What is it?
Agentic experiences or AX design involve AI working alongside designers and developers in real time, as it automates a variety of different tasks (Mischler, 2025). Traditional chatbots are more of a step-by-step process; they follow a prompt that is given, but can’t continue on their own without human instruction. AX, on the other hand, is more collaborative and can do a lot of tasks behind the scenes without constant human oversight and work with complex systems (IBM Technology, 2024). As Kat Holmes brings up in her post, “one strength that an AI agent brings to the workplace is its ability to learn across vast amounts of information, and find effective ways to solve problems” (Holmes, 2025).
It can help accelerate the design process, allowing designers to work with it while also placing more focus on other areas. She adds that designers ensure that these agents created are focusing on human-centered results, understand the interaction between the AI and user, and when to have humans interfere (Holmes, 2025). Designers are able to use and connect multiple agents to work together, pushing great advancements and boosting efficiency, which allows businesses to stand out (Mischler, 2025). AX is something that is becoming increasingly involved, which makes it more important than ever for designers to explore further, experiment, and adapt.
What I’ve designed with VoiceFlow
Experimenting with VoiceFlow, I wanted to create a chatbot named Sam for an imaginative company I called “HelpU,” with tasks to help users find, compare, and show the best deals and discounts on medication from various stores/pharmacies. VoiceFlow was able to form a collaborative team of agents, which includes the initial greeting, clarity of the user’s request/preferences, searching and comparing options, further information, and wrapping up. There are also exit conditions if the user isn’t clear with their requests, if there aren’t any suitable options that the user is requesting, or if the user no longer wants to continue. There are two ways of communication, either through chatting or by call, which in this situation chat system works more effectively, but it can depend on the user.
Through a few revisions by editing these agents, I’ve added a disclaimer and tried to reduce the amount of dialogue in certain areas that seem unnecessarily long. I also wanted to explore further the features that can be adjusted for each agent, such as toggling on the buttons for options that users can choose from if they are uncertain what to do, and carousels to display the medications and deals that users can easily compare.
The Design of Sam the Copilot for “HelpU”
| User Needs | 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? | How does the user interact with the agent? | What information is required to enable this functionality? |
| Find deals on over-the-counter medications or prescriptions. | Look at different sites online, browse stores or pharmacies for deals, or be unaware of discounts and programs. | Centralizes deals and processes in one place. Look for current prices and possible deals based on the store. | A chat system that users can personally message, select options, or even call. | Access to sites and up-to-date resources from different stories or pharmacies, GoodRX, programs, etc. |
| Get personalized recommendations. | Browse and locate the medications either online or in-person. | Filter deals based on the user’s preferences. | Chat system with carousel function. | User interaction and preferences and past engagements. |
| Financial assistance for prescriptions and out-of-pocket costs. | Looking for plans or deals to afford their medication. | Quickly find possible programs that can cover cost or provide financial assistance. | Chat system with carousel function to display these programs. | Access to sites and up-to-date resources. |
| Translation | Might ask, use tool to translate everything, or simply avoids or purchase the provide without enough context. | Can easily communicate and explain to the user. | Chatbot and/or phone call | By how the user messages. |
| Stay updated on time-sensitive deals or programs. | Keep an eye on multiple stores/sites, mail, or they don’t check until they need it. | Potential Value: if the agent is connected to an actual site in which users have their own profiles, these can be personalized and timely notifications | Chat notifications. | Connection to a site or app, profile, user’s past interactions, and preferences. |
Demonstration of my Agentic UX
The Limitations & What can be Improved
As I interacted with the agent, there were limitations that arose. First is the adjustments being made on my end. One must close the chat for it to fully process, and there were buggy moments when the agent started elongating its words or even misspelling some text when I asked it to be more brief. It’s very text-heavy, including when calling, listing the options, and even with the added carousel to compare deals. When I make these changes, I have to completely restart the chat and message until I get to a specific point at which I can see whether the changes work. The carousel and button functions, at times, don’t work or aren’t used as they are meant to. At times, it takes too long to process or doesn’t respond.
It’s also hard to undo drastic changes, and if I want to go back to a specific point. Since the agent isn’t connected to an actual site, it has a limited context of past convo if the patient chatted before. Sometimes, if I ask about the price for a particular medication at a certain store, it states that it can’t access the site, but previously showed deals. If it provides links to a certain site for a medication, it directs the user to just the homepage of that site. Last but not least, the AI voice is limited; eventually, a user will start to recognize it from different sites.
Some ways that it can be improved are to have real-time updates, have a section where we can manually change the AI voice, better ways to connect and access other various sites for information, being able to edit multiple agents in a workflow at once, have the agent remember past conversations, and better organization of prompts with pre-made templates that users can look through better.
Overall, VoiceFlow is interesting, and I enjoyed being able to see how we as designers can create our own virtual assistance. It still has its limitations and issues, but it has a lot of potential to evolve and improve.
Citations
Holmes, K. (2025, May 20). Welcome to a new era of experience with AI agents. Salesforce. https://www.salesforce.com/blog/agent-experience-design/
IBM Technology. (2024, November 18). AI Agents and AI Assistants: A Contrast in Function. YouTube. https://youtu.be/IivxYYkJ2DI
Mischler, H. (2025, August 13). How to embrace the great UX paradigm shift to agentic experience design. Salesforce. https://www.salesforce.com/blog/ux-shift-to-agentic-experience-design/
MARLEN SARMIENTO 
Leave a comment