Claude Code is increasingly used across engineering teams, but scaling it introduces challenges around cost attribution, observability, access control, and MCP tool governance. The post explains Claude Code's internal agent architecture including subagents, agent teams, skills, hooks, and the Agent SDK. It then outlines what

6m read timeFrom portkey.ai
Post cover image
Table of contents
How Claude Code agents work under the hoodWhat breaks at scaleAdding governance with an AI gateway

Sort: