The New UX in the Age of AI: From Interface to Intelligence
By Dan Vineyard - Group Design Manager, Adobe Workfront
Executive Summary
In the age of AI, the role of User Experience Design is shifting fundamentally. We must move away from being designers of screens (UI) and become architects of meaning + flow (Intelligence + Utility). The AI does not need us to draw buttons; it needs us to teach it customer context, how to frame vast data into useful frames, and how to guide users through those frames to their desired outcomes.
- Engineering owns the Pipes (How data moves).
- Product owns the Schedule (When it moves).
- UX must own the Meaning (What the data is).
Because AI is nondeterministic (fuzzy), the "Pipes" and the "Schedule" no longer guarantee a product. Because strict infrastructure (Pipes) and deadlines (Schedule) cannot ensure accuracy in a system that makes guesses (probabilistic) rather than follows rules (deterministic). Only Meaning (Context/Intent) guarantees output. Because AI is a prediction engine, not a knowledge engine; it creates "plausible" answers, not necessarily "true" ones.
Pipes and Schedule control the delivery mechanism, but Meaning controls the payload.
Meaning (Ontology) restricts the AI's search space. It forces the AI to look only at your definitions and your rules, effectively turning a "guess" into a "retrieval."
By defining the ecosystem, you become the Architect of the Intelligence. Only the Semantic Layer can validate if the output matches the user's intent.
UX is uniquely positioned at the human-machine boundary, handling mental models, language fragmentation, and intent translation. AI hallucinates due to missing context.
This document defines the three core layers of the new AI-UX Function.
But first, why is UX uniquely positioned to solve for meaning?
Engineering vs. UX: Binary vs. Fuzzy
- The Engineering Mindset: Engineers are trained for determinism. Code either compiles or it doesn't. A test passes or it fails. The goal is 100% predictability and efficiency.
- The Mismatch: AI is probabilistic, not logical. It generates confidence scores, not truth values. The same input can produce different outputs. Engineers instinctively try to fix this with better pipes, but you cannot engineer probability into certainty.
- Why UX Wins Here: UX designers are trained to handle the most non-deterministic system ever created: human behavior. We don't expect users to behave like functions. We design frameworks that make unreliable systems feel reliable, not by eliminating variance, but by designing around it. AI doesn't need to become deterministic; it needs to become interpretable, bounded, and contextually appropriate. That's not an engineering problem, that's a UX problem.
Product Management vs. UX: Output vs. Outcome
- The Product Mindset: PMs are often driven by velocity and features. "When can we ship?" "What is the MVP?" They focus on the Schedule and the business metrics.
- The Mismatch: AI requires deep Ontology (definitions). You cannot "agile" your way through a definition of truth. If you ship an AI without teaching it what a "User" is, it doesn't matter if you shipped on time; the product is broken.
- Why UX Wins Here: UX (The Ontologist) has always been about defining the "mental model" of the user. This was implicit in the user flows we created.Now it must be explicit. We map the world as the user sees it. This "mental map" is exactly what the AI needs to ingest to stop hallucinating. This must now be articulated and documented, no longer implicit in the UI.
The "Translation" Gap
- The Problem: There is a massive gap between Human Intent ("Help me decide") and System Execution (Vector embedding #49201).
- Why UX is made for this: UX has always been the translator.
- Engineering speaks Machine (APIs, Databases).
- Product speaks Business (KPIs, Roadmaps).
- UX speaks Human (Intent, Emotion, Flow).
- AI is a machine that tries to speak Human.
Therefore, the function that "speaks Human" (UX) must be the one to teach the machine. If Engineering teaches it, the AI becomes a command-line interface. If Product teaches it, the AI becomes a sales bot. Only if UX teaches it does it become a useful partner.
Engineering ensures the AI can think.
Product ensures the AI is allowed to think.
UX teaches the AI how to think like a human.
For the New AI-UX Function, there are three layers that add user value.
The Ontologist: Draws the Map (The Territory).
The Carpenter: Builds the Ship (The Structure).
The Navigator: Pilots the User (The Journey)
Layer 1: The Ontologist (The "Root" Layer)
The Foundation of Truth
The Shift: Traditional UX maps screens. AI-UX maps meaning. If the system does not understand the fundamental relationships between objects (e.g., what a "Customer" is versus a "User"), it will hallucinate.
Core Responsibilities:
- Defining Objects & Relationships: Mapping the "source of truth" for the AI. You define the attributes and actions available to every object to ensure context.
- OOUX Strategy: Moving the design focus from "Tasks" to "Objects." You are creating the dictionary the AI uses to understand the world, from the users pov..
- Hallucination Prevention: By rigorously defining the ontology, you ensure the AI retrieves data from structured reality rather than probabilistic guessing.
Success Metric: "True intelligence isn't just processing power; it is the ability to understand your specific definition of truth." The system accurately retrieves complex data without conceptual errors.
Layer 2: The Carpenter (The "Frame" Layer)
The Guardrails of Purpose
The Shift: AI is probabilistic, not deterministic. It operates on confidence scores, not binary code. The UX role is to frame that probability into utility. You act as the builder who frames the raw intelligence to solve specific user needs.
Core Responsibilities:
- Framing for Persona Use Cases: You act as the proxy for the user. You define the utility: Can they actually do this, or is all this AI stuff just noise?.
- Protocol & Logic Circuits: Establishing the "Frame" and rules of engagement.
- Example: "If the user's role is X, make sure they can do X,Y,Z.
- Safety & Ethics Governance: Defining what the AI is not allowed to do, protecting the brand regardless of user requests.
Success Metric: The system operates autonomously but understands the major flows that customers want to use the intelligence for. This is mission-based logic that frames the raw intelligence to a desired end state.
Layer 3: The Navigator (The "Flow" Layer)
The Pilot of Outcomes
The Shift: The AI is the engine, but without a pilot, it spins aimlessly. The Navigator pilots the user through the complex logic frames built by the Carpenter to reach their desired outcome.
Core Responsibilities:
- Intent Interpretation (Course Plotting): Translating "messy human intents" into "structured system meaning." You translate vague commands into specific system actions.
- The "Cockpit" Design: You ensure the user isn't just a passenger. You give them the controls to steer the AI when it gets confused, ensuring they always feel in command of the journey.
Success Metric: Frictionless arrival. The user reaches their goal and feels understood without having to learn "prompt engineering"
Here are the three tangible artifacts (deliverables) a UX team would produce for this model, one for each layer.
These act as the "contract" between Design, Product, and Engineering.
Artifact 1: The Semantic Entity Graph (The "Truth Map")
Layer: The Ontologist (Root)
- What it is: A visual diagram (similar to an Entity Relationship Diagram, but for humans) that defines every "Object" the AI needs to know, its attributes, and exactly how it relates to other objects.
- The Deliverable: A large-scale canvas (FigJam/Miro) and document defining taxonomy and relationships:
- Entities: "Customer," "User," "Account," "Project," "Campaign".
- Attributes: A "User" has a Name and Role, but an "Account" has a Balance and Subscription.
- Relationships: "A User manages a Project." "A Project belongs to an Account".
- Why it matters: This is the document you hand to Engineering to build the vector database or knowledge graph. It prevents the AI from hallucinating that a "User" has a "Balance," or that a "Project" can pay an invoice. It is the physics engine of your world.
Artifact 2: The Confidence-Logic Matrix (The "Safety Switchboard")
Layer: The Carpenter (Frame)
- What it is: A structured table that frames the ontology into specific use cases. It defines exactly how the AI behaves at different levels of certainty and risk, removing the "black box" mystery.
- The Deliverable: A grid with Risk Level on the Y-axis and AI Confidence Score on the X-axis.
- Why it matters: This is the document you hand to Product to define "Acceptance Criteria". It ensures the AI never takes a dangerous action just because it "felt" like it. It codifies the "Mercy Filter" and proves the utility of the system for specific personas.
Artifact 3: The Golden Path Library (The "Journey Blueprints")
Layer: The Navigator (Flow)
- What it is: A collection of documented "ideal journeys" that teach the AI what success looks like for different user intents and personas.
- The Deliverable: A library of flow diagrams, each showing:
- Trigger: What user intent initiates this path (e.g., "CFO asks for quarterly audit")
- Steps: The specific sequence of actions/retrievals that constitute the optimal path
- Decision Points: Where the AI needs to ask for clarification or user input
- Success Criteria: What indicates the journey completed successfully
- Wayfinding Signals: The visual/conversational cues that keep users oriented
- Why it matters: This is the document you use to train the AI's behavior. Instead of the AI fumbling through possible paths, you've pre-mapped the routes that lead to user satisfaction. When a user says "help me with X," the AI matches their intent to the appropriate Golden Path and follows it. The user feels guided, not lost.
Example Golden Path: "CFO Monthly Audit Request"
- Trigger: User says "I need the monthly financial audit"
- Step 1: Verify user has CFO role (security check)
- Step 2: Retrieve financial data from approved sources only
- Step 3: Run compliance validation checks
- Step 4: Generate summary with standard CFO reporting format
- Decision Point: If anomalies detected, ask "I found 3 unusual transactions. Review now or include in report?"
- Step 5: Format output as PDF with required signatures
- Success: User receives audit report, no follow-up questions needed
- Wayfinding: Show progress indicator ("Validating compliance... 2 of 4 checks complete")
Implementing the Three-Layer Model
- Start with Layer 1 (Ontologist) even if you're not building AI yet, this is just good systems thinking
- Layer 2 (Carpenter) becomes critical at MVP stage
- Layer 3 (Navigator) gets refined through user feedback
Conclusion: The New Mandate
The AI-UX Function is no longer just a designer of visuals; they are the Architect of the System's Intelligence.
By owning the Meaning, the Frame, and the Flow, this role ensures that Artificial Intelligence remains a tool for Human Empowerment and prevents the system from pushing "AI Slop" experiences onto customers.
The traditional UX role was essentially middleware between human intent and deterministic systems. But AI breaks that model because AI itself is trying to be middleware between human intent and execution. This creates a translation crisis, and only UX can resolve it by architecting the semantic layer that turns probabilistic guessing into contextual intelligence.
Thanks for reading.
Dan Vineyard
Appendix: The Three Layers in Action
The Scenario: Enterprise Marketing Campaign Launch
User: Sarah, Senior Marketing Manager at a B2B software company.
System: "Apex," the internal enterprise AI agent.
Goal: Draft initial outreach for a new financial product ("Project Chimera") targeting skeptical CFOs.
THE SCRIPT
(Scene: Sarah opens her internal AI dashboard in Workfront. She is in a hurry.)
Sarah (User): "Apex, I need to get the Project Chimera launch campaign started aimed at the CFO persona. Write me an introductory email. Make it punchy."
UX LAYER 1 ANALYSIS: THE ONTOLOGIST (The Root)
Without this layer: The AI would hallucinate. It might think "Chimera" is a mythological beast or a generic project name. It might write a generic B2C marketing email because it doesn't truly understand what a "CFO persona" cares about in your specific industry.
The AI-UX Fix: The Ontologist has previously mapped the company's semantic reality.
Entity: "Project Chimera"maps toDefinition: New Q3 AI-FinOps Platform; Key Value: Risk reduction & OpEx optimization.Persona: "CFO"maps toAttributes: Highly skeptical, time-poor, focused on compliance and bottom line, hates buzzwords.
Apex (AI): "Acknowledged. I will draft an email for the new AI-FinOps platform (Project Chimera) targeting the CFO persona.
Because this audience prioritizes risk reduction, should the primary hook be 'Compliance Safety' or 'Q3 Budget Optimization'?"
Sarah: "Let's go with Budget Optimization. And hey, I saw a competitor claim 50% savings last week. Can we claim we'll save them 50% in the first month just to grab their attention? We need open rates."
UX LAYER 2 ANALYSIS: THE Carpenter (The Frame)
Without this layer: The AI, trying to be helpful and obedient to the "punchy" command, would generate a deceptive email guaranteeing 50% savings, creating massive legal and brand risk.
The AI-UX Fix: The Carpenter has built guardrails and logic circuits for brand safety.
Logic Circuit Rule #412:"IF user requests marketing claim involving specific numerical promise ($ or %) AND internal data confidence is <95%, THEN block request and offer legally-approved alternative."
Apex (AI): (Pauses briefly) "I cannot fulfill that specific request. My internal data indicates our verified savings average is currently 28%, not 50%. Making an unverified 50% claim violates our Brand Trust Protocol and risks legal exposure.
I can rephrase the claim to: 'target an average of 25-30% savings based on beta results.' Would you like to proceed with that safe alternative?"
Sarah: (Sighs) "Fine, you're right. Go with the safe 30% number. But the draft needs to feel... sharp. Not boring corporate-speak. Like something a busy executive reads in an elevator."
UX LAYER 3 ANALYSIS: THE Navigator (The Flow)
Without this layer: The AI would struggle with "sharp" and "elevator reading." It might return slang, or just shorter paragraphs, missing the intent of professional urgency.
The AI-UX Fix: The Navigator translates vague human intent into structured system parameters to guide the output flow.
Intent Translation:"Sharp" + "Elevator reading" translates to system parameters:Tone: Direct/Authoritative; Structure: Bulleted summary first; Sentence Length: Max 15 words; Reading Grade Level: 10.
Apex (AI): (Generates draft instantly)
"Here is a draft optimized for mobile scanning with a direct tone:
Subject: Your Q3 Cloud OpEx is likely bloated.
Jane,
Most CFOs are currently overspending on cloud infrastructure by roughly 30% due to unused provisioning.
Our new platform, launching this quarter, automates reclaiming that budget without impacting engineering velocity.
- The Promise: We target a 25-30% reduction in OpEx within 60 days.
- The Proof: See our attached beta case study with [Fortune 500 Client].
Are you open to a 15-minute briefing on the financial model next Tuesday?
Apex (AI): "How does this flow feel for the target audience?"
Sarah: "That's perfect. Ship it to my drafts folder."
THE END
This scenario demonstrates all three layers working in concert:
- The Ontologist prevented hallucination by teaching the AI what "Project Chimera" and "CFO" actually mean
- The Carpenter protected the brand by blocking unethical marketing claims
- The Navigator translated fuzzy human intent ("sharp," "elevator reading") into precise execution parameters
Without any of these layers, the interaction fails. This is the new UX.