Staff Designer, Generative AI Lead · Design systems & training · ServiceNow · 2024–2026 · Adopted Org-Wide

I turned conversation design into an
AI behavior discipline

Helping designers specify AI behavior, not just write better copy

At ServiceNow, I helped a design organization of nearly a thousand people stop treating conversation design as copywriting and start treating it as AI behavior design.

From

Write a script for one exact exchange.

To

Write a system prompt the model can follow.

What I Did

I made conversation design teachable as a systems discipline.

I originated the framing that conversation design is not just a writing task. I wrote the principles that set the quality bar, got them published as an org-wide standard, and co-led the training that carried them across the design organization.

Then I designed the handoff model that helped designers specify behavior instead of scripting one ideal exchange.

Designers do not need to write every conversation. They need to design the system that produces good conversations.

From writing to specifying

ProblemA large design org was asked to design AI with artifacts built for scripted systems.
SignalDesigners handed off ideal flows. Engineers received examples, not behavior specifications.
ShiftI reframed conversation design as AI behavior design and built the principles, training and handoff to teach it.
Outcome600+ designers trained; standards required across 8+ business units; 25+ downstream artifacts trace back to the work.
What I owned
I owned
  • Reframing conversation design as AI behavior design.
  • Defining the five layers where conversation quality actually lives.
  • Writing the conversation design principles for ServiceNow.
  • Publishing the principles in Horizon, ServiceNow’s design system.
  • Leading the training section on how and when to use conversation as an interaction model.
  • Identifying the handoff gap between design and engineering.
  • Designing the Conversation Design Brief as a behavioral specification.
  • Making the discipline usable by designers who did not see themselves as conversation designers.
We partnered on
  • Internal discovery across designers and engineers.
  • Organization-wide training design and delivery.
  • Content design instruction and examples.
  • Adoption through design-system channels.
  • Workshop adaptation for ServiceNow Knowledge 2026.
  • Downstream implementation by product, design, and engineering teams.
01 / The Problem

The org was asked to design AI with no shared discipline

That created a predictable failure.

Designers thought they were handing off the experience.

Engineers were receiving one example of the experience.

The model then produced something different, because nondeterministic systems do not reliably follow scripts.

I worked with a junior colleague to map what was happening across teams. They observed sessions designers were already running for their own product areas, interviewed engineers, and traced where the handoff broke down.

Designers
Created one ideal flow and expected the shipped experience to match it
Were frustrated the product looked nothing like the design
Followed training best practices, but still under-delivered
Had no clear way to hand work to engineering
Engineers
Received scripts that did not say how the AI should behave across real variation
Had specific responses, but no AI behaviors
Were hard-coding conversations, which defeated the point of AI
Filled the gaps with unvalidated assumptions

The missing artifact was not better copy. It was a discipline.

What designers and engineers said

ServiceNow’s design teams were being asked to create conversational AI experiences, but most of the available design artifacts came from a pre-LLM world: happy-path flows, scripted dialogue, and polished example responses.

But the organization had no shared definition of good conversational AI, no common language for AI behavior, and no operating model for turning design intent into model behavior.

The pattern was consistent, and both sides were frustrated by it.

I wrote a script and gave it to my engineering team. Why doesn’t it look like what I designed?Designer
I can’t implement this. The AI will never produce those exact responses.Engineer

The handoff was asking scripts to do the job of specifications.

The issue was not collaboration.

It was that the deliverable was wrong for the medium.

02 / The Reframe

I reframed conversation design as five layers of AI behavior

AI conversation quality does not live only in the words. Tone matters. Clarity matters. Good writing matters.

Most of what people called a writing problem was a decision problem three layers down.

01
Surface response

People assumed: just the words

Actually: tone, clarity, and phrasing. The visible layer, but rarely enough to fix a bad answer.

02
Interaction pattern

People assumed: a flow diagram

Actually: what shape the exchange takes: ask, confirm, act, show.

03
Behavioral rules

People assumed: engineering’s call

Actually: what the system infers, decides, or defers to the user.

04
Context and data use

People assumed: a privacy review

Actually: what the system knows, what it can use, and what trust consequences follow.

05
Guardrails and evaluation

People assumed: QA at the end

Actually: what the system must never do and how that gets verified.

Conversation design lives at every layer of this stack, not just the response.

Why it mattered for enterprise

Naming the layers is what let a designer point at a problem and say where it lived.

I spent a long stretch working out where conversational AI was going once LLMs took over the response layer, and I concluded that the discipline had to change with it.

But in an AI system, the response is the surface expression of deeper behavioral decisions: what the system inferred, what context it used, what it decided to ask, what it chose not to do, and how it handled uncertainty.

At ServiceNow, conversational AI was not just answering casual questions.

It was helping people complete enterprise work: triaging issues, fulfilling requests, troubleshooting problems, summarizing information, and taking action inside business systems.

That raised the stakes.

A consumer assistant can often recover from a vague or overly broad answer. Enterprise AI has less room for that. Users need the system to understand role, context, permissions, task state, and risk.

They need to know when it is acting on their behalf, when it is asking for confirmation, and why a recommendation is safe to trust.

That is why I treated conversation as a behavioral system, not a tone layer.

03 / The Foundation

I gave the organization a shared quality bar

I wrote a set of conversation design principles for ServiceNow and published them in Horizon, the company’s design system. Publishing them there is what turned a point of view into a standard someone could be held to.

The conversation design principles published in Horizon, ServiceNow's design system
The conversation design principles as published in Horizon. An org-wide baseline, not just a doc.
View the principles in detail

The goal was not generic writing guidance. It was to state the behavioral expectations for enterprise AI in the place teams already went to look things up.

Know who you’re designing for

Design for a specific person in a specific moment.

Reduce every ounce of effort

Use context, automation, and inference to reduce the user’s burden.

Design for human agency

Give users real control, not the illusion of it.

Sound like a partner, not a product

No corporate voice. No sycophancy. No fake personality.

Design for uncertainty

Handle uncertainty through clarification, caveats, escalation, or recovery.

Design for trust

Trust comes from useful evidence, not confidence theater.

The principles were not meant to make AI sound nicer. They were meant to help teams make better decisions.

Know who you’re designing for meant the AI should account for role, permissions, task, context, and what the system already knows.

Reduce every ounce of effort meant the assistant should remove work, not add it.

Design for human agency meant users needed real control over action, confirmation, and escalation.

Sound like a partner, not a product meant competence, clarity, and collaboration mattered more than brand voice.

Design for uncertainty meant the system needed a plan for ambiguity, missing information, and recovery.

Design for trust meant the system had to show useful evidence: what it did, why it matters, and what the user can do next.

04 / Training

I taught designers to reason about AI behavior

Once the principles existed, I collaborated with 2 content design colleagues to turn them into an organization-wide training program.

The most important shift I taught was getting designers to ask better product questions:

Should this be conversational?
What does conversation make easier?
When should the AI infer, ask, act, confirm, or escalate?
What needs to be visual instead of verbal?

Training slide on agentic AI differences Training slide on Grice’s maxims: be truthful, informative, relevant, and clear

84% of attendees said they felt more equipped to approach conversation design one month after the session.

The material later became a hands-on workshop at ServiceNow Knowledge 2026, where I taught customers and practitioners how to identify and fix common AI conversation failures.

Alea Abrams presenting the conversation design workshop at ServiceNow Knowledge 2026
She spent time sharing her knowledge and helping me upskill, which made a huge difference in my understanding of the space.Content Design Coworker
Training structure and reach

I partnered with two content design colleagues to design and deliver three sessions across the design org. Together, we taught the fundamentals of conversation design, how to write for ServiceNow’s product context, and, in the section I led, how to decide when conversation should be used as an interaction model at all.

Part 01
Conversation fundamentals

What conversation design is and why it differs from general UX writing.

Part 02
Writing for ServiceNow

How to write clear, useful, enterprise-appropriate conversations for real product use cases.

Part 03 I led this
How and when to use conversation

Conversation as a unique mode of interaction, not a replacement for UI.

That timing mattered. A day-of number measures the session. A month-later number measures whether anything stuck.

The core message across all three parts was consistent:

Conversation should not be used because AI is available.

Conversation should be used when it creates a better path for the user.

Each session was recorded and the material was packaged for teams to run themselves. That is how 600+ people in rooms became close to 2,000 across design, product, and engineering.

People kept pulling it in without me being there.

05 / The Handoff

I turned shared language into a handoff model

Training created shared language.

The next step was making that language usable in delivery.

So the discipline needed an artifact, not just a vocabulary.

The Conversation Design Brief translated the discipline into a structured handoff:

User goal Trigger prompts Data boundaries Interaction shape Visual response patterns Edge cases Escalation paths Guardrails

Designers specify the behavior instead of scripting one example of it.

I’m not a writer, but I can do this.Designer · after using the brief

A designer who had ruled themselves out of conversation work found a way in because the job had been redefined as making decisions rather than producing prose.

Where the full Brief lives

The existing handoff model asked designers to produce polished examples, which left engineers to infer the behavior underneath them and fill the gaps with assumptions nobody had reviewed.

That is what an operating model is for. It widens who can practice the discipline.

The brief is what a discipline looks like once it is established enough to need tooling.

It is the point where principles stop being guidance and start being an input to delivery.

The brief mattered because it closed the handoff gap between design intent and model behavior.

The full brief, the conversations generated from it, and the engineering handoff belong in the evaluation case.

The full brief, generated conversations, and engineering handoff live in the evaluation case. Here, the point is simpler: the discipline became usable when it changed the handoff.

06 / What Changed

Designers stopped handing off scripts and started specifying behavior

The strongest signal of impact is that the work stopped being mine.

The standards became part of how the organization defines conversational AI quality, and they are now used by teams I have never worked with, on products I don’t need to review.

600+designers trained live.
8+business units using the standards as a required baseline input.
25+downstream artifacts tracing back to these principles and frameworks.

The work changed how conversation design was understood, practiced, and handed off.

Before After
Conversation design meant writing the words.Conversation design meant specifying system behavior.
Designers handed off ideal scripts.Designers defined decisions, constraints, and guardrails.
Engineers inferred behavior from examples.Engineers received a structured behavioral handoff.
Teams fixed AI responses at the surface.Teams could identify the deeper layer where the failure lived.
Conversation was treated as a replacement for UI.Conversation was treated as an interaction model with specific affordances.
The discipline depended on a few specialists.More designers had a way to practice it.
Closing

A discipline changes what people know how to notice

Building a discipline is not only publishing principles or running training. It is giving people language for work they could previously only feel.

Before this work, teams could tell when a conversation felt wrong, but not always which decision was missing. Afterward, they had language for the layers underneath the response: the pattern, the rule, the context, the guardrail, the handoff.

That is what I think a discipline does. It gives people a way to see the work clearly enough to practice it, critique it, and carry it forward without you in the room.