AWS released Strands Harness, a ready-to-run AI agent. It includes web search, shell access, memory, and file editing. Developers supply and can swap the model at any time. It supports models from Amazon Bedrock, Anthropic, OpenAI, and Google. A local Ollama model can also be used. It costs 26 percent less across seven benchmarks. AWS claims it outperforms Claude Code and Codex in token efficiency. It keeps long-term memory across runs and resumes with a session ID. It delegates subtasks to a helper agent and tracks work with a checklist. It manages its own context window and offloads bulky results to files. It is designed for operations, DevOps, back-office automation, and conversational experiences. It is available via PyPI, npm, and a CLI. It runs on laptops, CI pipelines, private servers, or cloud deployments.
What changed
Developers built agents from the Strands SDK with manual setup for context, tools, and prompts.
What it unlocks
Running a full agent with web search, shell, memory, and file editing by supplying only a model.
- 26% cheaper across seven benchmarks
- 77% cheaper than Claude Code on Terminal-Bench 2.1
What you need to act on it
- a model from Amazon Bedrock, Anthropic, OpenAI, Google, or a local Ollama model
Sources