HCDE Directed Research
Designing the Future of Human-Agent Interaction
Designing the Future of Human-Agent Interaction



A UW directed research group designing the future of human to AI agent interaction.
Project overview
Project overview
Project Overview
A UW directed research group designing the future of human to AI agent interaction.
A UW directed research group designing the future of human to AI agent interaction.
AI agents, systems that can take action over the web, are ushering in the future of AI and UX innovation. They can learn our preferences and act on our behalf, but they also risk acting against our instructions. When that happens, today's interfaces offer little control: you can watch or cancel. Timeline control introduces a third option: scrub back through the agent's plan, branch into an alternate path, and merge the best outcomes.
This project emerged from a 10-week Directed Research Group at UW, led by Kevin Feng and David McDonald, exploring design challenges in human-agent interaction. Work from this DRG may directly impact tech companies building agents. I worked the first 5 weeks synthesizing 20+ research papers, then 5 weeks designing, from sketches to user flows to wireframes to interactive prototypes to the high-fidelity videos shown here. All work is my own, refined through weekly critiques with PhD researchers and peer designers who each pursued independent concepts.
AI agents, systems that can take action over the web, are ushering in the future of AI and UX innovation. They can learn our preferences and act on our behalf, but they also risk acting against our instructions. When that happens, today's interfaces offer little control: you can watch or cancel. Timeline control introduces a third option: scrub back through the agent's plan, branch into an alternate path, and merge the best outcomes.
This project emerged from a 10-week Directed Research Group at UW, led by Kevin Feng and David McDonald, exploring design challenges in human-agent interaction. Work from this DRG may directly impact tech companies building agents. I worked the first 5 weeks synthesizing 20+ research papers, then 5 weeks designing, from sketches to user flows to wireframes to interactive prototypes to the high-fidelity videos shown here. All work is my own, refined through weekly critiques with PhD researchers and peer designers who each pursued independent concepts.
AI agents, systems that can take action over the web, are ushering in the future of AI and UX innovation. They can learn our preferences and act on our behalf, but they also risk acting against our instructions. When that happens, today's interfaces offer little control: you can watch or cancel. Timeline control introduces a third option: scrub back through the agent's plan, branch into an alternate path, and merge the best outcomes.
This project emerged from a 10-week Directed Research Group at UW, led by Kevin Feng and David McDonald, exploring design challenges in human-agent interaction. Work from this DRG may directly impact tech companies building agents. I worked the first 5 weeks synthesizing 20+ research papers, then 5 weeks designing, from sketches to user flows to wireframes to interactive prototypes to the high-fidelity videos shown here. All work is my own, refined through weekly critiques with PhD researchers and peer designers who each pursued independent concepts.
Role
Role
Role
UX Designer/
Researcher
(Sole designer)
UX Designer/
Researcher
(Sole designer)
UX Designer/
Researcher
(Sole designer)
Team
Team
Team
1 Designer
2 Research Leads
3 Design Peers
1 Designer
2 Research Leads
3 Design Peers
1 Designer
2 Research Leads
3 Design Peers
Duration
Duration
Duration
Research- 5 weeks
Design- 5 weeks
Research- 5 weeks
Design- 5 weeks
Research- 5 weeks
Design- 5 weeks
YEAR
YEAR
YEAR
September- December (ongoing)
September- December (ongoing)
September- December (ongoing)
the design
Timeline Control for ChatGPT agent
Timeline Control for ChatGPT agent
Timeline Control for ChatGPT agent
An innovative interface that gives users the power to scrub, branch, and merge an AI agent's plan: transforming passive oversight into active collaboration.
An innovative interface that gives users the power to scrub, branch, and merge an AI agent's plan: transforming passive oversight into active collaboration.
An innovative interface that gives users the power to scrub, branch, and merge an AI agent's plan: transforming passive oversight into active collaboration.
When the agent's plan goes wrong, scrub back and branch into a new direction.
When the agent's plan goes wrong, scrub back and branch into a new direction.
When the agent's plan goes wrong, scrub back and branch into a new direction.
You're planning a wedding in Hawaii, but you wonder what it would be like to have a wedding in Japan? Instead of starting over, you scrub the timeline back to when all the planning stages were set. The agent branches into a new plan, preserving everything that came before while exploring in a different place.
You're planning a wedding in Hawaii, but you wonder what it would be like to have a wedding in Japan? Instead of starting over, you scrub the timeline back to when all the planning stages were set. The agent branches into a new plan, preserving everything that came before while exploring in a different place.
You're planning a wedding in Hawaii, but you wonder what it would be like to have a wedding in Japan? Instead of starting over, you scrub the timeline back to when all the planning stages were set. The agent branches into a new plan, preserving everything that came before while exploring in a different place.
Switch between timeline branches and inspect exactly what the agent did at each step.
Switch between timeline branches and inspect exactly what the agent did at each step.
Switch between timeline branches and inspect exactly what the agent did at each step.
With multiple branches now in play, you need visibility and understanding. Quickly switch between branches and click on the branches to understand its reasoning. Expanding the plan history lets you view the plans in higher detail, see which steps are complete versus pending. You stay in control without micromanaging.
With multiple branches now in play, you need visibility and understanding. Quickly switch between branches and click on the branches to understand its reasoning. Expanding the plan history lets you view the plans in higher detail, see which steps are complete versus pending. You stay in control without micromanaging.
With multiple branches now in play, you need visibility and understanding. Quickly switch between branches and click on the branches to understand its reasoning. Expanding the plan history lets you view the plans in higher detail, see which steps are complete versus pending. You stay in control without micromanaging.
Merge the best outcomes from different branches into one synthesized plan.
Merge the best outcomes from different branches into one synthesized plan.
Merge the best outcomes from different branches into one synthesized plan.
Your branches explored different vendors and venues. Now you want the photographer from the Main Branch, not the one from the Japan Branch. Timeline merging lets you review and compare the best aspects across each branch. Using agent's system of auto-resolving conflicts and producing a single, optimized plan.
Your branches explored different vendors and venues. Now you want the photographer from the Main Branch, not the one from the Japan Branch. Timeline merging lets you review and compare the best aspects across each branch. Using agent's system of auto-resolving conflicts and producing a single, optimized plan.
Your branches explored different vendors and venues. Now you want the photographer from the Main Branch, not the one from the Japan Branch. Timeline merging lets you review and compare the best aspects across each branch. Using agent's system of auto-resolving conflicts and producing a single, optimized plan.
LET'S CHAT
I'm currently writing the full story behind this project, as it's ongoing.
I'm currently writing the full story behind this project, as it's ongoing.
I'm currently writing the full story behind this project, as it's ongoing.
The final product, research synthesis, early sketches, lo-fi wireframes, and iteration cycles that led here are in the works. In the meantime, if you'd like to learn more about my process or discuss the design decisions, I'd love to chat.
The final product, research synthesis, early sketches, lo-fi wireframes, and iteration cycles that led here are in the works. In the meantime, if you'd like to learn more about my process or discuss the design decisions, I'd love to chat.
The final product, research synthesis, early sketches, lo-fi wireframes, and iteration cycles that led here are in the works. In the meantime, if you'd like to learn more about my process or discuss the design decisions, I'd love to chat.