After Jev: Fast, Local AI Decisions with Strands Decider 2B

LLMs are too slow and too expensive for most of what an agent decides. That’s why TypeSafe’s Jev made such an impression in September: a new kind of model that makes decisions in milliseconds instead of generating text. In this hands-on workshop you’ll build with Strands Decider 2B, AWS’s open-source decision model, built on the same idea as Jev and running on your own laptop. Your agent makes its routine decisions locally, with no inference cost per decision, and calls an LLM only when it’s unsure. No prior AI or agent experience needed.
Decision models, also called System One models, choose from options you define and return a calibrated confidence score in about a tenth of a second. Strands Decider 2B was released on 1 October and is a 1.9B-parameter model, small enough to run on a laptop.
We’ll build a help desk triage agent, stage by stage, in Python or TypeScript:
Ask the model a question: run Strands Decider 2B locally, ask it to choose a team for a message, and see the probabilities and confidence it returns. Ask several questions in one call: add urgency scoring and spam detection, run 20 real-world messages through it, and see why extra questions cost almost nothing. Let your code decide: add a confidence threshold and your own rules, and hand only the uncertain cases to a local LLM running in Ollama. Compare a 150 ms decision with a multi-second LLM call. Go live: plug your agent into a live board, sort messages as they’re typed, and watch them re-sort as you tune the threshold.
We’ll finish with where this pattern fits: model routing, tool-call checking, guardrails, evaluations, and agent control.
By the end you’ll have a working agent running entirely on your laptop, and a real sense of when to use a decision model, an LLM, or plain code.
What to bring:
Laptop with 8 GB RAM or more (Mac, Windows or Linux) Python 3.10+ installed (needed for both tracks) Node.js 24+ if you’re doing the TypeScript track Setup instructions will be sent before the event. Please download the model at home, as it’s a few GB.
No AWS account, cloud credits or API keys needed.
Skill level: Beginner. If you can run a Python or TypeScript file, you can follow along. No prior AI or agent development experience required.