AI-LX · V2B hero · NOTE: business stakes paragraph and one participant moment still needed Read the editorial memo →
A major automotive retailer · 2025 Co-facilitator & lead researcher · Four Dimensions framework author

From "we should do AI" to a working vocabulary for AI in services

A two-day workshop that gave 50 designers a framework for treating AI as a design material — and the vocabulary to push back when someone walks in with a solution-shaped brief.

50
designers, researchers, executives
4
dimensions framework authored
Re-engaged
client follow-on commissioned

A 50-person design team had grown from ten people in 2018 to more than fifty by 2025. AI was the new ambient pressure on every product team: the most enthusiastic designers were leaving meetings to vibe-code prototypes in Figma; the most skeptical wanted to stop hearing the word entirely. Both reactions are symptoms of the same problem — AI being treated as either a destination or a contaminant, and almost never as a design material with properties, constraints, and an appropriate use in a larger service. The workshop's job was to give the team a vocabulary for making that distinction — and a practice they could carry back into their own product teams.

Chapter 01

A team under pressure to do AI — and no shared language for what that meant

The client's design leadership gave Harmonic a specific brief: help the team get more comfortable with AI as a design material. Help them build complete solutions that aren't AI-for-AI's-sake. Make it something they can carry into their own product teams — not a one-off lecture, a usable practice. Automotive retail is a high-stakes, trust-sensitive industry: high-ticket purchases, low-frequency transactions, and a customer relationship that lives or dies on whether the experience earns confidence at every step. An AI that misfires in a car-buying journey doesn't just fail to convert — it damages the relationship that makes the next visit possible.

I led the research and knowledge-base development ahead of the two days: primary conversations tapping my network, secondary sources pulled into a Harmonic AI POV document covering human-to-AI interaction patterns, front-stage and back-stage use case typologies, and emerging service design methods being adapted for AI contexts. During the two days, I introduced the AI design principles and the Four Dimensions of AI in Services framework — original work I authored because no existing taxonomy quite did what the team needed.

A workshop table viewed from above: hands sketching on a printed concept canvas in pen, several other canvases nearby with handwritten notes, and sticky notes clustered at the edge of the table.
Mid-activity at one of the workshop tables. Concept canvases were filled in by hand and small group, then walked back through together.

Chapter 02

The organizing move: put AI in the corner of the canvas

The instinct in the room, and in most rooms right now, is to start from "we should do AI" and look for somewhere to put it. That order makes the technology the design problem. Everything that matters about a service — the brand promise, the value exchange, the moments where it meets a customer, the feedback loops that tell you whether any of it is working — gets rearranged around the AI rather than the other way around.

The workshop's organizing move was to flip that order. Borrow from Henry Dreyfuss writing about industrial design in the 1950s: fit the new materials to people, not people to the materials. Treat AI the way an industrial designer treats wood or steel — a material with properties, constraints, and an appropriate use. Not a destination. Patrick introduced a value-centered canvas that made this concrete: value in the center, moments that matter at the top, brand tensions on one side, feedback loops below — and AI deliberately in the corner. That placement is the pedagogical move. AI is on the canvas because it belongs there, but it's in service of value, not the other way around.

The Intelligent Experience Canvas: Customer and Service as two large circles connected by Value Exchange arrows; Moments That Matter and Brand at the top; Feedback Loops at the bottom-left; The Role of AI deliberately placed in the bottom-right corner, not the center.
The value-centered canvas. Value in the middle. AI in the corner.

Chapter 03

The Four Dimensions framework — a vocabulary for choices the team was already making implicitly

Most AI design vocabulary is interface-level: chatbot heuristics, prompt patterns. The team needed something a service designer could use to reason about an AI's role across an omnichannel journey. I built the Four Dimensions of AI in Services for this workshop.

Actors: on the human side — customer, front-stage associate, back-stage associate, supplier. On the AI side — the customer's personal AI agent, the company's customer-facing AI, the company's employee-facing AI. Naming actors matters because the team had been collapsing all of these into "the AI." Interface: explicit (chatbot — the user knows), ambient (the lot reads a license plate), or mixed. Most teams default to explicit; ambient is where service design has the most to contribute. Collaboration model: sequential, parallel, independent, or orchestrated. Naming the other three opens design space the team was systematically missing. Control model: manual, assisted, monitored, automated, or autonomous. Control is a knob, not a binary. Every AI choice in a service is a position on this scale, with consequences for trust, liability, and brand.

The framework also underpinned an adapted service blueprint. Standard blueprints divide front stage and back stage. We added a further division: within each stage, separate human actors from AI actors. Once AI is named as an actor, the team has to say what it does, where its authority ends, and what the handoff to a human looks like.

Adapted service blueprint for a car-buying journey, with Frontstage AI and Backstage AI/ML rows added alongside their human counterparts — showing where AI acts, where its authority ends, and where human handoffs occur.
The adapted service blueprint — AI named as an actor in its own right, not a feature embedded in someone else's row. The Four Dimensions diagram (stand-alone) is being extracted from workshop materials.

Chapter 04

Paper, no laptops — why we slowed down to go faster

The workshop was deliberately low-fidelity: 50 people in a room, working on paper, sketching by hand. Some participants left and came back having built something in Figma. We let it happen, but the intent was clear — the two days were about practicing the thinking, not generating outputs. Laptops accelerate solutioning past the point where the framework can do its work.

The deeper bet: a 50-person team that has practiced putting AI in the corner of a canvas, naming it as an actor in a blueprint, and locating it on a control spectrum will recognize those moves later — in their real product teams, when someone walks in with a solution-shaped brief. The workshop isn't the deliverable. The recognition that happens six months later is.

Don't just start vibe coding when you don't have the foundational thinking done. Go all the way back to what you needed to do. Go all the way back to sketching.

— Whitney Masulis, during the workshop

The reframe

The team went in asking "where should AI go?"

The brief was to help the team get comfortable with AI as a design material. What the workshop actually had to do was harder: interrupt the instinct to start from "where should we put AI?" and replace it with a different starting question — "what value are we creating, and is AI the right material for this moment?" That second question is genuinely difficult to hold in a room where the loudest voices are already in solution mode. The framework is the tool for holding it. The two days of practice are what builds the muscle to reach for that tool under pressure.

What stays behind

A framework and practice that travel back into product teams

Three artifacts came out of the two days that live beyond them: the Four Dimensions framework, the value-centered canvas, and the adapted service blueprint with human and AI actor rows. The Harmonic AI POV document — built for this workshop — continues to be updated as the field moves. None of these were deliverables in the usual sense; they're tools for a practice that travels back into 50 product teams working on 50 different briefs.

The outcome signal: Harmonic was asked back for more work. Whether the team is using the canvas, the dimensions, and the adapted blueprint inside their actual product work is something we don't yet have ground truth on — and the case says so honestly. What the workshop produced, materially, is a team that has practiced asking "what value are we creating?" before "what AI should we use?" That practice is the real artifact.