March 26, 2026
AI DevOps or AI Overkill?
Show HN: Orloj – agent infrastructure as code (YAML and GitOps)
Hacker News freaks out over ‘Kubernetes for AI agents’ they’re not sure they want
TLDR: Orloj promises to be a master control center for many AI agents, using configuration files and databases to keep them in line. The community is split between people intrigued by the order it might bring and critics who say it’s yet another bloated, Kubernetes-style monster that will create more problems than it solves.
A new tool called Orloj just dropped, promising to run swarms of AI “agents” the way big tech runs servers, and the Hacker News crowd instantly turned it into a street fight over complexity. The name alone sparked its own side quest: one commenter proudly explained that Orloj is Czech for “astronomical clock,” derived from Latin, setting the tone for peak-nerd culture flexing.
On paper, Orloj sounds like “infrastructure as recipes”: you write text files describing your little AI workers, and Orloj schedules, watches, and controls them using databases and message systems. But the comments quickly split into two camps. On one side, curious builders asked real questions, like whether this thing can tame the chaotic, non-repeatable behavior of AI agents and help test them before they wreak havoc in production.
On the other side, the doomsayers saw nothing but tech debt. One user worried they’d be “taking on a lot of debt,” while another unleashed the spiciest take, accusing Orloj of “monolith ambition” and asking, basically, why does this need to look like buying Kubernetes all over again? The vibe: cool idea, terrifying weight. The meme energy is clear—everyone wants an “AI clock,” no one wants to be the one winding it forever.
Key Points
- •Orloj is a runtime for orchestrating multi-agent AI systems using YAML-defined agents, tools, and policies.
- •It supports DAG-based topologies (pipeline, hierarchical, swarm-loop) with fan-out/fan-in and model routing to OpenAI, Anthropic, Azure OpenAI, Ollama, and more.
- •Tool execution is isolated via containers, WASM sandboxes, or processes, with governance enforcing policies, roles, and tool permissions at runtime.
- •Production reliability features include lease-based task ownership, idempotent replay, capped exponential retries with jitter, and dead-letter handling, plus a web console for observability.
- •Quickstart provides binaries (orlojd, orlojctl), local setup, a starter pipeline blueprint, Go 1.25+ build steps, and guidance for scaling with Postgres and NATS JetStream.