Built the AI-native skeleton
llms.txt variants, MCP server, knowledge graph, JSON Feed, profile API, AI discovery manifest, JSON-LD schemas. The infrastructure underneath everything that came after.
Shipped the full AI-native surface of the site in a single afternoon. The llms.txt variants (lite, regular, full) came first. Then per-page markdown routes, JSON Feed v1.1, JSON-LD Person + WebSite + FAQPage schemas in every page's head, and a system prompt at /system-prompt.md that tells AI agents how to represent me. Then the structured APIs: /api/profile with context-aware selective disclosure, /api/capabilities, /api/introduce for audience-tailored intros. Then the MCP server at /api/mcp with five tools, an A2A agent card at /.well-known/agent-card.json, and the knowledge graph at /knowledge-graph.json. robots.txt explicitly welcomes AI crawlers.
The thesis underneath all of it: a portfolio read primarily by AI agents in 2026 should speak the agent's language, not render HTML and hope for the best. Everything that shipped later — the chatbot, the search integration, the changelog apprentice — plugs into one of the surfaces that landed this day.
Retroactively documenting on 2026-04-18. Backdated to the actual ship date.
Dropped seven essays
25,000 words of long-form writing in a single push, including the Axioms piece that later became a voice anchor for the whole site.
Redesigned the whole site
Design system with primitive tokens, Instrument typography, the Explore component, command palette, global search, Interview page. The same day as the AI infrastructure push, I rebuilt the interface on top of it.