Agentic AI: The Intersection of UX and AI
Through Voiceflow, it allows designers to learn how to create agentic AI without code; a prominent form of AI emerging today.
Insights
Oct 10, 2025
First, let's discuss what are agentic experiences, and what sets it apart from traditional chatbot experiences?
Agentic experiences are a new class of human-computer interaction. With modern-day LLMs, we are now able to generate responses dynamically and use multiple data sources to inform those responses.
Defined by the autonomy, how it adapts, and its goal-driven process, there are 2 kinds of agentic AI: One-way agents, where the bot recites predetermined content (not generative AI) and two-way agents, where the bot interprets input and crafts a personalized response (generative AI). These systems, known as AI agents, are designed to perform complex, multi-step tasks with minimal human intervention. In contrast, a traditional chatbot is limited by its predefined rules and consists of a lack of judgment. An AI agent can make informed decisions and act independently to achieve a goal.
Pre-LLM agent chatbots feel very robotic because of the limits of information generation. The latest roll-out of agentic design rests in multi-agent experiences, where there are multiple handoffs between specialists. An example is a ride-sharing app. The app itself can consist of a ride-hailing agent, a trust/safety agent, a customer support agent, etc.
2. Let's describe the agent I designed using Voiceflow
My agent is a personalized car shopping assistant to help customers search, identify, and determine the right vehicle for them. The agent starts off by collecting customer preferences on a type of vehicle that they may be interested in. The prompt states: Start the conversation in a friendly, stylish tone. Ask the customer what type of vehicle they're shopping for (e.g. sedan, SUV, sports car, truck, etc.), any specific brand or feature preferences, their ideal price range, and if there’s a primary use case (e.g. family, work, hobby). It makes it simple, easy, and streamlined for the customer in the car buying market to share their preferences and ideal features in a car, so the agent can make personalized recommendations that fits their user needs.
After the initial start, it then proceeds into clarifying the preferences, searches for vehicles that aling with the customers preferences (e.g. brand, model, style, features) through their in-depth database of vehicles, and then gives a list of recommended vehicles that fits their needs. Lastly, it offers the ability to save or email the vehicle recommendations to the user.
Explaining the underlying design of my agent.
My agent's design is focused on guiding a customer through the process of the car shopping experience and help the customer find and receive personalized vehicle recommendations. It moves beyond a simple Q&A format to an intelligent, goal-oriented system. It aids the customer in a more efficient scenario, helping the customer find what they want, and saving them time.
What tasks/steps did the user need to do prior to this agent experience? | What value does the agent provide? What tasks can it complete? | What value does the agent provide? What tasks can it complete? How does the user interact with the agent? | What information is required to enable this functionality? | |
|---|---|---|---|---|
User Need 1 | Calculate a vehicle need. The user had to have a specific need for a car, such as "I need a family car," or "I'm looking for a sports car" | Collect customer preferences. | The user provides information through conversational and natural language format | A list of all possible preference categories (e.g., vehicle type, make, model, budget range). |
User Need 2 | Manually search through different websites to find cars that match their criteria. This is often a time-consuming. | Search and recommend vehicles. | The agent presents the recommendations to the user in a clear manner. | An up-to-date database of vehicles. |
User Need 3 | Remember or write down the details of the cars they liked, and then figure out how to get that information sent to them for future reference. | Offer to email or save recommendations. | The agent makes a direct offer to the user who has to accept or decline. | The agent needs access to the user's email address, which is likely collected during the conversation. |
Narrated Demonstration
Some current limitations and opportunities for my agentic experience?
I tend to view the limitations of the Agentic AI being too reliant/dependant on the prompts. It required way more questions than I originally have thought, as if I already knew what I wanted. Additionally, it tends to still feel “robotic” at the time, due to the repititveness nature of the flows. Overall opportunities that could be presented is more open-ended experience, assuming what I may like already, and provide smaller responses, to feel more human-like rather than a more intelligent chatbot.
Agentic AI: The Intersection of UX and AI
Through Voiceflow, it allows designers to learn how to create agentic AI without code; a prominent form of AI emerging today.
Insights
Oct 10, 2025
First, let's discuss what are agentic experiences, and what sets it apart from traditional chatbot experiences?
Agentic experiences are a new class of human-computer interaction. With modern-day LLMs, we are now able to generate responses dynamically and use multiple data sources to inform those responses.
Defined by the autonomy, how it adapts, and its goal-driven process, there are 2 kinds of agentic AI: One-way agents, where the bot recites predetermined content (not generative AI) and two-way agents, where the bot interprets input and crafts a personalized response (generative AI). These systems, known as AI agents, are designed to perform complex, multi-step tasks with minimal human intervention. In contrast, a traditional chatbot is limited by its predefined rules and consists of a lack of judgment. An AI agent can make informed decisions and act independently to achieve a goal.
Pre-LLM agent chatbots feel very robotic because of the limits of information generation. The latest roll-out of agentic design rests in multi-agent experiences, where there are multiple handoffs between specialists. An example is a ride-sharing app. The app itself can consist of a ride-hailing agent, a trust/safety agent, a customer support agent, etc.
2. Let's describe the agent I designed using Voiceflow
My agent is a personalized car shopping assistant to help customers search, identify, and determine the right vehicle for them. The agent starts off by collecting customer preferences on a type of vehicle that they may be interested in. The prompt states: Start the conversation in a friendly, stylish tone. Ask the customer what type of vehicle they're shopping for (e.g. sedan, SUV, sports car, truck, etc.), any specific brand or feature preferences, their ideal price range, and if there’s a primary use case (e.g. family, work, hobby). It makes it simple, easy, and streamlined for the customer in the car buying market to share their preferences and ideal features in a car, so the agent can make personalized recommendations that fits their user needs.
After the initial start, it then proceeds into clarifying the preferences, searches for vehicles that aling with the customers preferences (e.g. brand, model, style, features) through their in-depth database of vehicles, and then gives a list of recommended vehicles that fits their needs. Lastly, it offers the ability to save or email the vehicle recommendations to the user.
Explaining the underlying design of my agent.
My agent's design is focused on guiding a customer through the process of the car shopping experience and help the customer find and receive personalized vehicle recommendations. It moves beyond a simple Q&A format to an intelligent, goal-oriented system. It aids the customer in a more efficient scenario, helping the customer find what they want, and saving them time.
What tasks/steps did the user need to do prior to this agent experience? | What value does the agent provide? What tasks can it complete? | What value does the agent provide? What tasks can it complete? How does the user interact with the agent? | What information is required to enable this functionality? | |
|---|---|---|---|---|
User Need 1 | Calculate a vehicle need. The user had to have a specific need for a car, such as "I need a family car," or "I'm looking for a sports car" | Collect customer preferences. | The user provides information through conversational and natural language format | A list of all possible preference categories (e.g., vehicle type, make, model, budget range). |
User Need 2 | Manually search through different websites to find cars that match their criteria. This is often a time-consuming. | Search and recommend vehicles. | The agent presents the recommendations to the user in a clear manner. | An up-to-date database of vehicles. |
User Need 3 | Remember or write down the details of the cars they liked, and then figure out how to get that information sent to them for future reference. | Offer to email or save recommendations. | The agent makes a direct offer to the user who has to accept or decline. | The agent needs access to the user's email address, which is likely collected during the conversation. |
Narrated Demonstration
Some current limitations and opportunities for my agentic experience?
I tend to view the limitations of the Agentic AI being too reliant/dependant on the prompts. It required way more questions than I originally have thought, as if I already knew what I wanted. Additionally, it tends to still feel “robotic” at the time, due to the repititveness nature of the flows. Overall opportunities that could be presented is more open-ended experience, assuming what I may like already, and provide smaller responses, to feel more human-like rather than a more intelligent chatbot.
Agentic AI: The Intersection of UX and AI
Through Voiceflow, it allows designers to learn how to create agentic AI without code; a prominent form of AI emerging today.
Insights
Oct 10, 2025
First, let's discuss what are agentic experiences, and what sets it apart from traditional chatbot experiences?
Agentic experiences are a new class of human-computer interaction. With modern-day LLMs, we are now able to generate responses dynamically and use multiple data sources to inform those responses.
Defined by the autonomy, how it adapts, and its goal-driven process, there are 2 kinds of agentic AI: One-way agents, where the bot recites predetermined content (not generative AI) and two-way agents, where the bot interprets input and crafts a personalized response (generative AI). These systems, known as AI agents, are designed to perform complex, multi-step tasks with minimal human intervention. In contrast, a traditional chatbot is limited by its predefined rules and consists of a lack of judgment. An AI agent can make informed decisions and act independently to achieve a goal.
Pre-LLM agent chatbots feel very robotic because of the limits of information generation. The latest roll-out of agentic design rests in multi-agent experiences, where there are multiple handoffs between specialists. An example is a ride-sharing app. The app itself can consist of a ride-hailing agent, a trust/safety agent, a customer support agent, etc.
2. Let's describe the agent I designed using Voiceflow
My agent is a personalized car shopping assistant to help customers search, identify, and determine the right vehicle for them. The agent starts off by collecting customer preferences on a type of vehicle that they may be interested in. The prompt states: Start the conversation in a friendly, stylish tone. Ask the customer what type of vehicle they're shopping for (e.g. sedan, SUV, sports car, truck, etc.), any specific brand or feature preferences, their ideal price range, and if there’s a primary use case (e.g. family, work, hobby). It makes it simple, easy, and streamlined for the customer in the car buying market to share their preferences and ideal features in a car, so the agent can make personalized recommendations that fits their user needs.
After the initial start, it then proceeds into clarifying the preferences, searches for vehicles that aling with the customers preferences (e.g. brand, model, style, features) through their in-depth database of vehicles, and then gives a list of recommended vehicles that fits their needs. Lastly, it offers the ability to save or email the vehicle recommendations to the user.
Explaining the underlying design of my agent.
My agent's design is focused on guiding a customer through the process of the car shopping experience and help the customer find and receive personalized vehicle recommendations. It moves beyond a simple Q&A format to an intelligent, goal-oriented system. It aids the customer in a more efficient scenario, helping the customer find what they want, and saving them time.
What tasks/steps did the user need to do prior to this agent experience? | What value does the agent provide? What tasks can it complete? | What value does the agent provide? What tasks can it complete? How does the user interact with the agent? | What information is required to enable this functionality? | |
|---|---|---|---|---|
User Need 1 | Calculate a vehicle need. The user had to have a specific need for a car, such as "I need a family car," or "I'm looking for a sports car" | Collect customer preferences. | The user provides information through conversational and natural language format | A list of all possible preference categories (e.g., vehicle type, make, model, budget range). |
User Need 2 | Manually search through different websites to find cars that match their criteria. This is often a time-consuming. | Search and recommend vehicles. | The agent presents the recommendations to the user in a clear manner. | An up-to-date database of vehicles. |
User Need 3 | Remember or write down the details of the cars they liked, and then figure out how to get that information sent to them for future reference. | Offer to email or save recommendations. | The agent makes a direct offer to the user who has to accept or decline. | The agent needs access to the user's email address, which is likely collected during the conversation. |
Narrated Demonstration
Some current limitations and opportunities for my agentic experience?
I tend to view the limitations of the Agentic AI being too reliant/dependant on the prompts. It required way more questions than I originally have thought, as if I already knew what I wanted. Additionally, it tends to still feel “robotic” at the time, due to the repititveness nature of the flows. Overall opportunities that could be presented is more open-ended experience, assuming what I may like already, and provide smaller responses, to feel more human-like rather than a more intelligent chatbot.