UiPath Agents - 4 Core Components

[00:00:00]Hello friends. Let's learn another new topic today.

[00:00:04]This is about UiPath agents.

[00:00:07]And we are going to see the four core components of an UiPath agent.

[00:00:14]Do you know what are those four components?

[00:00:16]Let me show you. The very first component is prompt.

[00:00:21]Now, what does this prompt means?

[00:00:23]On my right-hand side, while you are building agent, you would see there will be a system prompt and the user prompt. Basically, this is a natural language instructions that define the agent's role, goal, and constraints.

[00:00:40]So, this is the very first important core component through which the agent knows what is my goal, what is my role, and what exactly meaning what exactly the agent supposed to do. This is where you define the system prompt and the user prompt.

[00:00:56]Now, let's see the second core component.

[00:00:59]The second core component is the context.

[00:01:03]So, what is a context? What do you mean by context?

[00:01:06]What is a context? Context means information agent uses for decision.

[00:01:13]Such as knowledge bases, previous interactions, memory.

[00:01:17]For me to take a decision, let's take a small example. Let's say I have worked on an application. Let's say ServiceNow.

[00:01:25]My brain, my memory remembers what is that application, how how do I create an incident, how I create an RITM. Now, if anybody ask me, "Rakesh, can you do this on ServiceNow?" Then, because of my memory, because of I knew, I am able to do that action or I I know where exactly go and what to do.

[00:01:42]Similarly, before the agent could take any kind of an action or give you some kind of a data output or whatsoever you have asked the agent to do, if you provide him some knowledge bases, some previous interactions, some kind of a memory data, then using that memory data, it can take a better decisions. It knows what to be done.

[00:02:01]It's a kind of a context. It is also one of the important core components. So, prompt and the second core component is context.

[00:02:09]Let's see the third core component.

[00:02:12]The third core component is tools.

[00:02:15]So, here if you see, this is the tool.

[00:02:17]So, what what is the tool basically?

[00:02:20]For you to remember, the tool is basically hands of the agent.

[00:02:26]This is a brain. It is using ChatGPT, OpenAI. It is using Claude, Gemini behind the scene. That's the brain. Now, for it to take action, these are the hands of the agent. You can call it as hands. You know, just to remember things in a very easy way. Tools are hands of the agents.

[00:02:46]Now, actions the agent will take using tools such as invoking automation workflows. I can call a RPA workflow that I have built. And then I would using that RPA workflow, we will take some actions. Type into something. I'll get fetch certain data from somewhere.

[00:03:05]Using APIs, using other agents and robots. So, this is where you would be using the tools for taking actions.

[00:03:12]Clear? Now, let's see the fourth component. So, prompt we saw, context we saw.

[00:03:17]Third component we saw, tools. Let's see the fourth component.

[00:03:22]So, the fourth component is escalations.

[00:03:27]These are the escalations.

[00:03:28]Now, what does this escalation means?

[00:03:31]This is where it brings the human in the loop. For example, you have asked the agent to draft an email. It has drafted the email, but this email is going to a higher person and you want a human to review it before it is being sent by the agent automatically.

[00:03:47]So, here you would bring the human in the loop or the concept is called HITL.

[00:03:51]What is this called? HITL.

[00:03:54]How do you do that? You can use apps and you know using the action center apps, you can show our drafted message and then human can review and you can have a button approve reject things like that.

[00:04:06]You can use a messaging channel to show things. All these things can be done. So for approval, for review or assistance when needed.

[00:04:16]All right, this is where you use this.

[00:04:18]Now let's also see you remember not all tasks are agent friendly. So we will see what are the task which are agent friendly and what are the task which which are poor use cases.

[00:04:32]So what are the good use cases poor use cases? So good use cases.

[00:04:36]Drafting an email, drafting a message, summarizing content, ticket triage, categorizing the tickets. First pass customer interaction, you know, first interaction you want to do hello, we got your ticket, we got your email, we acknowledge your email and we are working on so and so. So those kind of customer interactions can be done. Now what are the poor use cases? Wherever there is a high risk or a high accuracy is needed.

[00:04:59]High risk and high accuracy needed, these generally becomes a poor use case.

[00:05:04]So for example, high risk financial transactions.

[00:05:08]Legal work.

[00:05:10]Regulatory work, complex data processing without a deterministic safeguards.

[00:05:14]So this all becomes poor use cases.

[00:05:17]I hope this content has given you some light on the core components of the agent and also to know about the good use cases and poor use cases. Thank you for watching. Please do let me know in the comment section what else would you like to learn in my upcoming videos.