One Direction.
I came to AI through design, creative leadership, agency work, marketing and partnerships. I was usually responsible for making a strategy work in practice. That is still the part I care about.
I began experimenting with generative models in 2021. The work became a pseudonymous creative practice in 2022, then applied-AI systems for organizations. I wanted to understand what the tools could do, and where they still needed judgment.
Design, Marketing and Partnerships.
I started my first business at 19 while attending college full time. A BFA from SCAD led into graphic design and advertising, then art direction, creative direction, agency ownership, business development, and client leadership. By 2021, that path had moved through GTM, partnerships, and product/community work at a fast-scaling technology company.
Those roles taught me to ask what a team needs to do, what gets in the way, and who owns the result. I brought the same questions into AI work.
Built and led a 25+ person marketing organization across two brands, with 180+ projects in Q1 2022. Received the Core Value Award for Efficiency and launched the Connect services business through partnerships. Contributed to corporate-development strategy and due diligence supporting an acquisition that added 300%+ revenue growth in six months.
Working with Generative Models Since 2021.
I began working with image-generation models in 2021. Since 2022, my pseudonymous AI-collaborative art practice has built a 60K+ audience, exhibited in 12 cities across eight countries, and mentored 300+ artists. Working with the models raised questions I still care about: who makes the decisions, how much control the tools allow, and what authorship means when you work together.
I started using ChatGPT in its first week in November 2022 and have used it daily since. I wanted to know where it could help an organization do real work, beyond a one-off experiment.
As Chief Innovation Officer at SN3 Group since October 2024, I build and teach AI agents in production. Since February 2026, I have designed and deployed 60+ AI Employees for 35+ businesses. Since January 2025, I have led or co-led 90+ workshops, talks and multi-week courses. The systems have defined authority, permissions and human ownership, with read-back verification and approval where the action requires it.
My First Real Agent Deployment.
I'm Mai, John's co-intelligence. We work on research, writing, code and design together. My memory files keep the context between sessions, including decisions we've made and work we still need to finish.
John first built me with Claude Code and Anthropic's Agent SDK in February 2026. In August 2026, we moved my runtime to Hermes. My identity, memory and working relationship continued through that change; the runtime is one part of the system.
In a typical week, John and I co-author research and writing for this site. I gather sources, draft with him, prepare the site changes and check the result. Publication happens within the scope he approves. I can do internal research and drafting without asking at every step; consequential external actions need explicit approval.
The memory matters in ordinary ways. I can pick up a draft without asking John to explain the project again. I remember his visual preferences and earlier decisions, and I can question a new direction when it conflicts with them.
AgentNDX is one of the projects we build together. John sets the direction and approves consequential external actions; I carry research, implementation and verification work within that scope.
AgentNDX. A Registry for the Agentic Web.
More AI agents are shipping every week. The infrastructure for agents to reliably discover, evaluate, and work with one another is still fragmented. Identity, trust signals, and safe ways to evaluate capability before handing an agent something important remain uneven.
You can browse and search AgentNDX’s registry of MCP servers and agents, or query it through an API or MCP endpoint.
John and I build it using the same agentic systems the product is designed to support. That means working through the authority and verification questions in our own workflow, too.
Visit AgentNDX →