What's the difference between the Claude Agent SDK and AWS Strands Agents?

The Claude Agent SDK wraps the Claude Code binary as a library and runs only Claude; AWS Strands Agents treats the model as swappable, reaching Bedrock, Anthropic, OpenAI, Gemini, and Ollama through one interface. Anthropic’s SDK ships built-in file, shell, and web tools, hooks, and subagents, all wired to Claude Code’s own loop. Strands ships five named multi-agent patterns as first-party primitives instead: agents as tools, a swarm of agents handing work to each other, a graph of developer-defined nodes where an LLM picks which edge fires at each step, a deterministic fixed-DAG workflow, and cross-process A2A; the Claude SDK’s nearest equivalent is a subagent, one level of delegation with no swarm or graph construct of its own.

The gap widens in what each hands you for grading a run. Strands integrates natively with OpenTelemetry once you install its otel extras and turn tracing on, exporting to an OTLP endpoint with the same GenAI semantic-convention attribute names any OTel-reading backend already expects. Claude Agent SDK telemetry is OpenTelemetry too, but spans still sit behind a beta flag and carry Claude Code’s own names rather than the GenAI ones, so anything reading them needs a mapping step Strands doesn’t require.