Designing Response Systems for Agentic AI Products.
Topics:
Agentic Experience
AIX Designer
AI/UX Design
Gen AI Experience
Anthony Todd
We’re entering a new age of how users consume content. Aside from designing traditional UX, I can also design agentic experiences that will exist in the ecosystems of other users and businesses. This year, I led the design of an AI tool that simplified our site experience, protected proprietary content from scraping, and ultimately be accepted into Google’s AgentSpace.
My role is to define the experience of the response, and how data was presented. I’m able to map how and when data should be surfaced, and present what considerations would be the most helpful for sourcing from content archives. In order to benefit our product, I’ve planned ways to identify intent signals from the user’s conversation, and encouraged sign-ups with persuasive value exchanges while aligning with the goals of my product and engineering teams.
I'm open to joining a product-driven team bringing AI experiences to real users in interesting ways. I’m able to design intuitive, explainable, and trustworthy AI experiences that align with both user needs and product goals. I can contribute my experience planning the foundations for UX strategy, human-AI interaction, and help bring it all to life with cross-team collaboration.
Sound good? Let’s connect.
Case Study
How are users responding?
User One
User Feedback
“I was still thinking when it just... did the thing.”
Why are they saying this?
Hypothesis: The agent is operating on task completion, not task readiness. Users are still in an evaluative mindset, but the AI is interpreting silence as approval. Ethical intent markers in our logs show a significant portion of regret-based prompts happening just after agent execution, suggesting a lack of consent in timing.
User Two
User Feedback
“I didn’t even realise it had started.”
Why are they saying this?
Hypothesis: Visual feedback is either too subtle or absent. Many agent actions, especially those that run in the background, start without strong interface cues. When users can’t visually track what the AI is doing, they lose the opportunity to intervene early.
User Three
User Feedback
“I couldn’t find how to undo it, so I just redid everything manually.”
Why are they saying this?
Hypothesis: There’s no obvious off-ramp. Once something is done, users are either forced to live with it or retrace their steps. Session recordings showed that in 42 percent of post-agent interaction flows, users deleted or rebuilt the output from scratch instead of editing it.
Where can we help users first?
First Priority
What part of the user journey?
Agent Engagement Phase (immediate response window)
Potential Strategy?
Design for Interruptability Before Completion:
Allow users to see the agent's intent before it commits to action. This can be done through preview states, timers, or visible confirmations. It’s important not to reduce speed, but rather to introduce an interrupt layer that respects the user's own rhythm without creating a lag.
Second Priority
What part of the user journey?
Mid-task Interaction
Potential Strategy?
Introduce Persistent “Pause and Revert” Controls:
Designing an always-visible pause or “revert to before AI” button gives users psychological safety. These are not panic buttons. They are creative affordances that allow a human to reassert authorship when the AI moves too far ahead.
Third Priority
What part of the user journey?
Reflection and iteration Phase
Potential Strategy?
Embed Agent Transparency in the Interface:
Many users didn’t know why the agent did what it did. By making the AI’s rationale, rules, or assumptions visible within the UI (even if only in expandable summaries) we could actually reduce surprise. This also helps train user mental models to align with agent logic, especially over repeated use.
What are users saying?
Best Case
What outcome are we hoping for?
Users feel more understood and begin exploring new playlists regularly, leading to a 25% increase in average session length and a 30% uptick in discovery playlist engagement. Positive word-of-mouth about “emotional accuracy” drives organic growth and differentiation in the competitive streaming market.
Worst Case
What outcomes should we try to avoid?
The agent becomes too passive, requiring manual confirmation for every small task. This slows workflows down and leads to a perception of inefficiency. Trust doesn’t improve, and now neither does speed.
Being Prepared
What should we be prepared for?
Test different thresholds for interruptibility. Not every action needs a pause, but those with irreversible or emotional weight definitely do. Build analytics around friction points and conversation-based regret signals to tune timing models over time.
Product Design
Full Time Work
Portfolio Samples
I'm currently open to full-time roles and freelance collaborations. If my work feels like a good fit for your team or project, feel free to reach out! Also, follow along on my social channels to see what I’m working on next.
Creating
and
collaborating
locally
in
Crested
Butte,
Colorado.
Anthony Todd’s Accumulating
Collection of Design Work
AI Experiences
Full Time Work
Designing Response Systems for Agentic AI Products.
Topics:
Agentic Experience
AIX Designer
AI/UX Design
Gen AI Experience
Anthony Todd
We’re entering a new age of how users consume content. Aside from designing traditional UX, I can also design agentic experiences that will exist in the ecosystems of other users and businesses. This year, I led the design of an AI tool that simplified our site experience, protected proprietary content from scraping, and ultimately be accepted into Google’s AgentSpace.
My role is to define the experience of the response, and how data was presented. I’m able to map how and when data should be surfaced, and present what considerations would be the most helpful for sourcing from content archives. In order to benefit our product, I’ve planned ways to identify intent signals from the user’s conversation, and encouraged sign-ups with persuasive value exchanges while aligning with the goals of my product and engineering teams.
I'm open to joining a product-driven team bringing AI experiences to real users in interesting ways. I’m able to design intuitive, explainable, and trustworthy AI experiences that align with both user needs and product goals. I can contribute my experience planning the foundations for UX strategy, human-AI interaction, and help bring it all to life with cross-team collaboration.
Sound good? Let’s connect.
Case Study
How are users responding?
User
One
User Feedback
“I was still thinking when it just... did the thing.”
Why are they saying this?
Hypothesis: The agent is operating on task completion, not task readiness. Users are still in an evaluative mindset, but the AI is interpreting silence as approval. Ethical intent markers in our logs show a significant portion of regret-based prompts happening just after agent execution, suggesting a lack of consent in timing.
User
Two
User Feedback
“I didn’t even realise it had started.”
Why are they saying this?
Hypothesis: Visual feedback is either too subtle or absent. Many agent actions, especially those that run in the background, start without strong interface cues. When users can’t visually track what the AI is doing, they lose the opportunity to intervene early.
User
Three
User Feedback
“I couldn’t find how to undo it, so I just redid everything manually.”
Why are they saying this?
Hypothesis: There’s no obvious off-ramp. Once something is done, users are either forced to live with it or retrace their steps. Session recordings showed that in 42 percent of post-agent interaction flows, users deleted or rebuilt the output from scratch instead of editing it.
Where can we help users first?
First
Priority
What part of the user journey?
Agent Engagement Phase (immediate response window)
Potential Strategy?
Design for Interruptability Before Completion:
Allow users to see the agent's intent before it commits to action. This can be done through preview states, timers, or visible confirmations. It’s important not to reduce speed, but rather to introduce an interrupt layer that respects the user's own rhythm without creating a lag.
Second
Priority
What part of the user journey?
Mid-task Interaction
Potential Strategy?
Introduce Persistent “Pause and Revert” Controls:
Designing an always-visible pause or “revert to before AI” button gives users psychological safety. These are not panic buttons. They are creative affordances that allow a human to reassert authorship when the AI moves too far ahead.
Third
Priority
What part of the user journey?
Goal-oriented Exploration Phase
Potential Strategy?
Embed Agent Transparency in the Interface:
Many users didn’t know why the agent did what it did. By making the AI’s rationale, rules, or assumptions visible within the UI (even if only in expandable summaries) we could actually reduce surprise. This also helps train user mental models to align with agent logic, especially over repeated use.
What are users saying?
Best
Case
What outcome are we hoping for?
Users feel more understood and begin exploring new playlists regularly, leading to a 25% increase in average session length and a 30% uptick in discovery playlist engagement. Positive word-of-mouth about “emotional accuracy” drives organic growth and differentiation in the competitive streaming market.
Worst
Case
What outcomes should we try to avoid?
The agent becomes too passive, requiring manual confirmation for every small task. This slows workflows down and leads to a perception of inefficiency. Trust doesn’t improve, and now neither does speed.
Being
Prepared
What should we be prepared for?
Test different thresholds for interruptibility. Not every action needs a pause, but those with irreversible or emotional weight definitely do. Build analytics around friction points and conversation-based regret signals to tune timing models over time.
Product Design
Full Time Work
Portfolio Samples
I'm currently open to full-time roles and freelance collaborations. If my work feels like a good fit for your team or project, feel free to reach out! Also, follow along on my social channels to see what I’m working on next.
Creating
and
collaborating
locally
in
Crested
Butte,
Colorado.
Anthony Todd’s Accumulating
Collection of Design Work
AI Experiences
Full Time Work
Topics:
Agentic Experience
AIX Designer
AI/UX Design
Gen AI Experience
Anthony Todd
We’re entering a new age of how users consume content. Aside from designing traditional UX, I can also design agentic experiences that will exist in the ecosystems of other users and businesses. This year, I led the design of an AI tool that simplified our site experience, protected proprietary content from scraping, and ultimately be accepted into Google’s AgentSpace.
My role is to define the experience of the response, and how data was presented. I’m able to map how and when data should be surfaced, and present what considerations would be the most helpful for sourcing from content archives. In order to benefit our product, I’ve planned ways to identify intent signals from the user’s conversation, and encouraged sign-ups with persuasive value exchanges while aligning with the goals of my product and engineering teams.
I'm open to joining a product-driven team bringing AI experiences to real users in interesting ways. I’m able to design intuitive, explainable, and trustworthy AI experiences that align with both user needs and product goals. I can contribute my experience planning the foundations for UX strategy, human-AI interaction, and help bring it all to life with cross-team collaboration.
Sound good? Let’s connect.
Case Study
How are users responding?
User
One
User Feedback
“I was still thinking when it just... did the thing.”
Why are they saying this?
Hypothesis: The agent is operating on task completion, not task readiness. Users are still in an evaluative mindset, but the AI is interpreting silence as approval. Ethical intent markers in our logs show a significant portion of regret-based prompts happening just after agent execution, suggesting a lack of consent in timing.
User
Two
User Feedback
“I didn’t even realise it had started.”
Why are they saying this?
Hypothesis: Visual feedback is either too subtle or absent. Many agent actions, especially those that run in the background, start without strong interface cues. When users can’t visually track what the AI is doing, they lose the opportunity to intervene early.
User
Three
User Feedback
“I couldn’t find how to undo it, so I just redid everything manually.”
Why are they saying this?
Hypothesis: There’s no obvious off-ramp. Once something is done, users are either forced to live with it or retrace their steps. Session recordings showed that in 42 percent of post-agent interaction flows, users deleted or rebuilt the output from scratch instead of editing it.
Where can we help users first?
First
Priority
What part of the user journey?
Agent Engagement Phase (immediate response window)
Potential Strategy?
Design for Interruptibility Before Completion:
Allow users to see the agent's intent before it commits to action. This can be done through preview states, timers, or visible confirmations. It’s important not to reduce speed, but rather to introduce an interrupt layer that respects the user's own rhythm without creating a lag.
Second
Priority
What part of the user journey?
During and Immediately After The Task Completion Phase
Potential Strategy?
Introduce Persistent “Pause and Revert” Controls:
Designing an always-visible pause or “revert to before AI” button gives users psychological safety. These are not panic buttons. They are creative affordances that allow a human to reassert authorship when the AI moves too far ahead.
Third
Priority
What part of the user journey?
Reflection and iteration Phase
Potential Strategy?
Embed Agent Transparency in the Interface:
Many users didn’t know why the agent did what it did. By making the AI’s rationale, rules, or assumptions visible within the UI (even if only in expandable summaries) we could actually reduce surprise. This also helps train user mental models to align with agent logic, especially over repeated use.
What’s the potential impact?
Best
Case
What outcome are we hoping for?
Users gain a sense of co-creation instead of being overridden. Task trust improves. Regret-based undo behavior drops by 50 percent, and users report higher satisfaction when working with autonomous agents.
Worst
Case
What outcomes should we try to avoid?
The agent becomes too passive, requiring manual confirmation for every small task. This slows workflows down and leads to a perception of inefficiency. Trust doesn’t improve, and now neither does speed.
Being
Prepared
What should we be prepared for?
Test different thresholds for interruptibility. Not every action needs a pause, but those with irreversible or emotional weight definitely do. Build analytics around friction points and conversation-based regret signals to tune timing models over time.
Product Design
Full Time Work
Graphic Design
Freelance Work
I'm currently open to full-time roles and freelance collaborations. If my work feels like a good fit for your team or project, feel free to reach out! Also, follow along on my social channels to see what I’m working on next.
Creating
and
collaborating
locally
in
Crested
Butte,
Colorado.