Frontline operations managers at enterprise companies like FedEx run on 6–8 separate systems — HR for headcount, scheduling tools for coverage, compliance dashboards, email threads. When a surge event hits — Mother's Day, peak season, a last-minute volume spike — they manually cross-reference all of it, build their own analysis, and make high-stakes staffing decisions without a complete picture. The operational cost: understaffing, compliance gaps, and missed SLAs.

ServiceNow's bet was to build an AI-powered employee workspace that eliminates that gap entirely — not a chatbot bolted onto existing tools, but a new surface built around an AI agent that knows your context, runs multi-step analysis proactively, and turns insight into action. I was part of a large, cross-functional product team brought in to shape that vision from strategy through pixel — defining the interaction model for how this new kind of AI agent should behave.

The fragmented operations stack a FedEx distribution manager navigates daily — each system a separate login, a separate mental model, and zero shared context

My contribution spanned strategy and execution. At the strategy level, I pushed for anchoring the entire surface around a single design question: what does a manager need to see at 7am before their shift starts? That framing shaped the Home dashboard as a proactive briefing surface — an Inbox of pending decisions, a contextual alert card surfacing the highest-priority business signal, and a Canvas for personal context they'd actually use daily.

At the execution level, I was deep in the pixels — prototyping the Otto agent's conversational thread, designing the step-transparency model (how the AI shows its 5-step reasoning in real time without overwhelming the user), and working through the visual language that separates "Otto thinking" from "Otto acting." Research validated our iterations through six rounds of usability testing; I partnered with the research team to turn findings into pixel-level decisions the same week they came in.

Left: Early information architecture explorations — mapping the "7am brief" and what proactive surface hierarchy earns manager trust on day one. Right: Otto's agentic step model — showing AI reasoning in-thread without overwhelming the user with process.

I drove the visual framework for this — designing two distinct modes in a single thread: "Otto thinking" (analysis, reasoning, data retrieval shown as a collapsible step list) and "Otto acting" (a structured action card with context, data, and a single confirm button). Getting that distinction legible at a glance was the core design challenge, and it required close iteration with research to validate that users actually felt in control — not just looked at it and assumed they were.

The "Otto Acting" pattern — a requisition card auto-generated with full justification, priority, headcount, and a one-tap confirm. The user approves the decision. Otto executes it end-to-end.

The workspace was demoed live at ServiceNow Knowledge — the company's flagship annual conference — as the platform's vision for AI-powered work. A FedEx distribution manager asks Otto one question: "How are we trending for Mother's Day?" From there, Otto runs a full surge analysis, surfaces a 37% staffing gap across the Memphis hub, auto-generates an HR requisition for 25 seasonal workers with full business justification, schedules 35 candidate interviews, and assembles a personalized Canvas — zero manual data entry, under 3 minutes.

What the audience saw wasn't a prototype. It was the gold standard for how AI-powered work should feel — the shipped interaction patterns our team built, validated through research, and pushed to production. Every card, every step list, every confirm action on that stage was a design system call we made together.

The agentic interaction model our team designed — Otto's step-transparency, the "acting vs. thinking" visual distinction, the confirm-before-execute pattern — became the gold standard for how AI agents behave across the entire ServiceNow platform. What started as a visionary demo became the design language for an AI product suite used by hundreds of enterprise customers globally.