PRODUCTION SCENARIO
Your support-tooling team wants an agent that reads repository files, runs tests, and edits code. It has to run inside the team's own VPC, hosted in the long-running TypeScript service they already operate, and they want typed message objects inside that process rather than parsing another program's stdout. They do not want to write the tool-execution loop themselves.
Which offering should the team build on?
Answering here is anonymous. Nothing is saved unless you sign in.
Show answer and explanation
Answer: The Claude Agent SDK, a TypeScript library with Claude Code's own tools and agent loop
The Agent SDK is the library form of Claude Code's tools, agent loop and context management, shipped for Python and TypeScript, so it drops into a TypeScript service the team already runs and leaves them no loop to write. Managed Agents is the right pick when you would rather not operate a sandbox or session infrastructure, but Anthropic runs the agent, which puts it outside the team's VPC. Driving the CLI as a subprocess is the documented route for languages the SDK does not ship in, and it hands back a stream to parse rather than typed objects in the service. The Client SDK is the right level when you want direct API access and intend to own the loop, which is the one job this team is trying to avoid.
Official Source: https://code.claude.com/docs/en/agent-sdk/overview ↗