AWS targets AI cost worries with open source agent harness

AWS launched an efficiency-focused agent harness on Monday as part of its open source Strands SDK, which AWS engineers say runs general-purpose agents at a 28% lower cost when using Claude or GPT models with nearly equal benchmark scores. Principal Engineer Arron Bailiss wrote in a blog post that default context management drove the token-efficiency and accuracy gains, with tool results over roughly 1500 tokens truncated and summarization compaction triggered when the context window exceeds 85%. The harness SDK is available on GitHub under an Apache 2.0 license as part of the Strands monorepo for building and running agents in Python and TypeScript.

As the cost of scaling AI agents gets higher up the CIO priority list, AWS has added an efficiency-focused agent harness to its open source Strands SDK for building agents. The Strands harness launched on Monday, described by AWS engineers as able to run general-purpose agents at a 28% lower cost when using Claude or GPT models with “nearly equal” benchmark scores. Principal Engineer Arron Bailiss wrote in a blog post https://strandsagents.com/blog/introducing-strands-harness/?ref=thestack.technology : “Our default context management largely drove the token-efficiency and accuracy: tool results over ~1500 tokens get truncated, and summarization compaction triggers when the context window hits above 85%.” The harness SDK is available on GitHub https://github.com/strands-agents/harness-sdk?ref=thestack.technology under an Apache 2.0 license as part of the Strands monorepo for building and running agents in Python and TypeScript.