Hey, I'm Marc

I spent more than twenty years building and scaling technology. I started as the engineer — search engines, data engineering — and ended as CTO, then GM at a global marketplace group. Each step took me a little further from the code. AI brought me back. The distance between an idea and a working system has collapsed. These days I build again.

What I'm building

• hippocampus — A Claude Code–native second brain with Codex support and Obsidian for browsing: drop in messy notes, get back an organized, linked wiki. Plain Markdown in a git repo, no vector database — the intelligence lives in the structure, not the infrastructure.

• stigmergy — The same idea at company scale, live: a team's knowledge, captured where the work happens, filed by an agent into a git-backed wiki. Gates decide what ships — a model can be argued with, a gate cannot.

• flightdeck — What enterprise AI adoption looks like when you run it on evidence: use cases as code, governed runs, and ROI computed from recorded outcomes instead of estimated on slides. It exists because most AI initiatives can't prove they worked.

• build-squad — A five-role engineering team — developer, tester, auditor, UX, documentator — built as Claude Code and Codex agents, with boundaries enforced by tooling rather than trust: the implementer never grades its own work. It's how everything else here gets built.

• imago — A template for Claude agents that keep a memory and a character between sessions, both plain Markdown. Agents read their memory reliably and write it unreliably, so the write has to be the task.

• crablite — A personal agent on WhatsApp, distilled into a few TypeScript files you can read in an afternoon. All its memory is plain Markdown on disk — no hidden state.

• crucible — Isolated agents analyze a high-stakes decision independently, cross-examine each other's reasoning, and a blind judge rules on every objection. It exists because one model agreeing with itself is not a second opinion.

• cortex — Makes a shared Google Drive answerable by an LLM without taking its word for it: every figure served is traced back to the source cell and re-checked by code, and the server refuses to answer before it guesses.

• chalk — An AI CrossFit coach for iOS and Android: balanced daily workouts, adapted to your gear and level.

• tvremote-app — A remote for my Samsung TV and its streaming apps.