← BACKOVERVIEW

CAPITAL ONE · AI / AUTOMATION · PROTOTYPING

Re-Engage

Designing what happens when AI starts acting — a system for giving an AI sales agent enough autonomy to be useful, while keeping people in control of when, how, and why it acts.

HUMAN–AI ORCHESTRATION — DESIGNING WHO ACTS, WHEN, AND UNDER WHAT CONDITIONS
ROLE
Product Design Leader
FOCUS
AI · Automation · Prototyping

Re-Engage explored a different kind of interface: one that doesn't always wait for a person to click something. The system could autonomously engage dealership leads, which shifted the design problem from creating screens to defining behavior.

The challenge became deciding how much autonomy the system should have, what context and constraints should govern its actions, and how humans could step in without breaking the continuity of the experience.

01/THE PROBLEM

A settings request became an information architecture problem.

What initially looked like configuration for an AI capability quickly expanded into a connected system of communication controls, messaging channels, lead rules, compliance requirements, notifications, and templates. The real question underneath it: how do you design the controls behind an AI sales agent — one with enough autonomy to be useful while people still govern how it behaves?

02/MY APPROACH

I treated autonomy as a system of states, constraints, and handoffs.

Instead of starting with individual screens, I modeled the relationship between the AI, the dealership team, and the customer — including who had control at each point and what caused that control to change.

  • Model the actors — clarify what the AI could do, what the human needed to control, and what the customer experienced
  • Define the states — explicit operating modes: AI active, human active, AI paused, and follow-up scheduled
  • Design the handoffs — how a person intervenes, and how the system behaves before, during, and after
  • Create the constraints — cadence, channels, lead rules, compliance, and when not to act
  • Prototype behavior early — functional prototypes that made autonomous behavior tangible while the system was still being defined
THE OPERATING MODEL — SYSTEM FLOW ABOVE, STATE MODEL BELOW, WITH CONTROL MOVING BETWEEN AI AND HUMAN

It changed how I designed it.

AI-assisted functional prototypes moved much earlier into the design process. Instead of asking Product, Design, and Engineering to react to diagrams describing how the system might behave, we could react to something that actually behaved.

Prototyping became part of the thinking — not just a way to demonstrate the answer after we’d found it.

03/THE SOLUTION

The interface became a control system for autonomy.

Re-Engage brought the controls governing AI behavior into a coherent system — giving teams ways to shape how the agent communicates, what context it uses, when it follows up, and when a person should take over.

  • Agentic AI — designing for software that can initiate actions rather than simply respond
  • Human on the loop — keeping people able to supervise, intervene, and take control when needed
  • Context + constraints — defining the information and rules that shape autonomous behavior
  • AI evaluation — creating ways to understand whether the system is behaving as intended
  • Functional prototyping — making behavior tangible before the complete production system exists
  • Cross-team alignment — giving Product, Design, and Engineering something concrete to react to together
CONTROL SURFACE (ABSTRACTED) — CADENCE, CHANNELS, COMPLIANCE, AND HUMAN TAKE-OVER

04/IMPACT

A vague idea of “AI settings” became a shared model for how the system should behave.

Making autonomy tangible gave Product, Design, and Engineering a common language for discussing behavior, intervention, constraints, and scale while the product was still taking shape.

  • Shared behavior model — Product, Design, and Engineering could reason about the same system
  • Reusable patterns — decisions about autonomy and intervention could extend beyond a single interaction
  • Faster learning — functional prototypes let teams evaluate behavior earlier

REFLECTION

Designing AI changed what I considered an interface.

Sometimes the interface is the decision the system makes when no one is touching it.

Autonomy becomes useful when people understand where it starts, where it stops, and how to step in.

I stopped using prototypes only to show ideas. I started using them to think.

NEXT PROJECT

Trade-In

Turning trade-in uncertainty into a shared understanding between shoppers and dealers before they meet.