Strands Harness

Commenters debated Strands harness’ launch and its benchmark claims. Strandstan and Albert Zhao emphasized Strands is open source and deployable outside AWS, and the blog touted lower cost and higher Terminal Bench 2.1 scores versus Claude Code. Several people pushed back: seizethecheese and techscruggs argued Terminal Bench 2.1 is saturated and small savings on a saturated benchmark aren’t meaningful, while theturtletalks, fxwin, jsw97, and cobolcomesback questioned benchmark choices - specifically why oh‑my‑pi was tested instead of vanilla Pi and whether comparisons were apples‑to‑apples. Johnmlussier warned that non‑native harnesses risk breaking as models and provider integrations change, and others noted Anthropic/OpenAI terms and managed offerings complicate portability.

Others focused on workflow and cost tradeoffs. Pizzuh, samusiam, and ticulatedspline said building a tailored harness is easy and often preferable for precise needs, and several users described hybrid setups using Pi, Ollama, OpenAI, and custom routing to balance cost and quality (everforward, ex1fm3ta, solarkraft). Stogot preferred a single open harness for consistency. Some criticized perceived AWS salesmanship and enterprise positioning (whattheheckheck, timeout), while agentdev001 and others asked for deeper methodology and follow‑up research to validate performance and real‑world costs.