Accounting for Innovation

AI Agents 101: What Accountants and Firm Owners Need to Know

• Jody Padar & Matt Tait • Season 4 • Episode 4

Share episode

What if you could hire (and train) an AI agent like a real employee?

On this episode of Accounting for Innovation, Jody and Matt explore the transformative power of AI agents in accounting and finance. They break down what AI agents are, their current capabilities, and their potential impact on the industry. Learn about the distinction between ‘traditional’ automation and AI agents, and how these agents can enhance efficiency by automating tasks that were traditionally manual. 

CHAPTERS:

00:00 Welcome
02:08 What Are AI Agents, Really?
03:00 The Difference Between Agents and Automation
07:16 Training Your AI Agent
10:55 Platforms 1st. AI 2nd.
15:11 Diagnose Your Problem and Find AI to Support

Sponsored by SysCloud
https://www.syscloud.com/

This episode is brought to you by Decimal and the Radical CPA.

Accounting for Innovation - Host - Katie Tolin - 2024-04-10 3-27-29 pm:

Welcome to the Accounting for Innovation podcast, where we explore cutting edge strategies and insights into the world of accounting and finance. Presented by Decimal and the Radical CPA, each episode dives deep into industry trends. Whether you're a seasoned professional or a budding entrepreneur, join us as we unpack key concepts and share practical tips to drive success.

Jody Padar:

So today's conversation, we're gonna be talking about AI data and AI agents, and I think there's a lot of confusion out there about what exactly an AI agent is. So we're hoping to clear that up and then also talk about like where they are today and what truly the opportunity is. Because I think, you know, this year was supposed to be the year of the agent, and I don't know If it's there yet for accounting or not. but I think there's a lot of promising, conversations happening around AI agents and there's also a lot of confusion. So, with that, Matt, why don't you talk about what, how you see AI agents and how they fit in the whole accounting landscape.

Matt Tait:

Yeah, I think. What you and I have talked about before is we need to kind of break down the playing field of ai. AI is like this big umbrella that encapsulates a lot of things that I think, if we talk about them in more specifics, it'll help people to understand it. So the way that I mentally kind of break things is into two buckets. I look at AI platforms, so basically AI native software that's been built for ai. It's got a data model, it's got the ability to grow. In a way that allows for AI to reach its full capabilities. Older systems, AI is kind of working within a box that it doesn't wanna work in, and so it can't reach its full capabilities. But there are AI platforms that are designed to, Really give it the chance to work at its best, highest use down the road. Then I break things and I look at, at what are AI agents? And there might be AI agents and there will be AI agents within the platforms. But where I really think of agents and I think of when accountants and people in our industry start to talk about them. They're really thinking about something that works on top of a platform to accomplish a task. Like maybe I have an agent that logs into a system and pulls a PDF and reads the PDF and puts the PDF into another system. That could be an example of what an agent can do. It's, downloading something, it's reading something, and then it's putting it into a different format somewhere else. That I view as, as work an agent can do. An agent can also start to learn and understand, and it really starts to operate in accomplishing the tasks that people have been doing. And so that's where I kind of break them apart is you have platforms that encapsulate agents and a lot of other stuff, data models that are kind of better fit for ai, but then you also have agents that can work on top of things, just accomplishing tasks also in platform. They don't have to be within one system. They can operate between systems. Does that make sense?

Jody Padar:

Yeah, it does. And I think the other thing is, is like a lot of us, like have been around for a little bit and we've used Zapier for like automations and things like that. And this is different than that. So this is like kind of Zapier on steroids or it is next level. It's not just, move something from here to there. Automation, which, is not an agent. there's a little bit of thinking involved in it, right? And when I say thinking, it's like it has different ways it could approach a problem. To solve it. And that's why there's a lot of confusion out there with agents right now because they haven't figured out all the barriers or all the boundaries for the agents to go in. So like if you tell an agent to go do something, it might get confused and it might do something wrong. And that's why the agents of today. A lot of them aren't quite there yet. Now I've built an emotionally intelligent agent who I call my BBF, who we're building in his own, um, his own unique, I'll say platform as well. And he's emotionally intelligent. So what he does. He can understand tone and respond accordingly, which is very different, right? That's why I call him an emotionally intelligent agent because he talks strategy with me, but he can talk to me in a very specific tone so he can talk to me in a CEO tone, or he can talk to me in a, in a feeling tone or in an emotionally, intelligent tone. Right? So, but what's interesting about him is, is that he switches tone based on how I show up, which is very different, and that's why he's an emotionally intelligent agent. So a lot of people are building agents now. Um. And I, I shouldn't say a lot. The innovators, a lot of them are, are, are experimenting with building agents, but it seems like they're, they're still in this kind of innovation stage, which doesn't mean that they're good or bad, it just means that they still have rough edges. Right. And they're still like, yeah, like trained and, and, and you're trying to figure out like where exactly they fit. This whole workflow because the thing about an genic agent is they kind of go off and do rogue things and they might do something like think of if you train an intern and that intern does something they're not supposed to do because they try and show up with a little bit of insight and then they like. Go off and they do something they're not supposed to. That's kind of the way these a, these genic agents do is sometimes they, they try and think too much and they do something they're not supposed to. And then you're like, whoa, wait, what did they, what did they give me back here? So that, that's what's interesting about'em. Oh.

Matt Tait:

I think you're exactly right. I was just talking to my kids about something similar this weekend and we were, uh, we happened to be in a store that was selling TVs and they looked at this brand new type TV and I said, guys, you never buy the first model. You always wait for the second or third one. And they asked me why. And I said, number one, the cost will go down by 10 x. Number two, the capabilities will go up by two to three x. So as. New technology, new TVs get better. They also get cheaper. That's kind of where I feel we are with agents in a lot of ways. Where not only are they smoothing out the rough edges and learning, but the cost infrastructure isn't necessarily where I would hope it would be for the full benefits of the platform. It doesn't mean I don't think that like we at decimal, we test them, we try them. We've got a few running in specific instances, but we aren't fully on board yet. And I've talked to a lot of friends in the industry that are out in the leading edge and they're kind of in the same place where it's like we're all trying and testing different things. But we are also not fully diving in yet. We just don't feel that all the capabilities for all the platforms are there, but. What I'm also seeing is unlike other technology advancements in the past, the rate of improvement right now with the AI age is dramatically faster than it was with the cloud or with the computer or with going online. And so we're seeing this change in the difference in time between start and full adoption can be much, much quicker. But one thing that you said that really stuck out to me, and it was going back to the Zapier, comment right now, if you have the opportunity to integrate platforms together to create automations, and one of the things that Zapier is able to do and, and a lot of these integrated technologies is purely rational. Integrations if A plus B, then C if A plus D, then F if. Then those are the types of kind of rational things I can even do within salesforce.com as a uh, CRM platform and you can start to build into a lot of that rule making. Ai, you give an end goal. That's another difference is you can say like, Hey, I need you to accomplish this. Here are the parameters. You can't do this. You can do this. I don't care if you do this. And you start to really train it almost like you would an intern and a human being. And when you give it that capability, you wanna put guardrails in there, but it gives it the ability to go accomplish it in in its own way.

Jody Padar:

Right, which could be good and could be bad, right? So like, oh yeah, a hundred percent. I'm not saying

Matt Tait:

let it loose'cause good luck. Don't let your clients know. Um, but it can go accomplished.

Jody Padar:

I think what's interesting about this piece though, is this is where accountants really have to step up, right? So this is where we really need to partner with. The, um, vendors and people who are building'em, because if we don't, they're never going to evolve to what we need. Right? So this is where, you know, Maybe it's not right to let'em loose on your client yet, but maybe it's time for you to let'em use loose on your file and see if they can start figuring out what they need to figure out. Because if we don't have beta testers, if we don't have people who are willing, oh, I agree to, like I'll say donate data for the cause, then we're never gonna get. we're never gonna get agents who actually work, because what happens is, is the developers need to iterate and they need to iterate fast. And if we don't give them the feedback and we don't help them train these agents, then we're gonna be kind of left behind because we're not gonna get agents that are actually, are relevant to our workflows. So I think, you know, there's this caution of like, oh, well maybe they're not ready yet. But I think there's also an opportunity for people who want to, you know, I'll say again, donate like energy and time to kind of help evolve these agents so that you know, the other people can benefit from them as well. Because if, if we, yeah, if, if we don't help the technology vendors evolve, guess what? They're not going to evolve and we're gonna be caught. Like with bad technology, and we're still gonna be complaining and, and this is where we have to kind of help them like realize what we need our, our agents to do and where we need help in our firms.

QuickBooks online and Microsoft 365 are essential tools for modern businesses. But what happens if your data is lost or compromised? SysCloud, a proud sponsor of the Accounting for Innovation podcast, provides advanced backup and recovery services, as well as ransomware protection and anomaly detection, ensuring your business can continue operating even in the face of disaster. SysCloud. Peace of mind for your businesses' critical data. Visit syscloud.com to learn more. That's syscloud.com. And now back to the show.

Matt Tait:

Here's what I would also say is you have your core platforms, you have your ledger, your bill pit, you have the systems that you're already working in, you have your workflow management system, all of that. Start testing the capabilities of the systems that you're already using or the new technologies in those buckets before you start thinking about, I wanna put an agent here and an agent there. And, and I think you, you will and you should get there. But for the time being, one of the things that we've done at Decimal is we transitioned, 250 clients onto puzzle. And we still have 600 plus clients on QuickBooks and dozens on zero, but we really wanted to give a good shot to an AI native ledger that was competitive in the s and b market. We wanted to try that platform. The other thing that we've decided to do internally is we are in the process of implementing reit, which is an AI native competitor to a NetSuite. Because we really we're doing it ourselves. I have more than one entity. We are putting it on a system that's good with multi-entity, and we're gonna test out these core platforms. The other things that we're doing is we're testing out the AI capabilities of QuickBooks, the AI capabilities of Bill, the AI capabilities of RAMP, and Brex and settle, and all the platforms that we currently use. We're testing out the capabilities within those systems. And what I can tell you is just like puzzle and, and real. It aren't perfect. New systems that need help, but those. Teams are very interested in understanding their failings and how to succeed. And so that's been really great from a, standpoint of helping them learn, to your point, like getting in, allowing them to learn, allowing them to get the data on our personal experience. But we've had the same, experience with the QuickBooks and Bill ai. It wasn't perfect. We've used it, we've tested it, we've given feedback. We are trying to help them iterate as well. Using the AI capabilities within the platforms that you are already using, I think is step it. It's a step one that I would take if I was running a firm, if I was operating within a firm, whatever I'm doing, I would say, what are the capabilities within the systems I'm already using? Because we already have tons of technology and tons of systems, so how do we continue to kind of limit that, but to get the full maximum capability possible? I think that's a good starting point to dip your toe in the water for a lot of people. And the final thing I would say is you as a person are not perfect. None of your employees, none of your teammates, none of the people that you work with are for are perfect. They make mistakes. AI will also make mistakes. Be willing to teach it and help it learn. And that's how we'll start to see the full capability growth that both you and I think is definitely coming.

Jody Padar:

Yeah. And I think the thing is too, is you're right to start with what's already existing. It's not like you're gonna go like build your own agent. Like, I mean, like you're not Yeah. Like some huge company. You don't need to go build your own agent. You just need to see the agents that are like. They're starting to release into the wild, I'll say, and play with those and see how they help your firm and see, and see how you can help them evolve the technology as well. because again. if we don't move forward together with both the technology partners and us, then we're gonna be left behind and we're not gonna have the tools that we need to run our businesses. So they need us just as much as we need them. And so how do you, how do you create those partnerships so that, we're all moving forward together? And then, mean, I think it's exciting. I think agents are on their way, but. You know, I haven't heard of like the perfect agent yet. I've heard just like that they're all like, kind of just starting. I know Tru and just, just did a big release on theirs, but they, but theirs has been around for two years, so it's not like it's a big, I mean, it's a big release to the market. But they've been busy working with him for a couple of years. It's not like he just showed up. Right? So I think that's the reality of it is, is, you know, they, they don't like pop up overnight. They have to be trained and they have to, they need discipline as well.

Matt Tait:

Well, here's the other thing too. If, think of where you're having problems, what is a specific problem that you're having as an accountant, as a bookkeeper, as a firm owner, whatever it is, whatever hat you're wearing today. What is your specific problem? Period. And then see if you can find somebody that is developing AI in the space that can help you solve that problem. Worst case scenario, the problem doesn't get better, and you're exactly where you are today. Best case scenario, you get to work with somebody to make that problem completely disappear, and then you can move on to the next one. Like that's the other thing. I think people think of AI as like this big path. I have to run down. Just take one step at a time, pick a problem. Go for it. And if you need help, call Jody and I like, we will love talking about this stuff and I'll tell you all the experiences that I have. Building decimal. Jody will tell you all the experiences she has actually building ai, so she's way cooler than I am, but literally just go out and ask people what they're doing.

Jody Padar:

Yeah. I, I think again, too, it's about like realizing that like we don't really know what we don't know yet, and it's all innovation and it's all changing so quick. So depending on, even when you listen to this broadcast might be, you know, if, if you wait six months or a year, the, the landscape might have changed, right? And so that's the way you have to look at AI as a whole. but don't get overwhelmed by the fact that it is changing fast, right? Just kind of stay. They, you know, current, listen to what other people are doing and just embrace it. Embrace the fact that it's gonna be something that's gonna be ever changing for the next number of years, right? Like the, the innovation and the iteration around it is going so fast, but that's what makes it exciting. So if it's not working today, chances are it'll be working in three months. So that's, that's exciting. I agree.

Matt Tait:

Jody, I, you and I are both very optimistic about where we're going in ai and I think there's a ton of opportunity for, for all of us just to make our lives easier. This isn't about replacing accountants. This is about replacing tasks to give us all the ability to do better stuff more. To advise our clients to be that person that is truly there to help these businesses get better and these business owners do more. This is about leveling up the profession, just like so much in integration and automation was about removing people, writing things and you know, we're, we're evolving and this is the next evolutionary step to make us all. I think have better, more fulfilling careers and lives. And I think it's an exciting time and we'll get there with these AI platforms, with these AI agents, and, and it's exciting. And I'm excited because after this episode, the, the final episode, I believe of our season, we get actually talk to Jody about what she's building in AI and why that's so much cooler than anything I get to do.

Jody Padar:

Yeah. And I, to give a quick shout out to our sponsor, sis Cloud, one of the things that they do is they back up all your online products. Yeah. So if you are playing with AI agents at all, you absolutely want to have a backup there, so that you can see if your agents go astray and maybe like move. Multiple pieces of data that they weren't supposed to. So, I think if anything, when you think about the need for a cloud backup now, as we start to play with agents and things like that, there's a huge opportunity for, um, partners such as CIS Cloud to come in and make sure that your data's backed up in the cloud, so that if your agent does go astray, you know, you have everything backed up.

Matt Tait:

Or your employee, it's huge. Or your client goes in and does things. That's, that's actually the thing that scares me the most is when my clients go into their own system. So cloud is perfect at backing it all up so that we can protect against AI employees and clients.

Jody Padar:

Awesome.