Rethinking the Browser Experience with Agentic AI. Phase 1
I’ve been using AI tools like ChatGPT more than ever, and I’ve become fascinated by how these systems work behind the scenes to translate complex queries into human-digestible information. This curiosity led me to start a new self-initiated UX project focused on agentic AI and browsers, a fast-moving and innovative space.
The Challenge.
The race to build the next-generation AI-powered browser is underway. OpenAI, Google, Microsoft, and Perplexity are each exploring agentic models, where AI doesn’t just respond, but acts on your behalf. Rumors suggest OpenAI will soon release a browser built around these ideas.
As a challenge to myself, I’m simulating a design sprint as if I were on OpenAI’s design team. My goal: create an agentic AI web browsing experience from scratch, then compare it to what OpenAI actually releases.
Initial sketch completed to kickstart project.
Personal Project Goals.
Deepen my understanding of agentic AI, user-agent workflows, and AI design principles.
Practice integrating AI into both UX process (via tools like FigmaAI, ChatGPT) and UX outcomes (via interaction flows with AI agents).
Explore and define how to build trust in AI systems — especially in areas involving privacy, transparency, and autonomy.
Identify and design multiple critical user journeys for final design concept.
Iterate heavily based on user feedback, from early concepts to mid- and high-fidelity prototypes.
Deliver and reflect on a full UX case study that bridges design, research, and speculative tech.
Agile Project Plan.
To help map out this project for myself, I created a project plan. Please note that due to my part-time job at Menards, portfolio updates, applying & interviewing for positions, and other life events — this plan did not follow the allocated time frame.
Project Plan.
Key KPI's to Hit.
In order to measure the success of this project, I created three KPI's to achieve:
Ensure users can complete core tasks with AI.
Reduce the time it takes to complete a task.
Reduce # of clicks it takes to complete a task before and after new browser design.
Measure how helpful users find the AI browser (likert scale).
The Current AI browser Market.
With ChatGPT by my side, I Identified the current advantages and disadvantages of current AI browser experiences to understand how AI is currently transforming the browser experience.
Comparative analysis displaying the advantages and disadvantages of AI powered web browsers.
Key Learnings:
Hybrid models, a mix of web control and AI co-piloting, enable traditional browser interaction with an added AI layer.
Contextual AI memory and session continuity (knowing where you left off) is still immature across most platforms.
Automation potential is untapped, only some (like Arc and Perplexity) hint at task chaining or content generation workflows.
AI enables many new features (assistant, task automation, research, etc.) to the web browser experience.
How to Design for AI Agents.
According to Microsoft, an AI "agent" as an AI assistant is designed to execute tasks, working with or for humans. The components of an agent include instructions, knowledge, actions, skills, memory, etc. Agents may have chat experiences but generally go beyond typical chatbots.
Diagram displaying how AI agents are leveraged.
Agentic AI systems (or “agent systems”) are made up of one or more agents. They can autonomously identify, plan, and take actions on behalf of an individual user, a group, or other agents from a general set of instructions with limited direct human supervision. An AI system may be considered “agentic” if it is autonomous, has more complex capabilities, operates in a more complex environment, and can scale to more domains.
Diagram showing how a main (core) agent connects to specialized agents.
The core of this browser is to understand and accelerate a user’s intent on the web by routing tasks to specialized AI agents — letting the user browse less and do more.
Reflects Natural Human Behavior
People talk to one trusted guide (the core agent), not 10 individual assistants.
Ex: You don’t want to say “Open Research Agent,” you just ask, “Can you help me find good laptops under $1,000?”
Simplifies Interaction Complexity
Keeps the front-facing UX clean.
Users don't need to know which agent does what.
Specialized agents operate behind the scenes, routed automatically via the core.
Scales Well with Tasks
New agents can be added modularly (e.g., a “Scholar Agent” for academic research).
Allows for continuous upgrade without changing the interface.
Supports Trust and Consistency
One voice and tone across all actions.
The core maintains memory/context across tasks (vs each agent being siloed)
I also looked into Microsoft's UX Design for AI Agents to identify principles that I must follow to create a successful and feasible design.
Diagram displaying Microsoft's UX-Agent AI design principles. Space, Time, Core.
I will later ensure these principles are achieved in my final design.
Interviewing to Uncover User Browser and AI Behavior.
I had many important questions that needed answers before I started to design.
What does a user's actual web workflow look like? What tools do they rely on to browse and synthesize information? How does they feel about web privacy, trust, and the growing role of AI?
To explore these questions and validate (or challenge) my initial assumptions, I conducted user interviews to better understand real behaviors and sentiment.
Overview of questions asked during user interviews.
After conducting my interviews, I practiced using ChatGPT to synthesize my results:
I uncovered diverse browsing behaviors and attitudes toward privacy, trust, and AI.
Three participants relied on multi-tab workflows for tasks like weather monitoring, collectibles price comparisons, and troubleshooting, revealing pain points around clutter and inefficiency.
Four out of five expressed frustration with irrelevant or sponsored results, underscoring a need for more relevant filtering.
Privacy views were split: 40% rejected personalization outright, 40% accepted it if helpful, and 20% held conditional acceptance, pointing to opportunities for transparent privacy controls.
Finally, AI literacy varied widely—some users integrated ChatGPT daily into their work, while others had little exposure but showed curiosity—highlighting a gap between experienced and novice users that could be bridged with clearer onboarding and contextual AI support.
Mapping out CUJs for Design.
Weather Tracking & Persistent Tabs
Shopping Research and Price Comparison
Troubleshooting & Learning via Forums and Video
Casual AI use for Summarization and Recommendations
What's Next?
Next, I will be validating these flows with users and design wireframes to begin user testing. After I get the wireframes into an intuitive interface I will explore adding additional novel features to the AI web experience.
Why This Matters (to me and the field).
This is not just a project about AI — it’s about preparing for the future of UX. As designers, we’re entering a new frontier where user expectations, mental models, and workflows are being redefined by AI. I’m doing this project to grow as a designer in that future, learning how to adapt both my process and thinking to match what’s coming next.
