The problem
Adobe’s Customer Journey Management suite helps enterprise marketers orchestrate experiences across email, mobile, web, and paid channels. As the product team began integrating generative AI capabilities, the design challenge wasn’t “how do we add AI”, it was a harder question: how does a human stay in control of AI acting at scale?
When an AI agent can draft, schedule, and send a campaign to millions of users across a dozen channels, the surface where the human gives direction becomes load-bearing. Get it wrong and you have a liability. Get it right and you’ve built the thing that makes AI actually usable in enterprise contexts.
What I designed
The work centers on what I’ve called strategic intent orchestration, the surfaces where a marketer’s high-level intent (a goal, a constraint, a judgment) gets translated into coordinated AI behavior.
This included:
- AI command surfaces, where marketers direct AI agents with natural language, structured inputs, and explicit approval gates
- Intent-to-action transparency, showing users what the AI understood, what it did, and where it made decisions that warrant human review
- Governance scaffolding, controls that let enterprise teams audit AI actions, define policy constraints, and override at any level of abstraction
- Confidence and uncertainty signaling, visual language for communicating what the AI is confident about vs. where it’s reasoning under uncertainty
The design challenge
The hardest part of this work isn’t the interaction design, it’s the conceptual work of figuring out the right level of abstraction. Too much control and you’ve just built a complicated form. Too little and you’ve built something no enterprise compliance team will approve.
The frame I kept returning to: design for the person who needs to answer for the outcome, not just the person who triggered the action. In enterprise software, those are often different people. The system needs to produce an artifact, a record of intent, decision, and execution, that is legible to both.
What it means for the practice
This engagement is live, ongoing work at the intersection of AI capability and enterprise governance. The design problems here are genuinely new, there’s no established pattern library for “how does a human supervise an AI agent operating across millions of touchpoints at once.”
That’s what makes it worth doing.