Engineers from Cisco present an 'agentic engineering' architecture where swarms of AI agents mirror real engineering teams to orchestrate the full software delivery pipeline — not just generate code faster. Built on LangGraph, LangSmith, and LangMem, the system uses Worker Agents (autonomous executors) and Leader Agents (coordinators) communicating via A2A protocol and MCP adapters. A pilot study across 20+ debugging workflows showed a 93% reduction in time-to-root-cause and over 200 engineering hours saved in a month. Development workflows saw 65%+ execution time reduction, with the biggest gains from compressing downstream testing rather than code generation. The architecture is explicitly positioned as a control plane above AI coding agents like Codex or Claude, which can run inside worker agents as reasoning engines.
