AI design · SNAPSHOT
AI
AI agents are thinking collaborators — and a new type of user. I design the workflows they run, the systems they consume, and the gates that stay human.
Agentic AI
Agents can take on work autonomously — reasoning, calling tools, multi-tasking, making decisions. My job is designing what that autonomy looks like, and where it stops.
Autonomous Job Application Agent with Claude Cowork
Jun 2026
Claude Cowork × Airtable × Notion · autonomous multi-agent pipeline
Architecture: robust AI design with a structured data pipeline, query mechanism, and full agent-loop workflows to autonomously apply for jobs directly from your local machine
Execution: autonomous multi-tasking and tool calling, generating full reasoning and decision audit logs at the end of each session
Agentic RAG / JIT: optimises the agent’s cognitive load
Decision Engine: operates on a “Minimum Viable Context” rubric — the core 3-way decision engine (Apply / Park / Skip)
Safety & Control: balances autonomy with HITL escalation gates
AI Architect
Data infrastructure
Query mechanism
Claude Cowork
HITL gates
Decision audit logs
RAG
JIT context
Amy’s Intelligent Agent with Claude Fable 5
Live
Live, fully functional AI agent, you can talk to her on this website
Deployed and governed with confidence, talk to visitors unsupervised
Configured with custom guardrails, rules, and safety gates
Built with Claude Code on Fable 5 in under 2 minutes
AI Architect
AI interface
Claude Code
Claude AI
Design System Component Automation
Mar 2026
Claude Code - Agentic Orchestration - 6 autonomous agents with escalation gates
Agentic Framework: architected to automate component generation
Multi-Agent Coordination: orchestrated token wiring, variants, and property generation
Human-in-the-Loop: experts utilised for escalation and governance
Claude Code
Agentic Framework
Agent Orchestration
AI Governance
HITL gates
Agent Experience (AX) — agents as users
AI agents are a new type of user — with finite context, real constraints, and their own experience of our systems. I've been designing for them as users since early 2026, before the industry had a name for it.
Bilateral pipeline with Claude Code × Figma MCP × Codebase
Mar 2026
Claude Code - Git - Figma MCP - Ratehub
Production codebase as the single source of truth
Automated token extraction and component generation
Bidirectional sync — code ↔ Figma
Claude Code
Figma MCP
Token Pipeline
Bilateral Sync
Design System Component Automation
Mar 2026
Claude Code - Agentic Orchestration - 6 autonomous agents with escalation gates
Agentic Framework: architected to automate component generation
Multi-Agent Coordination: orchestrated token wiring, variants, and property generation
Human-in-the-Loop: experts utilised for escalation and governance
Claude Code
Agentic Framework
Agent Orchestration
AI Governance
HITL gates
Token Management Sandbox
LIVE EXPLORATION
Codebase - Design system - Build live products - designing how agents consume, validate, and sync design tokens.
AI agents' needs: agents don't browse a token library the way developers do; they need consumption-ready structure within finite context
Just-in-time: Balance progressive disclosure with pre-load the whole systems
Governance layer and human-in the loop: When agents become users, the systems they consume have to be governed for them. → Systems Design
Claude Code
Agentic Framework
Agent Orchestration
AI Governance
HITL gates
AI-Human Collaboration
AI agents are a new type of user — with finite context, real constraints, and their own experience of our systems. I've been designing for them as users since early 2026, before the industry had a name for it.
Human-AI Design Team Collaboration Framework
Feb 2026
Full workflows for human–AI design teams across real conditions — new product, existing product, flawed or flawless design system
Automation Quality Matrix: mapped drift taxation and automation ceilings across 5 scenarios, including 0→1 builds
Drift detection, automation ceiling, uncertainty estimation, HITL gates.
Two kinds of drift: conformance (output deviates from spec) and ground-truth (the spec itself is wrong). A gate is only real when the human injects what the system can't supply — intent, tolerance, or external ground truth.
AI Governance
Quality Matrix
Drift Analysis
HITL Design
Expertise Mapping
Claude Code agents team
Design Automation Framework & Agentic Governance Framework
Feb 2026
Double Diamond Automation: integrated automation into the execution and delivery phases
System Maturity Analysis: mapped workflow breakdowns against design system maturity
Governance layer: quality matrix, expertise mapping Jr→Staff, escalation ceilings
Agent Workflow Design
Double Diamond
System Maturity
Claude Code agents team
Lean and full brief checklists
Feb 2026
Scalable Frameworks: tailored design briefs for lean vs. heavy project scenarios, scaled to a solo designer or a full team.
Agent Workflow Design
Double Diamond
System Maturity
Design Briefs
Claude Code agents team
AI-Powered Products
Not prototypes. Shipped, rated, in market — AI at the core of products real customers use.Same discipline as the Data Intelligence layer below — the mechanism shipped at consumer scale. NIQ runs it at 60B data points.
Xboost – 12 Shopify App Ecosystem (5-Star Rated)
Sept 2024
★★★★★
Architected the overarching design system, navigation and UX framework, guiding a junior designer in execution
AI-Driven Upsells: leveraged user browsing history to power personalized cart recommendations
Automated Workflows: designed review automation with AI moderation and customized fatigue alerts
Ratehub: AI-Powered Card Finder
Dec 2025
Designed trust signals — model confidence surfaced at the point of decision.
Trust signal design: designed approval-likelihood indicators to surface personalized confidence metrics at the exact point of agent action and decision
UX Optimization: improved visual hierarchy to reduce user cognitive load and denial rates, boosting business approval metrics
AI Matching: leveraged AI to analyze users’ inputs, matching them with cards most likely to be approved
Foodbit: AI-Powered Restaurant Platform
Jul 2021
Core UX Design: architected key interfaces including menu management, category overview and merchant settings
Smart Upselling: leveraged AI to recommend profitable menu combinations and automated size upgrades
AI Customer Profiling: built systems to log user preferences and feedback for highly personalized marketing and services
Data Intelligence - Enterprise scale
Enterprise-scale data — the foundation modern AI agents rely on.
Data at Enterprise Scale: The Interface Went Conversational, the Mechanism Didn’t
I've managed enterprise-scale data for a decade—ingesting, modeling, and refining it to answer real user questions, carrying context through reports and dashboards, and delivering actionable outputs. Today, that exact sequence is called an agentic loop.
NIQ recently launched Cadence, a system of 19 coordinating agents, sitting directly on this foundation. The interface went conversational and the language changed, but the core mechanism didn’t: the infrastructure I architected is the foundation modern AI agents rely on.