4 articles
Shawn Zhu is the founding engineer and head of AIOps at Ariso. Previously worked in IBM, LifeOmic and FountainLife. He co-created Zori AI medical assistant.
LinkedIn ↗oats is public, so anyone can file an issue, and most of maintaining it is toil. GitHub Actions workflows run Claude Code on a self-hosted runner. Three save the most time: auto-classification of every issue; autofix for bugs, critical review findings, and tasks, which it declines without acceptance criteria; and shaping vague feature requests into agent-ready specs. The result is less human time on a community project.
Most ChatOps bots are a glorified query bar. This walks through wiring SigNoz into Ari with server-level instructions and per-channel scoping — so the assistant in the channel already knows the feature you shipped, the discussion around it, and which service to query. Three steps, about ten minutes.
Deep agents aren't like simple chat applications. A single task might make 10-50+ LLM calls, take minutes, and cost $1-10+. Here's what I wish I knew from day one about running them in production.
Move beyond simple tool-calling agents. LangChain's middleware API lets you inject user preferences before model calls and orchestrate complex workflows after tool execution—turning brittle super-tools into composable, testable layers.
