Our summary
Strands Decider is a small AI model that chooses from supplied answers instead of writing replies. Marc Brooker, Mike Chambers and Fabio Nonato de Paula explain how it can run on your own computer. It belongs to the same broad class as Jev, but it is a separate model.
Their weather example catches an assistant guessing a city the user never named. Before the weather tool runs, Decider checks whether the proposed details came from the user and whether clarification is needed. Ordinary Python code turns those answers into a next step: ask for the city or allow the call.
The team reports tests of accuracy, confidence estimates and speed, not independent guarantees. Its weather example illustrates a hand-picked policy, not proven protection. The decision model runs locally, but the surrounding assistant uses Amazon Bedrock. Developers can study the released code without assuming the whole demonstration works offline.
Key takeaways
- Use a decision model for a bounded question; use a text-writing model when you need an explanation or reply.
- Check the exact proposed action against what the user said, and keep the rule for allowing it in ordinary code.
- Separate local decision processing from the rest of the assistant, and test reported confidence and speed on your own task.