What happened
AWS open-sourced Strands Harness, built on the Strands Harness SDK it released in August 2026.
Why it matters
The harness is open source, so teams can add features and customize it themselves.
What to watch
The benchmark is based on Fable 5, so results on other models may differ from the figures AWS published.
WHO IT HITSTeams that build or operate their own AI agents — internal tooling groups, developers running self-hosted models, and platform engineers inside enterprises — get an open-source harness they can modify instead of building the surrounding plumbing themselves.
Summaries like this, every morning.
Strands Harness sits on top of the Strands Harness SDK AWS released in August 2026. The pitch is one import instead of hand-wiring tools, memory and context handling. AWS says an agent built with it keeps per-session memory, so giving it a session ID resumes an earlier conversation.
The defaults handle context automatically: tool results over about 1,500 tokens are truncated, summaries kick in when the context window passes 85%, and context is restored inside the loop if it overflows. Out of the box an agent can run shell commands, read and write files and search the web.
A few lines of code spin up an agent, and swapping in another model is a single string change — for example from Claude Opus 5 to Google Gemini or Ollama. Agents can also be assembled from a command line and exported as TypeScript or Python. The open question is how much the benchmark result carries over, since it was measured on Fable 5; whether teams see the same cost and score on the models they actually run is likely to decide how widely the harness is adopted.
Pick your industry and the AI tools you use, and get news related to your work every day.
Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →
Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.
Google's Gemini 4 Carbon took the #1 spot on MathArena with an expected 85.4% score, displacing OpenAI's GPT-6…
In a test of PurpleDoubleD's locally-uncensored app (AGPL-3.0-only), a coding agent was told to fix a one-line…
The developer behind zero-trust agent tool ACP ended its development, renamed the repo to sealbridge, and publ…
On November 10, 2026, Anthropic released Claude Sheets in beta (building spreadsheets from prompts, 144 suppor…
A practitioner writes that after enriching specs and CLAUDE.md rules, AI still produced off-intent behavior, a…
The author fed the 'desired candidate profile' text from 145 job postings to Claude, turned the extracted labe…
