A Kubernetes operator that orchestrates AI coding agents

I've been working on itsacoo, a Kubernetes-native operator that orchestrates autonomous software development agents. You give it a concept in plain English, it produces a PRD, HLD, LLD, and task breakdown, then assigns work to AI agents (Claude, GPT, Gemini - whatever you want), manages a merge queue, and ships code. The interesting part: the coo-cli workspace and handoff features were planned by itsacoo's own AARE agent, coded by Claude workers running inside the operator, and merged by the COOMerger controller.

  The hardest problem we hit was merge conflicts. When you have multiple agents editing code concurrently, collision rates are brutal - we saw 80-90% on naive approaches. Our solution is a horizontal slicer that partitions work by domain and layer
  rather than by feature. If two agents would touch the same file, the slicer models that upfront and restructures the task graph to avoid it. Conflicts still happen, but they went from being the common case to the exception.

  The whole thing is modeled as Kubernetes CRDs - COOConcept, COOPlan, COOSprint, COOTask, COOWorker, COOMerger, and so on. This is opinionated and I know Kubernetes is not for everyone, but the tradeoff is real: you get state in etcd for free,
  self-healing reconciliation loops, a built-in audit trail via the API server, multi-tenancy through namespaces, and it runs air-gapped. We also got a 73% reduction in LLM costs by placing cache_control breakpoints on the system prompt and recent
  message context.

  There is a handoff mode that I think is genuinely useful even outside the full operator. Running `coo workspace create --concept my-concept` fetches live CRD state, renders a CLAUDE.md with full context, and injects it into the pod via the Kubernetes
  exec API. When you drop into the workspace, the agent already knows everything - what was planned, what is done, what failed.

  This is early and actively being built. Feedback welcome, especially from anyone running agent workflows at scale.

  https://itsacoo.com
  https://github.com/bobbydeveaux/coo-cli
  Full writeup: https://medium.com/@bobbydeveaux/we-built-an-ai-that-plans-codes-reviews-and-ships-and-then-we-used-it-to-build-itself-2c2a453b35b7