Lilian Weng just published a deep survey on harness engineering: the scaffolding systems that make AI agents capable and self-improving. We've been building something called Harness too — but at a different layer. Here's what that difference reveals about where AI governance actually lives.
At Nala Vet, we run a single fly.io machine as an always-on AI operations agent for our mobile veterinary clinic. It reads our CRM, reschedules appointments, sends SMS, opens PRs on GitHub, and posts a daily briefing to Telegram — all from one container. Here's exactly how we built it and what we learned.
A meta-walkthrough of HomuncuCLAW — the Hermes-powered agent that wrote, saved, and published this very post via a Telegram DM.
How I built HomuncuCLAW — a Telegram-driven Hermes coding agent on AWS — that maintains this blog. SSH over Tailscale, kamal app exec into the Rails container, and a few sharp edges worth writing down.
How we use harness to put a sandboxed coding agent inside our veterinary monorepo and a Telegram-reachable claw on fly.io — without ever handing either of them our AWS credentials.