Isaac from Opsa.co breaks down what makes an AI agent different from basic automation. An agent needs three things: it uses your tools, picks which ones to use and in what order, and learns from past interactions. In accounts receivable, a real agent decides which invoices to chase, writes personalized emails, and escalates on its own.
Chapters
Every week a new accounting firm owner tells me that they want AI agents, and when we sit down to talk about their use case, they don't need one. Hey, this is Isaac from Opsa.co, and in this video I want to share what an AI agent actually is, how it differs from an automation, and actual examples in an accounting firm environment for you to understand. So in the next few minutes, I'm going to show you the 3 criteria I use to test and tell the difference and walk you through some real accounting workflows. To show you all three in action. So let's talk about the definition. Most people think an AI agent is just like a traditional ChatGPT chatbot you talk to, and it can be, but it has a couple of things that make it different. Number one, it's that it takes action in the tools you give it. So if you give it access to your inbox, it can send an email on your behalf. If you give it access to your calendar, it can create a bin on your behalf. If you give it access to QuickBooks or Zero, it can create invoices or customers. Or vendors on your behalf. It can do useful things in the world. That is number one. But that alone doesn't really make it an agent. It has to have 2 other criteria and an optional 4th one. So the 2nd criteria, it is autonomous. What does that mean? From the tools that it has, it reasons and it thinks which tools to use and when. If you think about a traditional automation that executes exactly what you tell it to do, an agent, it's deciding what to do based on the command you've given it based on the goals you've set for it. The third criteria, it's adaptive, right? And what that means is that if it tries and fails, it's able to learn from that and try again. So what makes it adaptive is 2 things, right? It has like 2 sides of the same coin, which is adaptive means when it fails, it's able to remember and learn from it. So the learning component, but also the memory, which is it can read past context, past actions, and from that determine what's the best course of action from there. In summary, if an AI solution, if a digital solution or program can act on your behalf, pick from a toolbox which and when tool to use, it'll learn from its failures and it remembers the past, then it qualifies as an AI agent. Now let's talk about a simple example in the AR or accounts receivable or collections department that might better represent what's the difference between an automation an AI workflow, which is an automation with a little bit of AI, and then a full-blown AI agent just for illustration purposes. An automation would be something like, hey, when an invoice, it's 30 days overdue, send an email to the contact, prospective contact with a payment reminder and with this specific email template. And it doesn't personalize it or anything like that. It's reliable. It's gonna happen every time as long as the data is there and it's somewhat rigid, but it's gonna follow exactly what you told it to do. The example of an AI workflow might be, hey, I want you to pull the aged account receivables on a regular basis, send that to a large language model like a GPT or a Claude from Anthropic, and give me back a prioritized list of actions and some personalized emails to send to the people to make those collections. So it's a little bit of a smarter output. The process is still fixed and it's gonna Follow those steps, but now the AI is interpreting and personalizing the output it's giving you. If we were to envision instead an AR agent, you can set it to run every morning or every week, and it will review the AR report for various clients. And based on that, if you just tell it to just perform the AR function, and that's the prompt and no more than that, it's going to decide which invoices are worth chasing. It's going to look at the amounts, the date, outstanding, perhaps the payment history of that particular supplier or vendor, and it's going to determine whether this person is worth following up with and what to say depending on the amount, the days outstanding, and it might vary based on that. It might even choose to escalate to a human if it's perhaps a non-payer that has been, has shown to not pay on a regular basis, perhaps the balance is too large. The important thing is that you're not telling it specifically what to do. It is reasoning, creating a plan, remembering from past and payment histories in the past what action to take and what to do to fulfill its command or its goal. It's more intelligent but also has more room to grow, and that's sort of the trade-off— smarter but more dangerous at the same time. Now, knowing the difference between these 3 things is step 1. Step 2 is knowing Which agent can actually be helpful for your firm and how to actually go ahead and build one from scratch. And that's the answer of my next video where I show you and walk you through how to build your first AI agent. So check out that video next. And until then, stay efficient.
