Human, Amplified: Agentic AI That Knows When to Stay Quiet
What if the smartest AI in the room knew when to pause, to let the human lead?
That question guided my approach in designing a unified AI agent for HUMAN customer service reps.
The goal: help professionals handle calls faster and more confidently, without ever making them feel replaced.
Role: Lead Product Designer
Team: Cross-functional with 3 AI capability groups, researchers, and business leads
Outcome: Unified 3 siloed AI initiatives into a single, experience which is now the foundation for enterprise-wide AI customer service tools. Reduced tool-switching time in simulated tests by ~40%.
Challenges
-
Advisors were juggling 200+ legacy apps while on live calls
-
3 different AI initiatives existed in silos all competing for visibility
-
Teams misaligned on vision, and user outcomes.
What I Did
Lead design across the 3 ai streams
Created the design strategy, prototypes, and north-star roadmap
Helped align 3 AI teams and 20+ stakeholders toward a unified vision.
The Goal
Reduce handle time and increase client satisfaction.
Keep the human at the centre of the AI experience.
The user's POV:
A bit about our users: these CX professionals are on the phone, with an actual living client, listening to their problem and looking for a solution in a vast sea of ugly old.. cough.. disconnected applications AND to add insult to injury they are tasked with promoting a company offer to these clients on the same call. I already have a headache thinking about it.
Imagine a client calls you because they are unhappy with a service and you have to solve the issue AND also sell them something???! Yikes.
Here's a super super simple version of the journey for demo purposes. >
New call
Listen
Search
Summarize
Upsell
Resolve
Switching to tech and business speak for a second...
Three specialized teams were working on complementary AI-driven capabilities:
one focused on information search and retrieval, another on personalized recommendations, and a third on live conversation insights.
Each product was being developed independently, with its own roadmap, priorities, and definition of success.
My challenge as the design lead was to bridge these silos and align vision, user experience, and the hierarchy of importance across all three products. This meant synthesizing their capabilities into one cohesive flow, AND guiding each team toward a shared understanding of what should take precedence for the advisor experience (which required a fair amount of persuasion along the way 😫).
As it is with a lot of real world projects the textbook “Design Process” was out the window.
What we did to keep our sanity + what we learned:
Call listening
We listened in on 100+ recorded calls between advisors and clients (these calls had screen recordings from the advisor’s desktop, Jackpot!) it helped us see real use cases, real frustrations both on the client and the advisor side.
Lesson: App overload distracts advisors.
When we first watched the advisor screen recordings, it was chaos. Tabs flickering, multiple logins, and even Post-It notes on screens (all while actively listening to a customer). That moment made it clear: before we talk AI, we need to talk focus.while
Early Rapid prototyping
The fastest way to get the conversation going from vague requirements to concrete ones was to visualize what the solution could be. We couldn't wait for requirements because of the complexity of the project and the misalignment of the 3 teams.
Lesson: It's not enough to talk things out.
What the prototypes did was to bring everyone in the room, show them their own pre-cooked ideas and MINE and get all of us to talk about a single vision instead of being in our own heads.
Interviewing the customer service reps
We got talking with these Customer service reps and created a user journey map that was vetted with multiple stakeholders. We also showed them our rapid prototypes to get knee jerk reactions on what would be useful to them.
Lesson: Noise is a huge problem.
Noise was a big painpoint as these advisors had to focus on the client and fish through a lot of old apps to find a solution and implement it. We got great feedback on our rapid prototypes and already had areas we could improve on.
MVP to North Star
Even though the technologies we were designing for was considered cutting edge, we still had a lot of technical limitations. because of that I decided to use Backcasting to get to an agreed upon north star prototype, (and stress countless times that this isn't going to happen tmrw) and create multiple sign posts all the way back to the MVP to hash out all the iterations of this product it would take for us to get to the ideal.
I personally loved how this little strategy helped stakeholders and developers be more acceptable and less scared about the "out there" and futuristic idea I had for the North Star.
.png)
What I learned as a Design Lead
Lead through alignment: When teams disagree, prototypes become the common language.
Design for confidence, not just efficiency: The best AI augments human judgment, not replaces it.
Embrace ambiguity: Some of the best design work happens before the requirements exist.