Agentic UI - Views from a UI Engineer

After nearly two and a half years of satisfying work as a software engineer at Adobe, I’ve found myself traversing a new, AI-centric work environment. Adobe is no stranger to AI integration, and my work over the past few months has centered heavily on developing an agentic UI experience, where a user collaborates with an AI agent to achieve their goals more efficiently. Although I’m not tackling any low-level LLM integration or backend development, I work as a UI engineer who helps to bring fresh, agentic UI experiences to the end user. I’m excited to share some insights I’ve gathered so far across this new breadth of this work.

What I’ve observed: how can AI help or hinder the customer?

A common introductory question when working with AI agents is, what does the customer really want? As we’ve collectively ventured into new, uncharted territory within the tech sphere, plenty of AI experiences exist that customers simply don’t prefer to use. As opposed to the routine workflows somebody is familiar with, users may find themselves burdened by agentic experiences that don’t supply a satisfactory experience.

However, when used appropriately, various AI features can certainly make an agentic experience worthwhile. For example, time-saving tools can entirely transform a user’s experience by eliminating tedious and menial tasks. Additionally, agentic workflows frequently lessen user fatigue in the UI, such as reducing repetitive clicks and keystrokes. Due to the current mixed reception of AI tools, teams should carefully gauge the tradeoffs and benefits of implementing an agent, especially if it will be mandatory for customer use. As a precursor to any agentic development, it is crucial to understand what an agentic experience can provide that a traditional experience cannot.

Challenges I’ve encountered during agentic UI development

Up until this summer, I had only worked on traditional, non-agentic UI development. However, agentic UI catapulted to the foreground of my day-to-day tasks as new initiatives arose in my scope of work. So far, developing agentic UI has been a challenge, primarily relating to the collective novelty of this experience. It has been unusual, yet formative, to be thrown into the deep end with my senior-level coworkers; we’ve all needed to learn together. Adapting to the differences in agent-centered code repositories and frameworks has also been very challenging. New development ecosystems have been built to account for the rising complications in agentic development, and it’s our job to acclimate to them.

A baseline difference I’ve observed in agentic UI development is the need to be concerned with a collaborative chat experience—you typically don’t want to let the user to wander away from the chat without a good reason. This means UI features that may typically open in new windows, new pages, or in full-screen dialogs would instead need to be viewable in more compact ways. For example, using a resizable side-panel next to an AI chat helps to keep the chat UI in the foreground. Though I don’t sit in the meeting room as a UX designer, chiming in as a developer can be crucial in shaping UX outcomes as we work together as a team.

Aligning backend and frontend teams based on AI responses has also been more challenging than usual. Due to the volatile nature of AI-generated responses, a very strong data “contract” must be established with frontend and backend teams. Though AI responses may differ, the goal is for teams to firmly agree on expectations for the data shape that the backend sends and the frontend receives. The end goal of our UI team is to build dynamic components that can assume the form of these various AI responses. Once this data contract has been established between teams, the process of working with generated data in the UI becomes significantly smoother, allowing us to streamline the development of new, innovative deliverables.

Wrap-up

This experience working with agentic UI has required some adjustment, but it has been a great introduction to a new area of software development. I’m honored to be working on such an impactful feature just a few years into my career. It’s also important to mention that I’m very grateful to have such a supportive team around me at Adobe. I am aware that agentic UI will play a large role in the future of UI development, and being an adopter of this initiative at work is fantastic preparation for the future of my career.