Designing for Agentic AI: Moving Beyond Prompts into Intelligent Collaboration
Over the past few years, the design industry has experienced what feels like a generational leap. Generative AI changed how we create — but agentic AI is changing how we collaborate.
When I led product design for service and support experiences at Wayfair, I witnessed firsthand how automation and AI could transform post-order operations — from handling customer inquiries to optimizing warehouse routing. But the real inflection point wasn’t when AI started generating responses. It was when AI began making informed decisions autonomously — prioritizing tasks, resolving tickets, and learning from human feedback loops. That’s agentic design in action.
From Command to Collaboration
Most current AI systems rely on prompt-based control. Humans issue commands; AI executes. But as models gain autonomy, the role of design shifts from interface creation to defining agency and intent.
When I design for agentic systems, I start by mapping degrees of autonomy:
Assistive: AI helps complete a user’s action.
Advisory: AI recommends actions based on data.
Agentic: AI acts on its own, within human-defined constraints.
The third layer introduces the need for trust scaffolding — transparency, context awareness, and accountability. A well-designed agent doesn’t just “do” things; it communicates why and learns responsibly.
Principles of Designing for AI Agency
Design for Delegation, Not Dictation
Users should feel empowered to assign responsibility, not lose control. This means creating permission structures and “explainability moments” that help users trust automation.
Transparency Builds Credibility
Every autonomous action must have a visible rationale. When an AI agent reschedules a delivery or auto-approves a return, the reasoning — and the option to override — should be crystal clear.
Feedback Is the Interface
Unlike static systems, agentic AI relies on continuous learning. Feedback isn’t just data; it’s the UI through which trust is negotiated.
Ethical Boundaries Are Design Features
Designers now shape behavior loops, not just visuals. We’re responsible for encoding fairness, empathy, and inclusivity into systems that act without human supervision.
Case Example: AI-Powered Post-Order Experience
At Wayfair, we experimented with integrating AI agents to handle repetitive support flows — like shipment delays or damaged orders. Initially, these agents operated in “advisory” mode, suggesting actions for human agents. Over time, we evolved them into semi-autonomous systems that could act on behalf of humans within pre-set boundaries (e.g., issuing refunds under a certain dollar threshold).
The design challenge wasn’t technical; it was psychological. How could we make users — both internal service reps and customers — feel confident in outcomes driven by machines?
We redesigned the experience around contextual visibility: whenever the AI took an action, it surfaced a simple, human-readable rationale (“This order qualifies for auto-refund due to transit damage confirmed by carrier”). That design pattern became a trust anchor — people felt informed, not replaced.
"
The future of AI in design isn’t about control — it’s about collaboration. When humans and machines co-create, we unlock possibilities neither could achieve alone.
The Future of Agentic Design
As agentic systems become more embedded, design will move from “interface-first” to “intent-first.” We’ll design relationships, not screens. The best experiences will feel like partnerships — human and AI co-creating toward shared goals.
For design leaders, this means expanding the skill set: from craft to cognition, from flows to frameworks. We’ll be asked not just to design products but to design systems of accountability between humans and intelligent agents.
And that’s where I see the next horizon — building design cultures that understand how to orchestrate agency responsibly. When we get that balance right, AI won’t just make design faster — it will make it more human.
