Gee

Gee has more than twenty years in IT operations and has spent the last few years building AI automation, most often as the sole designer and engineer, and at times leading a small team.

Gee builds AI systems that do real work without a person watching, and builds them to be operated: scheduled jobs that leave a person a quick review instead of a manual chore, agents that propose actions behind approval gates, and audit trails that record what happened.

In client and employer work, that has meant department agents that do the routine work while every consequential action waits for a person, and AI receptionists that take an enquiry to a confirmed booking over SMS, voice or web chat, with code re-checking every slot. It has also meant the plumbing underneath: MCP servers that give assistants read-only access to business data, syncs that bring many systems into one store, and a scheduling platform taken from interviews and workbooks to production.

His own projects show the same habits. A small public reference agent uses a deny-by-default tool registry, checks its own output and keeps an append-only audit log. In a self-hosted assistant running open-weights models, a small model was measured treating rules in its prompt as suggestions, so the safety-critical answers were moved into code and every free-form reply passes one gate. A working web app was rebuilt on Cloudflare Workers and D1 instead of being patched, with deploys gated behind an automated test suite. His Claude Code skills have Claude, Codex and Gemini review the same work independently, then combine what they find.

Underneath the AI work is that IT operations background: Linux administration, networks and security, telephony, and backup and recovery.