AI & Agentic Experience Design

We help design how AI works—inside your team and inside your product.

From improving how teams use AI tools to designing AI-native product experiences, we help SaaS and enterprise companies turn AI into something usable, controlled, and built for the real world.

Experience
12+ years product design
Level
Senior / Director
Engagement
Remote collaboration
The Trust-by-Design framework

Why most AI products fail at the interface layer.

The bottleneck in AI adoption isn't model performance — it's interface design. Teams invest heavily in training sophisticated models, then wrap them in UX that either hides how they work ("just trust us") or overwhelms users with technical explanations they don't need.

The result is predictable: users abandon AI features. Not because the technology fails, but because the interface never taught them what to expect, how to course-correct, or when to intervene. They return to manual workflows because those feel more controllable, even if they're slower.

Good AI product design solves this by building accurate mental models — helping users develop an intuitive understanding of what the AI can and can't do, without requiring them to understand transformers or training data.

The four layers of AI trust.

Trust in AI systems isn't binary. It's built across four distinct layers, each requiring different design interventions. Our framework systematically addresses each layer through specific design patterns, ensuring trust compounds rather than fragments.

Predictability

Can users anticipate what the AI will do? Clear interaction patterns (copilot vs. agent vs. classifier) set expectations before outputs appear.

Transparency

Can users see why the AI did what it did? Not full explainability — contextual signals at decision points.

Control

Can users redirect, refine, or reject AI outputs? Agency keeps humans in the loop where judgment matters.

Accountability

When things go wrong, is there a clear path to correction? Error states, feedback loops, and escalation paths complete the trust loop.

Interaction patterns that match the model.

Not all AI features need the same UX. The interaction model should match the task and the model's capabilities. We help teams choose the right pattern, then design the full interaction model — including the edge cases, errors, and feedback loops most teams skip in v1.

Copilot patterns

For augmentation — AI suggests, human decides. Best for creative or judgment-heavy work where users want to stay in control.

Agent patterns

For delegation — users set goals, AI executes autonomously. Requires clear boundaries, progress visibility, and interrupt affordances.

Classifier patterns

For triage — AI sorts, routes, and flags while humans handle exceptions. Optimizes for speed while maintaining oversight.

Advisor patterns

For decision support — AI surfaces insights, human synthesizes. Information-dense dashboards with confidence indicators.

Mental model mapping: the missing step.

Most AI projects jump straight from "what should the model do" to "what should the screen look like." They skip the critical middle step: what should users believe the model does?

Mental model mapping documents the gap between how the AI actually works and how users think it works. This isn't about dumbing things down — it's about designing for the model users will inevitably build, whether you plan for it or not.

A well-designed mental model means users know when to trust the AI, when to double-check it, and when to override it — without needing to read documentation or understand the architecture.

Service offerings

How we can work together.

01

AI/AX Strategy & Assessment

Comprehensive audit of your AI product or feature set. Evaluate trust layers, mental model gaps, and interaction pattern fit. Leave with a prioritized roadmap grounded in user research and UX heuristics.

Timeline4–6 weeks
FormatRemote (async + live sessions)

Deliverables

  • —Trust-by-Design audit across all four layers
  • —Mental model gap analysis (user belief vs. system behavior)
  • —Interaction pattern recommendations (copilot/agent/classifier fit)
  • —Prioritized improvement roadmap with effort/impact scoring
  • —Stakeholder workshop + recommendations deck
02

AI Product Design Sprint

Rapid, collaborative sprint to design a new AI feature from scratch. Define interaction model, design key flows with edge cases, and prototype before engineering commits to an architecture.

Timeline6–8 weeks
FormatRemote (embedded with team)

Deliverables

  • —AI interaction pattern definition + mental model specification
  • —Core user flows including error states and feedback loops
  • —High-fidelity clickable prototype (ready for user testing)
  • —Design system components for AI-specific patterns
  • —Handoff documentation with rationale for key decisions
03

Fractional AI Design Leadership

Embedded design leadership for a major AI initiative (3–6 months). Own the full UX strategy from product definition through launch, working directly with product and engineering.

Timeline3–6 months
FormatRemote (fractional Director role)

Deliverables

  • —End-to-end product strategy and interaction model definition
  • —Complete design system with AI-specific component library
  • —All screens, flows, edge cases, and error states designed
  • —Iterative user testing and refinement (3+ rounds)
  • —Ongoing engineering collaboration through launch
  • —Post-launch metrics framework and iteration plan
04

Responsible AI Workshop

Structured workshop series to help your team navigate ethical, legal, and trust challenges. Design patterns for transparency, consent, and user control grounded in responsible AI principles.

Timeline4 weeks
FormatRemote (live workshops + async work)

Deliverables

  • —Responsible AI design framework tailored to your context
  • —Transparency and explainability pattern library
  • —User control and consent flow designs
  • —Risk assessment matrix for AI features
  • —Team training session on ethical AI design
Recent AI/AX work

Case examples.

Euphonic AI · Startup · SaaS 2024

Brand and interface design for an AI-led product

Partnered with the Euphonic AI founding team to define the brand identity and design the end-to-end interface for their AI-led SaaS platform. Established a visual system that signaled trust and intelligence without leaning on tired AI tropes, then translated it into a product UI where the model's reasoning, confidence, and limits are surfaced naturally inside the workflow. The result was a launch-ready product with a coherent voice across marketing site, onboarding, and core experience — and a design system the team could keep extending as the product matured.

ZS · Agentic AI · Strategy 2026

Agentic AI experience strategy for an enterprise consultancy

Led the experience strategy for ZS's agentic AI initiative, helping product and consulting leadership define how autonomous agents should fit into existing client workflows. Mapped the spectrum of human-in-the-loop, human-on-the-loop, and fully delegated patterns against real engagement scenarios, then specified the trust signals, escalation paths, and interruption affordances each pattern required. Delivered a strategy framework, prioritized opportunity map, and reference interaction models that gave the org a shared language for designing agentic features without compromising client trust or compliance posture.

ZS · Agentic AI · Design System Evaluation 2026

Design system evaluation for agentic AI surfaces

Audited ZS's existing design system against the demands of agentic AI interfaces — streaming outputs, in-progress reasoning, tool use, confidence and provenance, and recoverable error states. Identified gaps where current components quietly broke down under agent behavior, then proposed a layered set of new primitives and patterns (agent status, intent capture, plan preview, action approvals, audit trails) that extend the system rather than fork it. The output gave their design and engineering teams a clear roadmap for evolving the system without losing the consistency they'd already invested in.

Let's build AI that earns trust.

If you're shipping AI products and want interfaces that feel transparent, controllable, and genuinely helpful — not just technically impressive — let's talk.