AI & 2025 Predictions

There may be errors in spelling, grammar, and accuracy in this machine-generated transcript.

Hector Garcia: Welcome to the unofficial QuickBooks accountants podcast. I am joined by my good friend Alicia Katz Pollock, the original, the one and only Qbo Rockstar CEO and founder of Royal White Solutions.

Alicia Katz Pollock: And I have the privilege of collaborating with Hector Garcia CPA, the founder of Right Tool for QuickBooks.

Hector Garcia: In this episode of the unofficial QuickBooks [00:00:30] accountants podcast, we're going to talk about the state of AI in the accounting world or the business world, and some predictions that we may have had or we may have for 2025. Hey, Alicia, how are you?

Alicia Katz Pollock: I'm doing great and really interested in this topic, so I can't wait to hear what you have to say.

Hector Garcia: Well, let's kind of take a backwards here for a second and kind of talk about the layers or dimensions of artificial intelligence, because I think artificial intelligence sometimes is a little bit misunderstood. And I'll [00:01:00] fully admit that I'm actually not even close to being an expert or calling myself an expert in the subject matter. But I'm going to explain, basically, from my perspective, how I see artificial intelligence organized in terms of understanding what it is and where it is. And then maybe from there, we'll use that as a framework to discuss where we think each of these dimensions will be growing into. So you're ready, Alicia?

Alicia Katz Pollock: I'm ready. Okay.

Hector Garcia: So the first dimension of AI, we're going to call this the retail AI. These are the [00:01:30] chatbots. So the retail AI is the one that everybody has access to. Chatgpt cloud, Microsoft Copilot, Google Gemini, the Facebook meta. You know, some of the open source, um, open source. I like llama, uh, perplexity, you know, all these tools that people sometimes for free and sometimes for 20, 25 bucks a month, have access to a ton of computing power at their fingertips through a website and in some cases, through their iPhone [00:02:00] or whatever. So that's the that's the retail AI, the chatbots. So that's dimension one. Dimension two. Um, best way to explain it is the API based AI. So these are um, third party applications that create middleware for the lack of a better term that uses the power of AI through the API. So it connects to OpenAI, it connects to cloud, it connects to Google. Let's uh, those systems compute. Then they grab the results from that, and [00:02:30] then they turn that into something applicable for their particular software. So anytime you see any piece of software that says powered by AI, there's a 99% chance that they're already connecting to opening to OpenAI to cloud. They're already connecting to the the database or the core engine that powers the retail AI system. So let's not dismiss the the retail systems. This is how most people understand or start having a relationship with AI is through these chatbots, [00:03:00] but then understanding that probably the same exact engine being used to power those chatbots are being used to power other software applications that are claiming to have AI.

Hector Garcia: Now, Intuit specifically in particularly made multiple announcements about them building their own AI. Uh, but we're not 100% sure if they actually are using parts of, you know, maybe Facebook Llama, which has a, a, a open source AI available or maybe [00:03:30] using open AI. We don't know. We don't know if Intuit uses any of the third party AIS. They have announced that they have their own. But again, it's it's it's the core engine. Now some some applications that we see using AI at the moment, like for example Makers Hub, one of the one of the best apps in the ecosystem right now. Uh, they they use AI to interpret and interpolate data from PDFs, and then they use their [00:04:00] middleware to map the data that's being extracted from those PDFs into particular fields inside QuickBooks, for example. So for automating bills with line items and trying to figure out how to figure out how these line items connect to particular items or particular accounts inside the system. The makers of bills, the middleware. But the engine that powers the AI could be OpenAI, or it could be something else unless they develop their own. But anytime you see any software, you saying that they do AI. They probably are using one of these [00:04:30] third parties. You have any thoughts on this?

Alicia Katz Pollock: Uh, I think that's the one that most people have. Um, see happening is when they're hearing about AI, it's because it's showing up as powering new features inside their software.

Hector Garcia: Exactly. And we're going to go into some of the predictions and what's going to happen in these layers in a little bit. And the third layer of dimension of of of of I is the stuff that's happening behind the scenes that most of us don't get to see. [00:05:00] So one of the examples of this is machine learning is algorithms that are gathering data and then using that data to either build insights for executives to make high level strategic decisions, or at some point, using the insights to automate pop ups, information and actions inside software. So one of the examples of this would be like the way into it uses the data from all the the metadata from [00:05:30] all the users categorized in Chipotle into meals and entertainment, building the algorithms so the algorithm knows that next person that uses Chipotle is probably going to meat meals and entertainment. But then even more contextual where, um, companies that are, you know, in the marketing consulting industry, when they go to Home Depot, you know, maybe they're the those type of companies are putting Home Depot into repairs and maintenance, where maybe construction companies are putting Home Depot into tools and [00:06:00] materials in the cost of goods sold category. So that type of that, that, that work that's happening behind the scenes where the cross-referencing data across large data sets, using the metadata across multiple users to then feed an algorithm that then is going to make a decision to suggest one category or not, that's that.

Hector Garcia: That's that third category, which is like the stuff that's sort of happening behind the scenes. There's another layer to this is since I've been built, since I've been involved in the software [00:06:30] development world, and I don't write any code, by the way, um, I've actually firsthand gone to ChatGPT and showed a screenshot of the reconciliation screen in QuickBooks, gone to inspect the inspect window in, um, in Google Chrome, copy and paste the HTML into ChatGPT and say, by the way, I wish there was a button that I press on this screen and it would automate checking this, this and that. What would that look like in code? And then ChatGPT will turn back and go, well, if you were to write a JavaScript that would look [00:07:00] it would look like this. And then in some cases, I'll literally copy and paste that JavaScript back into the console, uh, into the browser, and you would actually press that and see things being automated on the screen. And I've actually built some features in write tool just by doing trial and error through ChatGPT and Claude, putting code and having built code for me and sent it to Mark. And then Mark goes back and goes, oh, this will work. And then he'll just put a little bit of his magic touch and then build that into the extension.

Hector Garcia: So you know how people using AI [00:07:30] to build software, it's kind of a big deal. And the example I gave is a very rudimentary one. Some companies have come back and and reported that they have ten x 20 x the speed of production because they've had, um, AI aid them in the process of building code. And the other super even more incredible layer to this is there already are like software in development that was [00:08:00] built 100% by AI, right? So like like the user went to AI and said, build me a piece of software that can do this one, two, three, four, all using basic English language. And the thing will build an entire layer of code where it doesn't require an actual developer to, like, tweak. It just literally just works of of, uh, of right off the bat that's already in the works. That's already like even with cloud, with some retail. Um, um, with some retail AI, you can go into its artifacts [00:08:30] screen and say, build me a pong game or bring me a Tetris game. And the thing will just go write all the code. And then literally without any tweaking, you'll be able to play the game immediately. So my so what do we do? We'll talk about predictions in a second, but that's already happening as we speak, which is actually pretty amazing.

Alicia Katz Pollock: Okay. So I want to kind of recap and summarize the three things so that I can make sure that that I'm clear on it. So the first use is what you're calling retail. And so those are the bots where we [00:09:00] as users go to the website and ask it for what we want, and it gives us back whether it's an image or turning an article into social media posts or, you know, we're asking it questions and it's giving us answers, like almost like a search engine. The second one is where they're building the AI into our software experiences, so that it's either getting predictive or automating some of the tasks inside [00:09:30] the software. So it's improving the software. And then the third one is kind of meta, where it's using AI to do the development of these other two things. Where we can use it to do the programing is that did I summarize that right?

Hector Garcia: Those are the three dimensions that we will discuss. There's a fourth dimension, sort of on the back end that we're not going to discuss too much, which is the hardware required to do all this. And there's a whole nother slew of development [00:10:00] that's happening with that. If you just look at the company, Nvidia, look at its history, his financial history in the past three years, it will tell you the story of AI hardware development. There's tons of hardware companies are fighting for compute. And what's what's even weird is that the power required by these computers is like more than the power than a lot of the things that we traditionally require power for and any of our machines. So, so [00:10:30] stuff like just computing AI, which is not very visual and tangible, is using more power than, you know, than big rigs that were, I don't know, doing automatic farming or whatever, like just like computing, like answering these questions through AI take more power than it takes to plow, you know, Acres and acres of fields. So it's just really amazing how what's happening in the world right now. There's going to be a huge fight for power resources, and that might be another discussion on the environment and all that stuff. But [00:11:00] we'll stick to the to the three that you summarized.

Alicia Katz Pollock: All right. Excellent. Yeah. I mean we've been increasingly using AI in royal wise and it really has increased our productivity. You know, whether it's I'm using ChatGPT, for example, to repurpose my writing. So if I have a transcript from a class, we can turn it into a blog article and then turn it into social media posts without me having to be the person doing all of the writing or [00:11:30] my voice. You know, there's some content out there that I didn't actually record that sounds just like me, and I can tell the difference, but you might not be able to. And so that's one of the things, some of the ways that royal wise realizes using that. Um, and, um, the we're also using it for that coding. You know, when we're working on the website, instead of having to struggle and trial and error as we're trying to make a [00:12:00] button work, we can just tell tell ChatGPT to give us the code, plug it in, and the button works. So it's really been a game changer for us.

Hector Garcia: Yeah, those are some some great examples of retail AI usage. So I'd like to talk about some of the predictions from 2025 that I think is going to happen, particularly in AI usage. So of course the models are going to get better. Uh, right now people, the AI can generate images really well [00:12:30] and it can create really short videos. And you've probably seen all over social media, all these little like ten second clips of like multiple things like, you know, like, uh, just imagine things that look like something from a Salvador Dali painting, like a very surreal looking things, you know, that you would imagine in the 60s where you were tripping, you know, on stuff, right? And you see all these things in social media being published because AI is now creating things out of nowhere. It's creating all these little small [00:13:00] video clips. I think in 2025, you're going to see long form content like videos, full videos of multiple minutes being fully created by AI, which means that within five years we'll probably see multiple. We'll see movies like full movies completely generated by AI, especially movies like the stuff that Pixar makes. So like companies like Pixar either are, you know, crapping their pants about, you know, what's going to happen or they're building, [00:13:30] you know, the AIS that are going to create these movies automatically.

Hector Garcia: But one thing I want to add to that is, you know, like right now, our kids, you know, go to YouTube and they pick, you know, Cocomelon or whatever. And this is how we hypnotize them for a little bit to watch all these animated videos. I think that 20 years from now, the parents are going to say, hey, my kid really likes soccer. And also he's into like Mexican food. And also they like pink and rainbows and and no unicorns. [00:14:00] And they like these type of artists. And can you create a custom TV show that contains all these components? And by the way, no scary stuff and, you know, do Spanglish. So my my my my my son starts getting some, some words in Spanish and words in English, you know, like it would be a point where parents actually custom create the TV shows that automatically get created. Now that's that prediction is for 20 years from now, but I it doesn't sound crazy at all that that could happen now.

Alicia Katz Pollock: It doesn't sound crazy [00:14:30] to me at all, because actually, one of the things that's happening in my own household is my son is an avid D&D player. Actually, both of us are, and he is developing an entire world and the you know, even a year ago, two years ago, he would have literally had to write the entire thing. And now he's able to say, all right, I need an element that, um, you know, I need a character that does this, and the paragraphs that it's spitting out are actually [00:15:00] eloquent and poetic and creative and beautiful, and I'm impressed. And I love the idea that he might actually be able to not just finish the project, but turn it into something that he could publish.

Hector Garcia: Well, now I'm big into dad, so I don't know the mechanics of the game with scoring and and power structure and that sort of thing. But I'm an avid board gamer in general, and I know [00:15:30] that what makes a good game, a just game in general, like I don't want to get off like the topic of accounting and that stuff too much, but what makes a game good is when the rules are clear And when the powers are balanced, right. And then the decisions that each player makes, um, it's a trade off. Like I'll, you know, I'll give away defense for offense or give away offense for defense type of stuff. And I assume that with the context of, of of a game design and balance [00:16:00] of power, and then using a real life example like back in reframe, uh, 2024, we had 18 tables and I had to use make sure that in every segment, people were swapping across the tables, and people with the same last names didn't were not in the same table to avoid families and and spouses to be in the same table and then tell ChatGPT, these are all my people. Can you design a table rotation where there's the most amount of new people talking to each other? So like balancing numbers [00:16:30] are very difficult for humans to do.

Hector Garcia: Like just having all at all times balancing of numbers is something that, um, ChatGPT can do really, really well at the same time for training purposes. When I go to my clients and I want to train them on, you know, QuickBooks, for example, I'll go to ChatGPT and say, hey, by the way, um, I have a client in Miami, and I want you to create a hundred vendor names with addresses that are in Miami that I can import into a fake QuickBooks file, [00:17:00] so I can have data that my client feels familiar with. And also, you know, they sell mostly to Brazil. So can you give me 100 names of typical Brazil people and addresses so I can load it in my in my fake QuickBooks file so that the training for them, like ChatGPT can. It's really good at randomizing, randomizing, balancing, and contextualizing information that we can find usefulness in the accounting world. You know, like just trying to draw some examples from the whole ND, uh, example from your son [00:17:30] that you gave.

Alicia Katz Pollock: Yeah. And so where do you see all of that going?

Hector Garcia: Well, what I one of the things that I think, um, companies like QuickBooks can do really well is to help us create custom sample files, right? So knowing the power of this retail and randomizing and storytelling device that this could be, you know, could I go to QuickBooks and say create fake sample file for me. And I would say, [00:18:00] okay, fake sample file for a marketing firm based in New York that does business in Canada and the UK, uh, throw a bunch of customers in there, throw a bunch of transactions in there, you know, create a story about what this business is, make sure that the transactions, when I pull the PNL, it kind of tells the story. And then I can use this to train people in accounting. Can you imagine sample files being built like this? I mean, and it's the AI has power to do this.

Alicia Katz Pollock: Yeah. [00:18:30]

Alicia Katz Pollock: As soon as you said that, like, oh my God, I just kind of like jumped out of my seat. I love the idea of being able to create a sample file that I can use for a class and be able to tell it all right, I need it to have three duplicate transactions. I need to have it have three receipts that somebody typed in the wrong dollar amount, and be able to actually create it without me having to actually sit here and think about and craft every single transaction [00:19:00] to have the examples that I need for what I want to teach. Oh my God, that would save me hours.

Alicia Katz Pollock: Love that.

Hector Garcia: So so so so the prediction that I have is that maybe someone will develop this in the next year. Maybe they maybe they listen to this podcast that were inspired. They can we could do it, you know, over a right to a world. We could do it. I don't know if I want to spend the time doing that, but but that this could be built like sample files could be created so powerful. So. And look, there's a huge market here [00:19:30] too, because if you if you think of third party software developers, they need sample data all the time. And then they and then you go to a conference and you have them showcase their software. And they have this really like cookie cutter sample data that contains, like Marvel characters or contains, um, Star Trek characters. And I find that fun and stuff. But like, people are going, oh, these guys are just playing around. This doesn't feel like a real thing. So imagine if if they'd come up. By the way, we have AI that builds random data so we can sample this. People go, oh, okay, [00:20:00] that's cool. You know, like it's just that, that just feels like a cool application of sort of like retail AI. And there could be a combination of retail AI. And then the API can push stuff into a QuickBooks file. So it's really not that far fetched.

Alicia Katz Pollock: Yeah, I think the trick would actually be getting it into a sample file, since that's one of the things that we have trouble with. Access is getting sample files that we can actually save as sample files, but that's a whole other topic. We can cross that bridge if we get the reality of being able to create these.

Hector Garcia: Right [00:20:30]or you could technically put it in a 30 day free trial account. Yeah. So so so my other prediction on on retail AI is that they will be very soon, probably this year. A lot of quote always on AI right? Always on AI. Like right now we we might not have the computing power for that. This is why it requires us to be in a chat, send a text, read a text, send a text, read [00:21:00] a text. Because by the way, the reading of the text is great because it reduces the amount of power that the the computer and the back end needs to do. So. Like it almost. It needs that break, right? For you reading what it says for you to write the next thing but always on AI is going to be a complete game changer. So let me give you an example of what what really subtle retail always on AI would look like. So imagine [00:21:30] I had a chain that had a little tiny device that was all the time listening to everything that I was saying, all the conversations I had with everyone. It would be recording it in my phone, probably through Bluetooth, and then maybe every hour. It could just like summarize what just happened in the hour, or it could summarize it, send an email to someone, or for God's sake, send my dad a text message to let me know what I'm up to so he doesn't have to [00:22:00] call me every hour to ask me what I'm up to.

Hector Garcia: So like, there's all sorts of there's all sorts of really powerful, always on AI type of things. And by the way, it's not that far fetched because we turn on location in our phones so we can track to see where our kids are all the time, and we're sort of comfortable with that. So why wouldn't we be? I mean, there's also implications of like knowing that that person's always recording everything because it changes the dynamic of the conversations and all that stuff. But I think that a lot of folks are going to turn into that because, [00:22:30] like, I get into so many fights with my wife about he said, she said, and I'm thinking, man, if I had this thing that was recording, I could play back to her to let her know that that's not what I said type of thing. So I think that a lot of people are going to be comfortable with this stuff because the practical applications are there.

Alicia Katz Pollock: Yeah, I mean, I've had that idea many times. It's like, God, I wish we could actually go back and find out what actually was said there. There was an episode of Black Mirror where they could do that. They could, like, blink their eyes and actually repay the video [00:23:00] of what? Of what was said. And that had a big creep factor to it also. But there you go. Now, I think that actually already exists, though, Hector. I mean, it's not an always on, but I saw a necklace advertised on Facebook that had a little button that you could press the button, and then it would start recording and then it would give you those summaries. It wasn't always on. And like you said, that a lot of this takes a huge amount of computing power. So it probably shouldn't be always on. But you didn't have to like tell [00:23:30] Siri to start it or anything, you could just click a button and go.

Hector Garcia: But yeah, but there's still there's still a button, right? That you just press a button and it still has to finish recording, transfer the voice and process it. What I mean by always on is, for example, I tell my thing to go, by the way, any single time that you hear somebody saying, remember, whatever the next things that come in after, remember, like my wife says, remember on Tuesday, we have to do this. Do me a favor. Just [00:24:00] go into my calendar and throw it in there. I'll go back after I say so. Somebody's telling me, remember Tuesday at three? We have to do whatever I go. Yes, I know that my always on machine is capturing that and putting it in my calendar automatically. That stuff is going to come very, very soon.

Alicia Katz Pollock: I definitely like that idea. Now I keep hearing about the term AI agents. Is that what you're referring to?

Hector Garcia: So, so so that's actually the exact next thing that was coming on my list, which is I think [00:24:30] that the always on AI is going to be the game changer for AI agents. So let me go backwards and explain what an AI agent is. And this is a very soft term. Like it could mean different things to different people. But the, the, the concept of an AI agent is we preprogram what we want it to do. We say, hey, you know, if this happens, then do that. If this happens, then do that. And it's sort of waiting [00:25:00] and listening for a command to, to do that. So in many ways, Zapier is actually an agent. Okay. It's actually not an AI agent. It's a it's an agent because Zapier, you tell it, you go, hey, every hour, read this spreadsheet. If you see a new line being added to the spreadsheet, go into QuickBooks and add the customer. Or every every five minutes, go into this database and if you see something change, go into my zoom webinar registration and change that registration. [00:25:30] Right. So like Zapier does this stuff like I'll read from one database and based on the condition you give it, change the other one when I agent will be will be just like that.

Hector Garcia: But you're not going to actually give it the parameters of what is, what triggers it and what needs to be updated. What you're going to give an AI agent is a goal. So if so, if so, with the example with the calendar, I say, hey, my goal is to make sure that every single thing that is told to me that needs to be [00:26:00] done fills my calendar. So and then the AI agent now has to extrapolate what information fills the calendar and what isn't right. Like if somebody comes in and goes, hey, I think on Thursday is Martin Luther King, uh, birthday. Like it won't put that in my calendar because it contextually knows that, like, I don't know Martin Luther King and we don't celebrate, you know, that person because it's not in my family. It's not. Doesn't have to be like an appointment. Right? So like, an AI agent would be smart enough to discern [00:26:30] from all the noise what actually is a triggering point and what is and where we Sapir you have to say, if it shows up on this database, then update it. So it needs to be like a line item on the database in order to show up.

Hector Garcia: This one could use natural language to turn to turn it into actions. That's what what an agent is. Um, and part of my predictions for 2025 is 2025 will be the year of agents. A lot more agent software will come out where you can describe what you wanted to do, and [00:27:00] then it will just do it. So like an agent that could happen in the software world is there could be an agent that reads QuickBooks feedback where somebody says, this is broken. Uh, the agent has the power to access QuickBooks, test it, realize that it is truly broken, read the code, read the error, and then put it on the to do list for the developers to go fix it. So that's an example of like an agent that Intuit could use [00:27:30] where they could actually automate and not just fix it. It would also suggest what code to add, and the agent would just have to actually add it and approve it. I mean, the the developer would have to add it and approve it at the end. So an agent will do that. It can it can grab natural language and know what to do with it.

Alicia Katz Pollock: Okay. Tell me if this is an example of that. The learning management system that I use to deliver all of our qbo courses hasn't the ability [00:28:00] that we can tell it to like, make a course about the Marvel heroes and it will actually structure and populate a course based on AI just at, you know, just build that course and then we use that framework to develop further. Is that an example of an agent?

Hector Garcia: No. Because you still because the the you're still telling it to build it, I think. I think a better agent would be [00:28:30] like I would say, let's say, for example, you have a bunch of people subscribed to to royal wise and then they can't find a particular solution, and then they go into a chat box and goes, how do I, you know, reverse a transaction where a check bounced? So the AI agent grabs the question and does maybe three things. One, it goes through all the transcripts of all your videos and says there are three videos that have a mention about a bounced payment. It might [00:29:00] be specifically what you want or not. And these are the these are the timelines in which it feels natural that the discussion started and ended. So try these three things, then go into maybe the QuickBooks knowledge database and go, by the way, based on the QuickBooks Knowledge database, these are the three things based on a Google search. These are the three possible things. If none of these satisfied you, I created a prompt that you can use to maybe ask your own eye or what to do. And I'm automatically sending, you know, [00:29:30] the Alicia, an email that says this, this was the question. These were the three suggestions. This is the, um, you know, where the gap could be and automatically put it in your development plan for the next course to create like an agent would do all of that, you know what I mean?

Alicia Katz Pollock: Okay. Yeah. Because I mean, some of what you said, we are already working on the rudiments of like a, a chatbot. We actually call it Ask Alicia where it has all the transcripts of all of the videos. And [00:30:00] somebody can put in a question like, how do you refund a credit memo? And then it just pulls out the answer based on the course and says, oh, it's over in this video over here, right.

Hector Garcia: That's an API. That's an API based application of AI, like it's a single function. You're saying, here's a question, read the text and then extrapolate the answer with the transcripts in here. So that wouldn't be an agent, because it's doing a single action based on a particular command, based on a predetermined amount [00:30:30] of things. I think an agent would have a little bit more autonomy and will solve the problem based on the natural language.

Alicia Katz Pollock: Yeah. I'm trying to wrap my my head around the difference of where that's more like the, you know, the API world that you started with versus an agent.

Hector Garcia: Well, an agent will use the API. So like so the agent is not like a third layer to this. An agent will be an agent for the most part will use APIs. Right. So it will [00:31:00] use a language model engine like OpenAI to understand the question. Remember the question still needs to be understood, which is the biggest challenge. The biggest challenge of this whole thing is not the actions to take. The actions to take is actually the easy part. The hard part is to extrapolate what actions should be taken based on the question or based on the prompt. Okay.

Alicia Katz Pollock: So so is it like me saying, um, like, you know, book an appointment with Brenda for [00:31:30] Thursday at ten and it goes and it books the appointment creates the the zoom link sends it out. And I didn't even have to, like, touch my computer to do it.

Hector Garcia: Yeah, that that could be one thing. But you're telling it specifically what what what to do. I think an agent would have a natural conversation with people, and then the agent would realize that they're not answering the question and therefore suggest to make an appointment with Brenda and then suggest the date that it's available. [00:32:00] And also emails Brenda. And let's let's them know that this question happened. It creates a tentative appointment, waits for the response to the question Brenda will ask, will answer something like, well, I'm not even sure. Is that even possible? The agent would understand that then needs to cancel the appointment and let the person know that we can't help them. You understand what I'm saying? Like. Like the agent. The agent extrapolates multiple tasks and multiple, um, multiple things [00:32:30] to do, and and it's not limited to A to just happening once. It's limited to a, a multitude of things that are happening in a period of time. That's kind of what an agent is.

Alicia Katz Pollock: So it's basically automating. It's giving. It's giving a new definition to the term virtual assistant.

Hector Garcia: I think in many ways an agent is a virtual assistant, right. And then you you tell it like you still give it limitations. Like you, you could give it access to your email or not. You could give it access to a certain spreadsheet [00:33:00] or not. You can give it access to QuickBooks or not give it access to Calendly or not like you. Obviously through the API, you decide what this agent has access to and then based on what it has access to, it decides what action to take and where to grab information or push information into.

Alicia Katz Pollock: Okay.

Alicia Katz Pollock: Yeah. So we're basically stating our intention and it's just going ahead and doing it for us. Correct.

Hector Garcia: Now my my last prediction here would be on the kind of a combination [00:33:30] of the API slash behind the scenes kind of AI that is happening. I think 2025 will see at least 10 to 12 real competitors to QuickBooks that were all built 100% by AI, with AI as its driving engine. So in many ways, like the developers won't even develop debits and credits and personnel and balance [00:34:00] sheet logic into the software, they will tell it this is accounting software. Go figure it out. And it literally builds pianos and balance sheets without, um, the developers actually even structuring it. You know, like from a design perspective, it just builds it. And then the software itself will not just build a panel and balance sheet. You would say, give me a panel with tons of details, or give me a panel with limited details, or give me a comparative panel that allows [00:34:30] me to focus on my biggest expenses, and it will know which expenses to expand, which ones to collapse, which ones to organize, maybe even regroup that regroup and rename the chart of accounts and reorder it based on the on on the description that the user gives it. So like my prediction is 2025. We'll see that way before Intuit can even like fathom the idea of doing this, because Intuit takes so long to develop anything and anything that they develop is sort of one dimensional, where all these software companies are building things, [00:35:00] using AI as the builder of the, of the, of the software as its core. So like, I'm going to predict, we're going to predict we're going to see tons of that in 2025.

Alicia Katz Pollock: That's interesting because, you know, there's there's a lot of people who have been dissatisfied with some of the the choices that Intuit is making in. Their direction and but are kind of locked into QuickBooks because there aren't any other software [00:35:30] that that equals it in terms of best in product and best industry. And so, you know, as much as I actually love using QuickBooks, I think having competitors that approach it in a different way, it might actually disrupt the whole industry.

Hector Garcia: Well, the real reason why I'm particularly thinking about QuickBooks, not just because our podcast is called the unofficial QuickBooks Accountants Podcast, but it is because I think QuickBooks is the the one software that everybody wants to disrupt [00:36:00] because like Photoshop for Adobe Photoshop was very difficult to disrupt, extremely difficult to disrupt. But Canva did it right. Um, and then you look at, you know, things like Microsoft Office Suite, very difficult to to to disrupt Google Drive and Google Sheets did it and for free. Like people don't even have to pay for it. Um, so like almost every single big name in the software industry has been [00:36:30] disrupted except for small business retail accounting. Like it's still a 80%, QuickBooks Live 5%, zero and 15%, you know, bunch of random programs out there. And I think people want a piece of that 80% power that QuickBooks have. And I think many software companies are going to try to go after I mean, they've been trying for years and they haven't been able to. Um, and I think they haven't been able to because of how entrenched QuickBooks [00:37:00] is in the accounting world.

Hector Garcia: And accountants trusted and or trusted to work in a trusted is a complex term, but trusted to work to at least produce a PNL and balance sheet that 95% of the times they can depend on. Okay, so like that's the kind of the the value proposition that QuickBooks has in the accountants hearts and minds at the moment. But because I allows people to build things completely reimagined, just conceptually completely reimagined, um, they will literally come out of left field with some things [00:37:30] that no one will will think of, like, like the best QuickBooks have come up with to make accounting easier is to call accounts categories instead of accounts. That's literally the best thing that Intuit has thought about. You know, when it comes to making QuickBooks, I mean, accounting easy for small business owners with I users don't even have to think about that. Like they'll build it in such a way that the user will type whatever nonsense they put in the chat box, and it'll understand what nonsense they're saying, and turn that into a real [00:38:00] financial statement.

Alicia Katz Pollock: That's really that's interesting. You know, I like the idea of having a whole different approach to it. You know, like at first we had ledgers and then we had spreadsheets, and then we had quicken, and now we had QuickBooks. And, you know, it's as time goes on, nothing. Nothing is permanent. So it'll be interesting to see if it is 2025 or if it does take [00:38:30] another 5 or 10 years. You know, I think it's going to take I think there'll be a longer development cycle until it's something that can rival everything that QuickBooks online brings to the table. But you know, it's you know, nothing is forever.

Hector Garcia: Yeah. For sure.

Hector Garcia: Now it will be most interesting to go back a year from now. Listen and watch this and go, wow, we were crazy. Or wow, we were right. I mean, I think it would be interesting either way. Probably the most interesting thing to look forward to the whole year is to see how [00:39:00] much of these predictions were, um, were actually actually came true. So anyway, with that being said, Alicia, what's going on in your world?

Alicia Katz Pollock: Well, this is my time when I am actually looking at how to implement AI inside royal wise. And we are. One of the things that we're doing is playing with language and multilingual. And so we are looking at ways of leveraging AI to deliver our content next year [00:39:30] in Spanish and possibly other languages as well. So that's one of the the biggest adventures that we are on right now, as well as, like I said, putting together a chatbot that allows me to access, you know, I've got probably 100 hours of qbo video transcripts, at least 100 hours, and, you know, another thousand pages of written material, and being able to synthesize all of [00:40:00] that into, um, ways that people can do research on QuickBooks using our content. So those are some of the things that we are working on. So that's my prediction for 2025 for royal wise.

Hector Garcia: Well, it's your own prediction per se, which is kind of cheating, right. Because you control that 100%. Um, well, what's going on in my, in our world, particularly within this topic, is, um, the folks that write tool and I, we're, we're thinking deep about how I [00:40:30] will affect the development of right tool. What more what more features we're going to add that are AI driven. You know, do we double down on AI now that Intuit is releasing more AI based tools? Do we develop in parallel? Do we just like give up on the AI stuff and go somewhere else? That's those are our deep thoughts in that area we did recently. I don't know if you got a chance to play with it. Alicia. We released a feature, um, in write tool. Where in bank feeds, there's a little button that says Ask Perplexity. [00:41:00] And basically, if you don't know what a particular expense category could be because you don't recognize the vendor, you click on perplexity and it opens up a pop up, and it searches that potential vendor name and perplexity. And it's much more powerful than using ChatGPT, because perplexity does sort of a combined combined AI plus Google type of search. So it actually combines the best of both worlds, not just the AI knowledge base, but also goes out and [00:41:30] googles it. So it does two things for you. It Googles it, uses the knowledge base and merges the two things to then suggest what potentially that could be for you. So it's just it's just really cool and it's just within this. So we're gonna we're trying to figure out what are the really cool, easy, low hanging fruit things we can build for developers that contain some sort of interaction with AI. Nice. Alrighty. So thank you very much for listening to this. One of the the last or maybe one of the last episodes of the year, and we'll see you in the next one. [00:42:00]

Alicia Katz Pollock: See you next year.