AWS has launched Strands, an open-source agent harness designed to simplify agentic AI development. The tool promises to eliminate the complexity of building custom agents by offering a plug-and-play solution that requires only a single line of Python or TypeScript code to work with any AI model provider, addressing the friction developers face when assembling agent primitives from scratch.

AWS claims the Strands harness consumes 28 percent fewer tokens than Claude and GPT models across six benchmark tests whilst achieving nearly equivalent performance scores to established competitors including Claude Code and Codex. The harness achieves this efficiency through default prompt caching, context window compaction when usage exceeds 85 percent, and automatic truncation of tool results exceeding 1,500 tokens. However, the article notes an important caveat: whilst AWS describes Strands as a general-purpose agent, it was benchmarked exclusively against coding-specific harnesses, raising questions about how it would perform against purpose-built general agents.