AWS debuts Strands Harness, 26% more efficient than peers

AWS is pitching Strands Harness at a specific moment: developers have gotten comfortable building AI agent prototypes on their own laptops using tools such as Anthropic's Claude Code and OpenAI's Codex, because those setups "just worked" locally. Moving those agents into a scalable cloud environment is where things break down. Strands Harness is AWS's attempt to bridge that gap by offering a ready-to-use agent that runs locally or on any cloud, not only AWS.

The design choices reflect that goal. Instead of making developers build bespoke tools for each task, Strands Harness leans on tools the underlying model already knows how to use, and it manages its own context window by offloading tool results to files and caching reused request parts. It also keeps long-term memory across runs and can delegate open-ended subtasks to a helper agent. AWS says these choices paid off: the agent was 26% more efficient than agents built on other frameworks using the same model, and in one test using Anthropic's Fable 5 model it cost 77% less than Claude Code on the same tasks while scoring higher on Terminal Bench 2.1. A primary use case is rapid prototyping, and one AWS team used it to build Strands CLI, a command-line interface that lets even non-coders prototype agents with natural language and export the harness code.

The stakes hinge on whether developers accept AWS's benchmark results as representative of their own workloads; the comparisons come from AWS itself, so outside validation may determine how quickly the harness is adopted beyond early prototyping, particularly among teams that already have a preferred coding agent.