AWS has introduced an open-source agent harness called Strands, which the company says consumes fewer tokens than competing agent systems. The harness is designed to coordinate AI agents on coding tasks, and AWS positions it as a lighter-weight alternative to established tools.
According to The Register, Strands was benchmarked against Claude Code and Codex and achieved near-parity on those tests. However, the evaluation was conducted by AWS itself and focused only on coding agents, meaning the results have not been independently verified.
The token-efficiency claim is the main differentiator, but the self-assessment leaves open questions about how Strands would perform in broader, real-world agent workloads. The project is open source, so outside developers can inspect and test it directly.