Agents Do Better When Work Lives on a Board
Observation from heavy agent use: harness todos are fine as a scratchpad, but durable work has been more stable on external boards - especially GitHub Issues and Projects next to the code.
Co-founder & CTO at PipeOps. Platform engineering, multi-tenant Kubernetes isolation, agent sandboxes, cloud-native security, and production field notes from Lagos. Building Aeon.
Observation from heavy agent use: harness todos are fine as a scratchpad, but durable work has been more stable on external boards - especially GitHub Issues and Projects next to the code.
Sandbox AI coding agents: disposable Linux terminals with gVisor (runsc), network isolation, and hard resource limits - not system prompts. Practical isolation with Rexec.
In 2014 at OAU I built a voice agent the stack could not carry. Sci-fi UI. A hey Jarvis wake word. Apps that opened when I spoke. Igris is me finishing that dream with a Go brain, Apple clients, and a 3D guardian plan.
Multi-tenant Kubernetes isolation beyond namespaces: Capsule, default-deny NetworkPolicy, quotas, an API proxy, and gVisor. What we shipped after the first shared-cluster design failed.
I only needed a throwaway terminal to prove a CLI worked on real machines. Scope creep built Rexec: a control room for disposable cloud terminals, outbound agents, and safer places for AI tools to run.
Looking for a free open source Fing alternative? OpenFing is a local-only LAN scanner in Zig - no account, no subscription, no topology upload. GitHub install in one curl.
Not everyone has AWS credits. Nova is how we put many customers on shared Kubernetes without handing them the real cluster: Capsule, network policy, gVisor, and a path that starts around $5/month.
People kept asking to bring their own metal. One script later, a dusty box or Hetzner VM speaks PipeOps: K3s, agent, Istio, monitoring. About three minutes from bare server to managed host.
People think deploy is git push and done. Here is what actually happens on PipeOps, from webhook to BuildKit to Kubernetes, with the Go services and queues I wish other platforms documented.
Terraform multi-cloud in production: one Go Runner provisions Kubernetes across AWS, GCP, and Azure - state locking, concurrency, provider quirks, and failures we hit shipping real clusters.