Randy Johnston and Brian Tankersley, CPA, discuss the latest developments in agentic AI, including explaining what it is, how it can help accounting firms and their clients, and what their preferred AI platforms are. Watch the video, listen to the audio, or read the transcript. The Accounting Tech Lab is an ongoing series that explores the intersection of public accounting and technology.
Or use the below podcast player to listen:
===
Transcript (Note: There may be typos due to automated transcription errors.)
SPEAKERS
Randy Johnston, Brian F. Tankersley, CPA.CITP, CGMA
Randy Johnston 00:11
welcome to the accounting Technology Lab. I’m your host. Randy Johnston, with co host Brian Tankersley, we are pleased to talk to you about the latest agentic AI developments. Many of these were released over the summer. Now, you do know that there are large language platforms in the market, just like we talked about in the you know, new AI action plan for the US. We believe right now the safest of the platforms is Claude, followed by copilot, followed by perplexity, followed by chat GPT, and then Gemini. But popularity, it’s hands down. Chat GPT at 61% market share and copilot at 14, followed by Gemini, perplexity and flaw. Now I just lay out the large language models, because what we’re talking about today is not that we want you to get productivity from the large language models, but where we think AI will really be a winner is in the use of agents. Now, Brian, you know that there are several advantages to agents. What would you like our listeners to know about that?
Brian F. Tankersley, CPA.CITP, CGMA 01:17
Well, agents will actually take action without you having to do much of anything, okay, and so it will, they will actually go through and they will do multi step processes, okay, you know, if you’ve used generative AI, you’ve noted that you can get it to create. You can get it to, for example, go out and create an outline for you, and then you can say, okay, flesh this outline out and fill it out. And it will start to bulk as you start trying to do this, but you can go through this thing, but you have to manually kind of feed the pieces in here. Okay, think of agentic AI as as, I guess, kind of like a, kind of like the paper tray on a on a printer, as compared to when you have to manually feed things in okay. Like, you know, some of you, know, some of you, whenever you print out something that’s going to be on letterhead, you you manually feed the paper in. You have it set to use a different tray, and then you have to sit there and feed the things in, you know, my labels I put on my honey jars, I have to, I have to manually feed them in, okay, one at a time, and that’s kind of like what happens when you’re using the traditional tools. Now, if I had agentic, however, it would go ahead and it would make the list, and then it would go in and do and do multiple, multiple steps. You know, we’ve seen this with the perplexity search, with with the deep research tools that are in some of the platforms. The The idea here is that, is that it will do complex things, and then it will take its own actions as needed on things. I remember there was a, there was a guy from, you know, this is probably 10 plus years ago. There was a guy that was a, did a presentation at AICPA engage on self healing networks about some of the things that that happened. For example, they had, they had bots that would actually detect when, when there was behavior that looked like ransomware, and it would automatically shut down that that that that action taking place on the network to stop ransomware from being able to read and write files if something was acting like ransomware. And so this, this whole, this whole point of doing this is that, again, it gives you, lets you have work better workforce specialization gives you better trustworthiness. It lets you, it lets you be a little more innovative. So again, you think about managing complex it, you know, what if something could detect that you are running low on, you know? And we’ve seen this with some printers now, where they they now, if you subscribe to toner, and they detect that you’re printing a lot and you’re low on toner, they just automatically ship you some. And so there are services out there that do this for customer service, again, taking actions on your behalf. You know, I wish that. I wish I had an agent that would go out, for example, and when, when there’s a greater than 80% chance of severe thunderstorms near an airport where I’m flying. I wish that I could get the agent to go out and automatically book me a backup itinerary, okay, without me having to do anything kind of the same thing an executive assistant would do. But again, sales support, healthcare and social care, content creation. There are a lot of things like this that agentic AI does, and it does use sophisticated reasoning and iterative planning. You know, what if? What if somebody you know, when they sent you an email, the agent, you know, the tool, would automatically generate a proposed response to it. Or you could put a few items in, and again, that’s where copilots coming in, where you can actually have it generate responses to these things and but again, when we move into agentic, now, it’s really going to work and act intelligently and independently to solve these problems. Okay, so it really is in this world, the bots are going to take action. Now, the hard part about it is that is that these. Bots are going to need access to more and more and more of kind of the keys to the kingdom, sensitive information, like your passwords, like your email, like your your text messages on your phone, and all those things if they’re going to generate these responses and these, these other, other pieces of things in here.
Randy Johnston 05:18
Yeah, so you’ve heard from Brian that really the ability to work autonomously is a big deal. With agentic AI, they tend to hallucinate less. They are really self directed. And if we contrast agentic AI and generative AI, it’s for agentic AI, it’s autonomous action and decision making. The autonomy is high. The adaptability is great. Goal Setting capability is inside the agentic AI and human oversight required is very minimal, whereas with generative AI, you know, it’s content creation based on the prompts, and the autonomy is very low, like Brian cited with the paper feeding issues, there’s some adaptability in the Gen AI engines, for example. Agents have been added to chat GPT, so the big llms are going to be trying to build more and more agents for your use, but we think agents are going to be built in products, which is what this session is really about. And with generative AI, there’s very little independent goal setting, and you have to have human oversight, because what’s sometimes produced is just inaccurate. I was actually using generative AI for a report yesterday and realized it had just spewed some garbage, which was kind of logical sounding, if I would have been a little more tired, it could have slipped through the cracks, and my piece that I had it right would have been wrong. But, you know, there’s a lot of difference between file cabinet, CS and go file room, and it crossed those two up.
Brian F. Tankersley, CPA.CITP, CGMA 07:00
Yeah. Yeah. And, you know, that’s the thing that’s, again, the thing about the genetic here is that it’s it really uses this four step process for solving problems. It will perceive or and see an opportunity or see an action that needs to be taken. It will then reason what the appropriate reaction is. It will then act, and then it will learn from the feedback. And so, as we’re again, as we’re as we’re thinking about this now, it can take inputs from a wide range of different things. You know, for example, one of the things that I wish my my cameras would would model would be, you know, if it if, if my cameras model, and they see a skunk in one of my bee yards, it should put something on my to do list, to go out and trap the skunk and get it the heck out of there. And so this, you know, again, it will. It can take input from, again, unstructured data sources like text and email and web pages and things like that, and databases, as well as unstructured data sources like cameras and pictures and other things like this. But I will tell you that my AI on my cameras is just not very good. You know, the yesterday, yesterday I had, I was looking at my my cameras I have on one of my bee yards, and two of my dogs walked through and and it flagged the motion in there. But unfortunately, the the AI in it identified the two dogs that were walking next to each other, or walking, excuse me, one in front of the other. It identified them as a car, which I thought was kind of sad. So, you know, it’s, you know, again, with this, you are going to need to have higher, higher cognition in here. So again, you’re going to have, again, the thing I would say here is the agentic is going to have generally better executive function, where you where? Again, you can say, I want you to do this. And it will come up with a list of tasks, and then we’ll go attack those tasks sequentially. One of the things that’s also different about some of the new some of the new agenda agentic AI tools, is that some of them are launching multiple threads at the same time to solve these problems, and so it’s going to use geometrically more computing power, because it’s because of the approach of engaging with multiple, you know, multiple, for example, large language model queries at the same
Randy Johnston 09:13
time. So Brian, now I think you’ve helped to create our next product together, skunk detection for technology. Okay? Launch an agent to find out where the Skunks are at in the technology. And boy, there’s a lot of skunks out there right now. You
Brian F. Tankersley, CPA.CITP, CGMA 09:28
just want, we just want a world where it doesn’t stink. Nobody wants that.
Randy Johnston 09:32
Well, we have 14 different announcements that we want to cover with you today, 14 that have happened within the prior 60 days of our recording this podcast today, the first one we’re going to call out, and probably was a trigger for some of this, was intuits agentic AI announcement on June 27 now, in this particular case, Intuit has been building their AI engine for about six years. It’s. They currently have a policy data, yes, and in fact, that’s really one of the key points here, is there is no option to note. There’s no way to opt out for any client or yourself for any of the accounting data. So it’s one of the downsides. As we see it, they’re trying to deliver upside with four generative agents that can autonomously work across QuickBooks for payments, accounting, finance and customer support. And these agents can automate payment reminders and reconcile bank transactions and Forecast Cash flows and draft client emails, and they can do all those things without agent, without human interventions completely done by agents. Now, IDC claims that agents will be a $25 billion business within three years. The productivity gain here is genuine into its proprietary Genos operating system is doing all sorts of actions. So Brian, I’ve probably taken a lot of thunder on this, but what would you also want our listeners to know about intuits agentic AI announcement and release?
Brian F. Tankersley, CPA.CITP, CGMA 11:10
Well, they came out with another release about four weeks after this, and I didn’t see you know that it was really focused on the Intuit enterprise suite, which is really their their addition to QuickBooks, QuickBooks Online, advanced, you know, they’re adding on to it a number of marketing and other tools there to call it an enterprise suite. And just remember here, this is not enterprise. This is enterprisey, like it integrates, okay? But it’s not enterprise like SAP and Oracle, okay? So nobody in Germany at SAP needs to worry about getting displaced by into its enterprise suite. Okay, but it is a, it is, you know, kind of lower, a lower end, mid market product. And again, it’s going to be interesting to see how, how some of these other AI related tools that we’ve that we’re going to talk about here in a few minutes, that are, you know are going to affect it, because it’s Intuit is, is really gone all in for this, and they’ve made a number of choices that have alienated accounting professionals, including pretty much, pretty much ending, well, ending new sales of of QuickBooks Desktop, and kind of signaling, not so subtly, that QuickBooks Desktop, and probably QuickBooks Enterprise, may, may be ending in the in the intermediate term. So I guess what I just what I just generally would say here, is that they’re gone all in on AI, and it’s going to be interesting to see what our relationship looks like as a profession with Intuit based on some of the choices they’re making, which tend to be tend to favor their use of AI, their use of sensitive data for purposes that make most of us pretty uncomfortable.
12:49
So anyway, well,
Brian F. Tankersley, CPA.CITP, CGMA 12:50
said, lot of announcements
Randy Johnston 12:52
here, there is a lot of announcements. So we’re going to start with the smart vault product there. Their DMS has a new feature called Smart request AI, which automatic document collection and client intake process, and their smart vault believes that this could take save 60 to 90 minutes per return, which is a pretty big deal. Of course, smart vault has also integrated with a lot of the primary tax products out there, and the key capabilities of smart request is the AI powered questionnaire and document request list that they can also do a bulk creation of client client request to save some time. They’ve got a firm dashboard to monitor the status, and then they actually organize a work paper generated from the completed request, ready to import to the tax software. Now, whether that will automate the AI tax prep products like black oil or filed or magnetic those type of players or not, is to be seen, but also can just import straight into tax. And I think they’ve done a nice job with this capabilities, because it reduces a lot of the back and forth email and does save time. So that’s one example of an agent that’s been added to a platform. So Brian, you want to maybe talk about SoVi from the sales tax people, Sovos.
Brian F. Tankersley, CPA.CITP, CGMA 14:27
SoVi is really just a sales tax tool that that, again, is that’s in here. And I think this is indicative of, again, remember, Sovos is the is really more of an enterprise, AI, enterprise, AI, enterprise, excuse me, sales tax tool. So it really isn’t going to play as much in the small business space in there, but again, it’s, it does give, give some, some pretty significant embedded AI tools, you know, I’ve talked about into its ai, ai that they added to enterprise suite. And, you know, in their. Are some additional things there, but nothing, nothing super earth shattering, I guess data Snipper added AI agents to to ask audit support questions. Randy, you want to talk about that a little bit? Yeah,
Randy Johnston 15:11
that was done on July 29 and you know, of course, the vendor had released their financial statement suite earlier, but this was a follow up to improve document so you can now do a single query against content. So it would be possible, for example, to ask, what evidence do I need to collect to perform a SOC two user access control? Or can you extract the right evidence related to the SOC two control and match it to my sample list, and are there any documents with risks that need my attention? Those are all pretty interesting questions. Philo Richter, who’s VP of product and engineering at data Snipper, said that as the ecosystem of agentic commerce, execution evolves, software development must evolve with it and with data Snipper agents running on Azure, we want to empower auditors with agents that work alongside them to automate critical tasks. So there was some very interesting support from Microsoft on the announcement, but data Snipper itself says that new customers favor packages with AI products and agents will improve those experiences. So that’s a lot of rant on data Sniffer, but that’s what they’ve announced. So that probably gets us to the Bloomberg piece, Brian,
Brian F. Tankersley, CPA.CITP, CGMA 16:34
yeah, and again, they they announced some AI assistant research features in here. This is similar to what you’ll see in here, Walters, Kluwer and and Thomson, Reuters and others. Again, I think, I think most of the people with heavy duty content want to leverage that content to train AI models, and they’re still trying to kind of kind of figure those pieces out. We talked about the CCH. And
Randy Johnston 16:58
before you go on, Brian, sorry to interrupt you on that, but Brian and I did record a separate accounting Technology Lab on the Bloomberg tax AI capabilities earlier in the year, but the new AI assistant announcement from July 25 is actually a pretty big breakthrough in terms of Adding additional pieces, and that has no additional cost. And it evolved out into experimental development on Bloomberg tax Innovation Studio. So again, when you look at these announcements, sometimes there’s fluff, sometimes it’s like, yeah, this is part of the wall arc of where the product is heading. Now you and I both like cch answer connect and their AI capability. So I need to turn you loose back on what you were going to say.
Brian F. Tankersley, CPA.CITP, CGMA 17:46
Well, I think generally that that these, you know, we’re covering this primarily because there are, there is a lot of big change taking place. And I don’t want to get too deep into any of these particular tools and necessarily that much detail, because I don’t know that we have that much time left, but, but I will just say generally that the tools like cch answer connect, that pull data out of the tax applications, and then actually will will come up with planning opportunities and other things like that. And again, the the level of capabilities that they’re adding are is certainly worth looking at. And it’s and I will assure you that if we record another episode on this, on this topic, in six months, there will be many more innovations beyond what you see on this slide in the next slide.
Randy Johnston 18:32
Yeah. So it turns out that the WK announcement, which summarized documents with AI, you know, that was pretty straightforward, but a big time saver, as it turns out. And of course, you know, ramp just took some more public investment money, sorry. And you know their announcement were AI specifically for controllers looking at low risk expenses and suspicious receipts and other things. And you know, Brian, since you’d written a new expense session for us this year, you probably had your head wrapped around that pretty well, but it was one of those that, you know, it probably is worth adding that even to our presentation, just because there are others doing that. But that’s a new agent
Brian F. Tankersley, CPA.CITP, CGMA 19:19
well, and it’s been very interesting to see how much media presence ramp has had. You know, they’re I’m seeing them pretty much everywhere,
Randy Johnston 19:26
so they seem to be pretty wired for sound on that. And of course, Ledger did their AI tax assistant for rental property owners, where they were doing real time tax calculations and automating expense classifications and adding properties and so forth. I mean, it’s if you’re in the property management business, their AI agent does some stuff that has to get done. And I think next up was to Pulte, wasn’t it?
Brian F. Tankersley, CPA.CITP, CGMA 19:52
Brian, yeah, they had some AI native Treasury automation statements in here to try to make it easier to do more easy. Easier to do again, things, things with their treasury. Again, more of a big company kind of thing here. But also tax file came out with, you know, the interesting part about tax file and Integra global and this, and some of these, some of these other tools here is the combination of offshore, offshore labor and agentic AI and offshore labor and generative AI, I think, is changing the outsourcing game pretty significantly too, you know, because a lot of the people that are doing pure play, data entry, their work is we’re literally being displaced by the by the advanced machine learning that we’re seeing today, as well as the AI tools. And so there, there are people out there that have some some familiarity with sitting in front of a computer all day that now they can deploy to solve, to solve problems that the AI is not quite ready to solve yet.
Randy Johnston 20:52
Yeah, yeah. In fact, a lot the AI tools, like txf intelligence, I think, will change the game of outsourcing notably. And you know, we saw that in other episodes that we’ve recorded in these areas as well.
Brian F. Tankersley, CPA.CITP, CGMA 21:07
Now, now the ignition, the ignition, AI powered pricing feature is actually pretty interesting in my mind, because pricing is one of those things that I think a lot of us, a lot of us people that tend to tend more toward being technical accountants, and less less toward being the the sales accountants. You know, pricing is something that’s hard to figure out what the right strategy is, and so forth, and so, putting something into your process, into your proposal, solution for AI powered pricing is actually a pretty interesting approach that that we have. So I’m glad to see ignition doing that.
Randy Johnston 21:43
Yeah, now on the tax planning tool from instead Henry argue. We did get to speak with at AICPA engage, and he showed us some of the tax return analysis reports and tax plan reports and tax strategy reports. Again, very interesting. Agentic AI driven
Brian F. Tankersley, CPA.CITP, CGMA 22:03
reports. And let’s not forget that instead, has also had a like eight or nine figure investment from from the iris group to create a new tax software, so a new tax prep software. So that’s a that’s another interesting initiative there, again, where there’s, you know, it’s, it was announced in second quarter, I want to say so we may, may have covered it in this, but it’s a huge amount, huge amount of investment from IRIS to effectively create a, you know, along with, instead, to create a tax prep engine,
Randy Johnston 22:40
yeah. And then you turn your attention to all of the HR and payroll companies announcements we just picked on PAYCOM, who’s done an AI powered ask here tool, and it’s intended to Do employee questions, mass employee inquiries, standardizing company responses, and in effect, be a self serve communication center. And as I think about payroll regulatory or HR regulatory environments, having agents that are responding to the common questions that are, you know, national or state or international in scope and filtering that with your own company’s policies. That’s pretty interesting, because, you know, if you’ve ever been around an HR office or around payroll processing, employees always have so many questions, which, frankly, are routine in most cases.
Brian F. Tankersley, CPA.CITP, CGMA 23:40
Yeah, absolutely. Well, very good. Well, it’s a, it’s an interesting time to be sure, there is a lot of money being spent in this area. You know, we talked about in a previous podcast about the, about the hundreds of billions of dollars that are being invested in AI right now. And, you know, I don’t know that hundreds of billions are being invested in accounting, but a significant, significant amount of investments taking place. You know, I don’t know what, what the future holds for us, but it’s a it’s a very exciting time to be in the profession. Be sure. Any any other thoughts? Randy,
Randy Johnston 24:14
well, no, I appreciate you mentioning that, because you’re right. If we recorded this again tomorrow, not even six months from now, the answers and the observations would probably be different, but we tried to pull out announcements that were accounting centric to help you just get a flavor of how much is going on out there. And we clearly don’t want your thinking about AI to be affected by what I would call flood the zone marketing. You’re going to get hit from every which away with lots of information for a while. The information that was being delivered was, frankly, not really right, but most of this right now seems to at least have some. Some essence of factual claims on the new generation products that are coming now. So with that said, it has been a pleasure to have you listening in again today, and we’ll talk to you again soon in another accounting Technology Lab. Good day.
Brian F. Tankersley, CPA.CITP, CGMA 25:17 Thank you for sharing your time with us. We’ll be back next Saturday with a new episode of the technology lab from CPA practice advisor. Have a great week.
= END =
Thanks for reading CPA Practice Advisor!
Subscribe Already registered? Log In
Need more information? Read the FAQs