CI/CD meets agent QA

OpenObserve’s engineering blog says it built a “Council of Sub Agents” — eight Claude Code agents — and expanded coverage from about 380 to over 700 end‑to‑end tests while cutting flaky failures roughly 85%. (openobserve.ai) Airwallex published a post describing a Claude Code sub‑agent workflow that shortened a previously two‑week integration test cycle to roughly two hours. (careers.airwallex.com) Public repositories cataloging community tooling now list dozens or hundreds of specialists: VoltAgent’s “awesome-claude-code-subagents” curates 100+ subagents, and a separate GitHub project documents a production system with 112 specialized agents and 16 orchestrators. (github.com/VoltAgent) (github.com/wshobson) Anthropic’s Claude Code documentation details the official “sub‑agents” feature and shows each sub‑agent runs in an isolated context with distinct prompts and tool permissions, a capability critical for parallel QA tasks. (code.claude.com) A developer post and plugin notes show a recent fix (Claude Code 2.1.30+) that allows sub‑agents to access SDK/MCP tools reliably — a change teams cite as necessary before delegating tool‑invoking tests to sub‑agents. (github.com/leary-comos/claude-night-market) Companies already using agentic QA are combining those sub‑agent fleets with API automation pipelines — for example, an engineering blog from Altruist describes an “API Agent” that autonomously generates and maintains executable API tests across a test suite. (altruist.com) The visual CI/CD frames that went viral mirror standard pipeline templates — commit → build/test → deploy → monitor — which are the same stages used when teams insert agent‑driven QA as an automated testing layer before deployment. (miro.com)