Simon Gonzalez De Cruz builds agent tooling that keeps working after the demo: typed tool surfaces, service-layer authorization, deterministic fallbacks, and the operational discipline of 12+ years running enterprise learning systems before AI existed as a job title.
Every claim below links to something that actually runs.
A free, open-source MCP server, Python library, and CLI that gives AI agents typed video-editing operations instead of raw shell commands: cutting, subtitles, audio, effects, repurposing, and preflight checks that catch bad output before it renders.
github.com/KyaniteLabs/mcp-video - 67 stars / 18 forks (live, 2026-07-06)
Two Macs, a GPU NUC, and a cloud VPS networked over Tailscale, running background workers against a self-hosted Git and CI control plane, with identity-verification scripts so a worker can never act as the wrong node. Built to find where sub-agent systems actually break at scale.
Service-layer authorization enforced once and shared across every surface: MCP tools, a REST API, a CLI, and a web dashboard all call the same permission checks, so access rules cannot drift between interfaces.
Before the first MCP server, Simon spent 12+ years running the training and compliance systems that large organizations depend on. That background is why the guardrails in his agent tooling are not decorative.
Administered Workday Learning for 8,000+ associates, coordinated 10-15 global training programs a year, and diagnosed a broken compliance-training workflow before "guardrails" was an AI buzzword.
Project and program analysis inside SAP SuccessFactors, turning operational reporting needs into something leadership could act on.
Bilingual training operations and disaster-response logistics, delivered in English and Spanish under real time pressure, not a classroom exercise.
Long Beach, CA - remote-friendly. Delivers technical and stakeholder communication in English and Spanish.