AWS Strands Decider Enables Local Decision Making

The real threat is not another frontier model API, it is that AWS is turning decision making into a cheap, local building block. Strands Decider is a 2B parameter open model for picking options, checking tool calls, routing requests, and scoring outputs, and AWS publishes code, training recipes, and hardware requirements. That means teams can own the model, the weights, and the serving stack instead of paying a per call toll to an outside API.

• Strands Decider is aimed at the same narrow jobs Jev targets. AWS lists model routing, tool selection, argument checking, triage, guardrails, and low cost evaluation as core use cases. In practice, that means deciding which queue an email enters, whether a tool call should run, or whether an answer is good enough.
  1 github
• The open source angle matters because AWS made the model reproducible and easy to experiment with. The repo says the full training recipe can run in about 11 hours on one RTX 3090, and the model can be served on Apple silicon. That lowers the bar for internal fine tuning on a company's own labels.
  1 github
• TypeSafe's counterposition is not openness, it is product behavior. Jev is sold as a hosted decision model with typed outputs, calibrated probabilities, and thresholds that let software auto act when confidence is high and send edge cases to review. That is strongest when the task changes often or when confidence quality matters as much as raw accuracy.
  3 typesafe 4 sacra 5 sacra

This market is likely to split in two. Stable, repetitive decision tasks will move toward open or in house models tuned on proprietary data, while higher drift workflows will favor services that can keep calibration, latency, and developer ergonomics better than a self hosted stack. The companies that win will be the ones that become the default control layer inside agent workflows.