I design work.
You've shipped the platform. Hired the team. Funded the AI pilot. And the work still doesn't move like it should. That's a design problem.
Most AI initiatives never make it to production.
The model rarely fails. The system around it does — the workflow, the decisions, the data, the governance. That's the layer I design.
of enterprise generative-AI pilots deliver no measurable impact on the P&L. The cause is integration, not model quality.
MIT — The GenAI Divide: State of AI in Business, 2025of organizations scrapped the majority of their AI initiatives in 2025 — up from just 17% a year earlier.
S&P Global Market Intelligence, 2025of generative-AI projects get abandoned after proof of concept — on poor data, weak controls, and unclear value.
Gartner, 2024Design makes AI work.
My work involves these industries and platforms. Now I apply my expertise in experience design to AI systems architecture.
The human layer
Select a layer to explore it.
What I help you do
Systems are the pipework that keeps agents, workflows, and people moving together, even as the work shifts.
Decisions are about who owns the call, how it's made, and the trade-offs the business will stand behind.
Governance is the guardrails that make agents safe to run in production, without turning the work into theater.
Selected work
Every email sent successfully. None arrived.
Setting up business email looked like a five-minute task. It wasn’t. A stale mail route, a typo’d address quietly eating leads, and an email pipeline that reported success while sending nothing — all hidden by fail-soft code. How doing it right surfaced what “good enough” had buried.
Infrastructure & Reliability · 2026
PlanFlow: Structuring how plan reviews get decided
A plan submission and review system for agencies managing permits and approvals, with structured workflows, role-based reviews, and AI that sharpens human judgment instead of replacing it.
Government & Permitting · 2026
The bot was secure. The system wasn't.
This site’s own lead bot had 17 security layers and answered every question politely — while running completely unprotected. How a routine audit caught it, and what fail-safe really means.
Applied AI Systems · 2026
- Claude
- OpenAI
- GitHub
- Google Gemini
- React
- Next.js
- Supabase
- TypeScript
- Neon
- ElevenLabs
- Convex
- Streamlit
- AWS
- Pinecone
- Slack
- Telegram
- Microsoft Teams
- Confluence
- Jira