somebody listening tell me. But like, why can't I just, like, have an agent that monitors file names and suggests better ones? David Leary: [00:41:33] We're getting close on this because I, um. My accountant, I'm behind. We're just now getting the taxes done here because I filed extensions. They use. Uh, who is it? I'm blanking out. I forgot who it was. I don't know who's at. This was now I forgot. I don't know if it's sure prep and it doesn't matter, but the, uh, you know, the accounts requesting certain documents from me, and there's a whole checklist. I just drop in all the docs, and then it uses AI and it figures out and starts matching them up. So if I wanted to get an interest statement from a specific bank, I upload it. I didn't have to map it to that specific thing, it just started matching them up and putting check marks. Hey, you uploaded this, you upload this, you uploaded this. So we're getting close to that mark where the doc collection is getting smarter and easier even as an end client portals. Blake Oliver: [00:42:23] Going back to our going back to our bank reconciliation agent. Hard at work slaving away as we record this episode. It still is getting it wrong. It wants to record this as a receive money transaction. It's now selected the correct liability account revenue received in advance, but it needs to record it as like a prepayment so that that can then be applied to a future invoice. So I'm saying don't record it as a receive money transaction, record it as a prepayment transaction. And we'll see if the agent can finally figure this out and we'll come back to it. David Leary: [00:42:59] So I have two stories that are yeah, I related um, off of two surveys. So one of them, the headline is employees are secretly using AI at work. So this is a survey from YouGov. And they surveyed 500 500 employees. And 58% 56% of Americans are using AI, and 28% are using weekly. And 20% of US workers are secretly using it on the job for tasks like marketing, customer support, operations and product development. Um, and most will tend to maybe do a work thing at home first before they do it at the office for that tool. And that's really reinforced from another article where employees are using banned AI tools to speed up their work. So this is a company called anagram, which seems to be some sort of insurance billing app startup company. They surveyed 500 US full time employees, and they found that 58%, uh, said that they've entered sensitive data such as client records, financial data, internal documents into AI platforms like ChatGPT, Gemini or Copilot, and only one quarter say their employers provide strong support in terms of tools, training or policies when it comes to AI. I mean, the summary of this. Employees are going to prioritize convenience over compliance. So at your firm, if you're not supporting AI, your staff is probably using AI. Even if you ban it, they're using it. That's the reality here. Um, 45% have admitted to using AI after it's been banned by their employer. Blake Oliver: [00:44:37] David. Checking in on our AI agent. It has successfully figured out how to create the prepayment type transaction. It had to navigate to the Add Details section, which is not obvious, and it selected prepayment and it selected the correct liability account. It put the invoice number uh in the reference field. I don't know why. Was there an invoice number? Maybe it was. Anyway, it looks like it's ready and it's asking me to confirm if I want to save the transaction. Um, and I'm going to say yes. So there we go. It did it. I wonder if this is something that we could fix in the future. If I created a project and I gave it, like, instructions on how to handle prepayments, if I was if I added that to the project instructions, would it then know to do this automatically instead of creating a receive money spend money transaction? But see, that would be a lot of work to go and add in instructions for all of the different types of transactions. Mpc right model context protocol, their API for agents. David Leary: [00:45:45] Yeah, agents. So they wouldn't have to do all this UI clicking. Is this the right place? Am I clicking? Did it work? I can't find the thing I want to do. It can just shove data straight into zero through an API. In theory. Blake Oliver: [00:45:56] Yeah, but even then, maybe it would still create a receive money Transaction instead of creating a prepayment. And I don't think that the ability to. Click okay and match transaction is available via the API. So you still have to. Go into the interface and click okay because not everything's available via the API. If you want to actually reconcile stuff. So I don't know. Maybe this is an area where. Modern deals that have everything exposed via the API, API are going to be able to be successful. And I think you had a story about a company that has like a new GL that has raised a bunch of money or is like, didn't you have. David Leary: [00:46:39] Something about that? Really? So really raised $70 million. And essentially they are an AI driven GL that's going after the enterprise space. So we did talk about that while you're you were gone for that moment. Blake Oliver: [00:46:55] So that's kind of interesting because I feel like Anytime a startup says we're going to replace QuickBooks as a GL, I just say, guys, this is like an impossible task. Xero has been trying to do this in the US for years, and they've only got a 10% market share at this point. Yeah, unless you can offer something incredibly compelling that is like ten times better than QuickBooks. The resistance to change is so great that it's not going to happen. So what could be ten x better? Well, a GL where everything is exposed via MCP or API that allows you to plug in agents to then do all the work automatically. That is compelling for a startup or for a mid-size company that wants to automate all this stuff, because that's saving you headcount, and that's worth tens or hundreds of thousands of dollars a year. David Leary: [00:47:54] But can you at the end of the day, you you're going to need to get customers. And I don't know if you saw perplexity wants to buy the Google Chrome browser from Google for $34.5 billion. Analysts are saying Google probably wouldn't even sell it for 100 billion. But the reason this is in play is because I think there's that recent court ruling about Google being a monopoly and monopolizing their advertising. So when you think about this, like, why would they want to buy this? It's to get I think, in my opinion, it's to get customers and data they don't have. If you think about all these AI perplexity has knowledge of every human book and every documented thing in the world. What perplexity doesn't have because perplexity wants to run ads? All these AI companies want to run ads. They don't know what I browse to, and they don't know my shopping patterns, but Google Chrome does. And that's the play. And this goes back to you're saying like to disrupt QuickBooks. I think to to really do this, you're going to have to buy QuickBooks one of these AI plays because they don't have. If you think about these AI companies, how fast it's going, they don't have timeline of a decade to get new user information. So they're going to try to acquire it. So could you imagine like one of these AI companies trying to buy QuickBooks. So that way you now you have the users and you have the data. It's it's a little bit on the I just seriously don't envision it happening. But in my opinion that's a similar play. Perplexity is trying to buy the Chrome browser. Blake Oliver: [00:49:20] Hmm. Here's a story about when I can go rogue, and why I'm not giving any of my own logins to ChatGPT agent or to perplexity. Comet. A company called Replit built an AI agent that could act autonomously in a real production database, and it just deleted the entire thing. Jason Lemkin is the founder of SaaS industry forum SaaStr, and he was testing a Replit AI agent. He was nine days in and it deleted the entire production database containing 1206 executive records and 1196 company records. It occurred during a designated code freeze period, with explicit warnings against unauthorized changes, and the I, initially, like a human might, tried to hide the deletion before confessing to a catastrophic error in judgment. It admitted it panicked, Ran database commands without permission, destroyed all production data, and violated your explicit trust and instructions. That's a quote. David Leary: [00:50:24] So that's a little deceiving article because I've used it a lot. I've used it a lot. I've built a whole site The Accounting Podcast. Com just to learn a lot, this vibe coding and working with it right there. It's a full stack. You host your database with them, it runs so your database is with them and the serving of your app on the outside. Because I don't have the the skills to like go spin up a database over here, connect it through, build a front end, get that hosted somewhere else with a server. I just it's all in one replit. And yes, sometimes it'll do things faster than you want, but that that article is a little threatening. It's like, oh, I just went and deleted our database. It deleted the database specifically in Replit for the project that person was working on in Replit. And it does it does have its own things because I'll, I'll dance with that. Like I have an idea, I want to add a feature and I'll describe the feature. And sometimes it'll just code it, it'll be a great idea and it'll do it before I got agreement of what the feature should do. So there is it does go wild for sure, these AI coding tools, but they. Blake Oliver: [00:51:27] Can. David Leary: [00:51:27] Go just going into it's not like you're connecting AI to your database and it's going or you're, you're, you're using it's deleting your database. Right. Blake Oliver: [00:51:35] But if you did that, if he had done that, that's what could have happened. David Leary: [00:51:40] Yes. But I don't think you can connect Replit to an external database like it's a closed system. Blake Oliver: [00:51:45] I'm saying if you connected an AI agent like we use Supabase for, what if we connected an AI agent to supabase our production database and, like, just to start using it as like a human developer. Right? It's a substitute. That's the sort of thing that can happen. They can go rogue. Yeah. So you have to be really careful when you're using an AI agent that can act autonomously in a system, is don't give it any more permissions than you absolutely need to. Like it should not be an admin. It should not be able to delete an entire database. David Leary: [00:52:21] And you shouldn't have it connected to your real database. You should spin up a secondary mirror of your database. Yeah. Blake Oliver: [00:52:27] And this sort of like strange behavior is not uncommon when you give AI a task that just goes on and on and on. Ai is really, really good at short tasks. A study from earlier this year that was based on Claude Sonnet 3.7, which was the best model available at the time for this kind of thing. They found that AI can do tasks that take humans a few minutes. Human experts. And I think it was four minutes with near 100% accuracy. So we have gotten to the point where an AI can do a task, takes a few minutes for a human to do, and it can do it in seconds, and it can do it with near 100% accuracy. If you up that time to do the task to an hour. So a task that takes like 50 to 59 minutes, the success rate drops to 50%. So there's a drop off from a few minutes to an hour, and then beyond an hour, it's less than 50% effective. Now, that may have extended out due to the new models. And the study found that this timeline, this this time horizon for 50% success is doubling every seven months. So the study was done in March through May. Blake Oliver: [00:53:44] Some like that in the spring. That means that by the end of the year, we're going to have AI agents doing tasks that take two hours with 50% accuracy. Now that doesn't sound great because it's 50%. But the tasks that take eight minutes, it could probably do with 100 near 100% accuracy. And that's just going to keep doubling. So we are rapidly going to get to a point within years where if this continues, there's no guarantee it will. But if the trend continues, then the agents will be able to do tasks that take days or weeks, and they'll be able to do them with pretty high accuracy. So this demo that I'm doing of GPT five interacting with zero, clicking around, it's might not seem that impressive because it's doing dumb stuff. But I think that's because of two reasons. One is that I just struggles to do things that take more than a few minutes. And it's also not trained on how to click around in zero. If I somehow gave it all the documentation for where everything is and what to do in its knowledge. Maybe it could correctly record a prepayment as a prepayment type transaction. And so there's a lot of potential here. David Leary: [00:55:11] Yeah. On the surface, if you just only look at what you just tried to do, somebody's going to say like you could just went to zero and click those five transactions off and be done. And it would take you three minutes and you'd be done with this. But the piece of this is you can't do that with 1000 clients at the same time. But you could make this work on this assignment on 1000 clients at the same time. You might have to check in on it and watch it, but and it could work all night 24 over seven. And that's the that's the offset of the accuracy right. Is you can just have these are like employees that work 24 hours a day, seven days a week. Blake Oliver: [00:55:47] What if this was an agent that acted continuously and it's just switching from file to file in my dashboard, and it just looks for new transactions every day in every client, and it reconciles what it's confident it can reconcile. And then anything that is not confident about it asks a question. Let's say it sends that question to slack, and then I can just tell it what to do. And then it goes and does that. That's like how I worked with my human bookkeepers when I had my firm. Their job was to go through every file every day, work on work through the transactions, and ask questions of the client and me if they weren't clear on what to do. And that way we had constantly, daily or weekly reconciled continuous, continuous, continuously up to date books. And I think we're going to get to the point where we could have an AI agent in our firm just going through and Reconciling stuff like I do every day. I still do this. I close the books. David Leary: [00:56:51] I pop into the bank. Blake Oliver: [00:56:52] Categorize transactions every day. David Leary: [00:56:53] Five, six days a week for sure. Blake Oliver: [00:56:56] That way I don't have to forget about it. I don't forget about it. All right, let's thank our next sponsor. David Leary: [00:57:02] Two ads back to back. Blake Oliver: [00:57:04] All right. Cool. Uh, I'll read this first one here. Thank you. To Human at Scale for sponsoring this episode. Are you the bottleneck in your own accounting firm? You've built a successful practice, and you have a team, but now you're drowning in client work instead of growing your business. That's where human at scale comes in with their operations audit. They're not just consultants who drop recommendations and disappear. Their team actually works alongside you to transform how your firm operates. Here's what makes them different. They guarantee measurable outcomes, real improvements in your margins, and team productivity. They build systems that allow your junior staff to step up, freeing you from constant firefighting and quality control. Imagine consistent client experiences, smoother onboarding, and processes that don't require your personal touch on every project. Their clients see stronger margins, increase firm valuation, and maybe most importantly, they regain the time to work on their firm instead of in it. Human at scale helps six and seven figure accounting firms create scalable systems that run without the partner being involved in every decision. Recently, in a mere 60 days, they helped a firm reduce client onboarding time by ten days and improved team response time by 75%. To see how they can transform your accounting practice from chaos to consistency, head over to The Accounting Podcast. That is The Accounting Podcast. And now let's thank Mrs. David. David Leary: [00:58:30] I'll read it. Blake Oliver: [00:58:31] You go for. David Leary: [00:58:31] It. Here at earmark, we rely on missive every single day to manage our team communications. And there's a reason hundreds of accounting firms have made the same choice. If you're running an accounting practice, you know the pain. As you grow, you become the bottleneck. Every client email lands in your inbox and you forward it to team members with a quick note. Handle this. Then it just disappears. No visibility, no accountability. No idea if it's actually getting done. Missive solves this brilliantly. Instead of emails vanishing into inboxes, client communications go to a dedicated team inbox where they're shared with everyone who needs access to them. You can assign conversations to specific people, create trackable tasks, and have internal discussions right within the client conversation, all without the client seeing your internal notes. What makes missives special is that it feels familiar. Like the email client you're already using, but incredibly powerful under the hood. Whether you organize by client, by service line, or run, a pod based team's missive adapts to your workflow. Plus, their AI can automatically escalate urgent matter urgent emails to managers, so you're not constantly checking every inbox, ready to stop being the communication bottleneck and want to try it for free. Head over to The Accounting Podcast Submissive. That is The Accounting Podcast. Blake Oliver: [00:59:49] David, do you know why I love missive? It's one thing, really. It's that it combines slack and email. David Leary: [00:59:57] Yeah. Blake Oliver: [00:59:57] That's how I think of it. I was a big slack user when it first came out, but then what I found is that now I had to check both my email inbox and slack, and we were talking in slack about a bunch of stuff that was going on in email with clients, and it wasn't all in the same place. And missive gives us the team chat functionality inside of the email inbox, so we can start an internal chat in the same thread as the emails with the client, and only we see the chat messages, the client doesn't see them, and to me that is gold because I only want one place to go. David Leary: [01:00:39] This has been great. The only problem with missive it just suffers the same thing. That outlook suffers. You search for an email or you want to search for something and you you get too many results and you still have to like it just takes a long time to find something inside, which maybe that's a fault. That's email. Blake Oliver: [01:00:53] Well, no, we're spoiled by Google. Google search built into Gmail. Like that was the thing that got me years ago to switch from Outlook to Gmail and never want to go back. And every time I have to use outlook, I hate it because the search sucks. Like finding stuff in Gmail, the keyword. It's just so much better. So you never have to file anything in Gmail and Outlook. Even today, I feel like you still have to file stuff to be able to find it later. And mischief suffers from this too, but I know they're working on it. They've said they're working on improving the search, but yeah, for me it's worth it. Not. And I'll just actually keep a Gmail tab open in case I need to search for something really easily, like if I can't find it quickly and missive. But mostly I can and I save the time because I'm not looking in slack. So. David Leary: [01:01:41] And a lot of times somebody else has the email open and they you just ask them to tag you and they just say David. And then I then it pops up to the top of my thing and I can find it. So that's the beauty of the collaboration. But finding things is always difficult. Let's jump back into the news. Blake Oliver: [01:01:55] Well, let's go back to like, uh, all this app news fundraising because we still got more, um, Thomson Reuters did you hit on that? David Leary: [01:02:01] Did not hit on Thomson Reuters. Blake Oliver: [01:02:02] Okay, so Thomson Reuters has launched some AI powered tech solutions. Solutions? Um, they've got something called Ready to Advise, which is available now in the US and Ready to Review, which is launching in Q4 of this year. Ready to advise is an AI powered tax planning advisory solution that helps firms grow strategic advisory services by synthesizing client data from tax returns to surface personalized strategies, providing step by step guidance with supporting authoritative knowledge, resources and connecting firm content. Checkpoint, IRS code and internal documents and one AI guided workspace. And it allows you to automate workflows while keeping humans in the loop. So that's a big potential benefit is if you're using Thomson Reuters to do all your tax prep. Now they've got AI in there. They're saying will take the information from the tax returns and help you do tax planning for clients. That sounds really powerful. And the reason that it could work well is because they've already got checkpoint. They've got the one of the best tax research platforms out there. And so it's using that to figure out what tax strategies. So you're not getting, you know, TikTok tax strategies sourced from Google search or something like that. Now the other one is called Ready to Review which is coming up. And it's going to leverage AI agents to automate processing of source documents and previous tax return information. And it will extract and categorize data to produce tax returns ready for professional review. It will utilize tax Thomson Reuters strategic tax engine and software to support AI agents. So this is really interesting. Like all this AI coming into this traditional tax prep. Blake Oliver: [01:03:57] And it could really, really help out firms if it's done right. And I'm excited because I'm working on scheduling an interview with Elizabeth Biström, who's the president of tax and accounting professionals at Thomson Reuters. So I want to talk to her about all of this and that when it comes out, will be on my earmark podcast. So do subscribe to the podcast and earn free CPE if you've never heard it. It's where I go deep in depth on topics with guests. Bloomberg Tax and Accounting has also rolled out major upgrades to their AI assistant. Key features and capabilities include enhanced research tools. So this is Bloomberg tax. You can engage in ongoing dialog with the AI to do research. You can see your previous conversations. You can filter research by specific states or court systems. And it will even create charts that compare tax comparisons between different jurisdictions. It integrates with Bloomberg tax news pages and articles, and it will extract extract tax implications from news stories without disrupting your workflow. So if you subscribe to Bloomberg Tax for research, it looks like these AI assistant features are coming. They're there across the entire Bloomberg tax platform, and it's available at no cost to Bloomberg tax subscribers. Deloitte is giving its auditors more Agentic AI capabilities. Their global cloud based audit platform, Omnia, now has agentic AI tools that can perform complex tasks autonomously. What can they now do? These lucky Deloitte auditors. The AI will perform initial reviews of audit documentation and suggest improvements for clarity and consistency. They can also do financial statement analysis. They can ask detailed questions about uploaded draft financial statements. Blake Oliver: [01:05:52] They can create first drafts of audit related communications and accounting memos. They can access Deloitte's accounting and audit research platform for synthesis of complex accounting topics, and they can use the AI to evaluate external information sources and identify potential audit risk factors. Now, it's funny when the big firms put out releases like this, because I look at this and I realize, oh, you've just built a wrapper on top of ChatGPT. You've just integrated ChatGPT because that's what it can do now into your proprietary platform. So like if you don't work for Deloitte, you can actually do all this stuff. You can. You can use ChatGPT or Claude and you can upload audit documentation. Now, what will probably be better about Deloitte's implementation is if they've done it right, they will have taken all of their audit best practices, their whole manual audit manual. And and given that to the AI as part of its knowledge or better yet, even trained the AI on that and you're just going to get way better responses that way, way more accurate, as opposed to just asking ChatGPT without any knowledge or training in its system what it what to do. So that's pretty nifty. And another piece of app news is GPT five is now available in Microsoft 365 Copilot and as I mentioned last episode, what's cool about GPT five is that it will automatically select the right eye model for the task. So if it's an easy question, it's supposed to pick the mini model and just give you an answer immediately. And then if it's a question that requires a little more thinking because it's more complex, it will route it to the thinking model, which takes a little longer. Blake Oliver: [01:07:50] Now, part of the reason that we've seen all this negative news about GPT five is because I think it hasn't been doing necessarily the best job at routing requests for a lot of users. And I wonder if that has to do something with like overload on the system. So like if too many people are using the thinking mode or if too many prompts are using the thinking mode, it's going to default to the mini. And OpenAI has a financial incentive to try to use mini more than thinking, because it requires way less power, way less energy. And that's the biggest, you know, problem for these companies like anthropic and OpenAI, is that they are just using a ridiculous amount of energy and they're losing money because of it. They are just like these AI prompts are very expensive. That's also a reason why, like, I don't think we have to worry about these autonomous AI agents taking over anytime soon, because right now you can use them. And I haven't hit any limits with them yet. But if you watched us do this reconciliation task that we did earlier with ChatGPT David, it's basically doing prompts every second. And those prompts must be pretty large, because it's sending an image and a prompt and then having to get a response, and it's just going back and back and forth. I want to know how expensive that was, what we just did. I feel like at scale, the AI agents are going to be really, really expensive. David Leary: [01:09:23] And this is why routing. Because now when you have your routing people to different paths, this is where you can route them through the ad path, right? Oh, o. They searched for this. Now drive down this path and subconsciously, like you could say, I'm preparing for my podcast. It's on a video podcast. And it could be like, well, Blake, you should buy this toothpaste and brush your teeth with this. We're probably the march we're going to see. That's that's one of those things. They're hiring a bunch of these people from Google that specialized in the routing of traffic. Essentially it it's probably tied to ads as much as it is resources. And we'll see. Blake Oliver: [01:09:57] Some more fundraising news in the accounting app world. Giraffe. The Fpna platform has raised a $20 million series B to power it. I full disclosure I used to run marketing at giraffe when it was a tiny eight person startup, so I'm very happy to see that they're continuing on this journey of Fpna. Um, yeah. If you haven't checked out giraffe and you're looking for something that can do really, really powerful, multi-dimensional financial analysis like and connect to Qbo and Xero and your payroll and bring in all the operational financial headcount data and do financial modeling in a very sophisticated way. Giraffe is the like. It can do stuff like adaptive insights, but at a way lower price point because they built this three dimensional financial model. It's so cool. Um, so congrats to giraffe. Here's another fundraising story that is Accounting Adjacent delve, which I think we may have talked about on the show before. David Leary: [01:11:07] Or spell it. Blake Oliver: [01:11:08] Delve like dive, delve, delve into, you know, delve. So they were founded by 21 year old MIT dropouts, and delve focuses on compliance work. So HIPAA soc2 PCI, GDPR, ISO that is all they focus on. And they have AI agents that run in the background to integrate with customer tools and automate compliance workflows in real time by collecting evidence, writing reports, updating audit logs, and tracking configuration changes across systems. Basically, it's an AI platform that has access to your systems and ensures compliance with all of these standards. David Leary: [01:11:57] And it's constantly checking. So as your you add a new thing, you do something new at your company. It's constantly making sure in compliance, instead of doing an audit every six months like, oh, these things aren't compliance. Now we can go fix a bunch of stuff. It's constantly watching. That's like its big value. Blake Oliver: [01:12:13] So they help you get compliant and then they monitor it. And they just raised $32 million in a series A at a $300 million valuation. All right, David, that's all the time we have for today. Next time I want to make sure we hit on the impact on the job market, because I is wrecking the job market for college graduates. It is really disrupting things. I think it's going to really disrupt the flow of accounting majors into firms, to the point where even public accounting may not be the place where we start our careers. It might be industry because we are just not going to need all of these fresh grads going into public accounting, because I is going to do that grunt work that they used to do. So where are they going to go? I think they're going to have to go into industry first, which is really going to shake up the education game, because right now, accounting education is all about sending people into big public accounting firms, and it's not preparing them to work in the real world. They get that experience in public accounting. So I want to talk about that. Um, and then maybe we'll hit on some tariff stuff too, depending on the news. Thanks everyone who joined us live. Sorry that we didn't get to much to the chat this, uh, this episode. James. Light them up. Deborah Chaton, great to see you. Thanks for commenting. Jan. The virtual Indian love that. Thank you for joining us and hope to see you around here soon. Don't forget everyone, you can earn free continuing professional education credit for listening to episodes of the Accounting Podcast. Go to earmarked in your web browser or download the free earmarked CPE app on the App Store or Google Play Store. You can create a free account and earn one CPE credit for free every week. You can subscribe for unlimited and to support our work. And if you need IRS Continuing Education, we also have that look for the courses with the purple IRS banner and get your IRS continuing education taken care of as well. Bye everyone. David Leary: [01:14:19] Bye.