https://blog.mikegchambers.com/Personal notes and observations on artificial intelligence, machine learning, and building with foundation models. 2026-10-08T09:52:37+00:00 Mike Chambers https://blog.mikegchambers.com/ Jekyll © 2026 Mike Chambers /assets/img/favicons/favicon.ico /assets/img/favicons/favicon-96x96.png 2026-10-08T00:00:00+00:00 2026-10-08T05:44:40+00:00 https://blog.mikegchambers.com/posts/decider-gated-memory/ Mike Chambers

We listed memory as a use case when we launched strands-decider. I went and built it: a decision model gating long term memory retrieval in a Strands agent. Eight memory calls became three and 147 injected records became 25, with the answers unchanged. The wording of the question mattered more than anything else.

2026-10-01T00:00:00+00:00 2026-10-08T09:52:10+00:00 https://blog.mikegchambers.com/posts/introducing-strands-decider-2b/ Mike Chambers

Strands Decider 2B is a small, open source decision model built for fast experimentation, local development, and innovation. A 2B model that picks between options and scores them in tens of milliseconds, with calibrated confidence.

2026-09-29T00:00:00+00:00 2026-09-29T00:00:00+00:00 https://blog.mikegchambers.com/posts/jev-is-this-tool-call-ready/ Mike Chambers

Gating an agent tool call before it runs, using Jev as the semantic classifier and the Strands Interventions API as the control point. Jev answers a narrow typed question, ordinary Python makes the decision.

2026-09-16T00:00:00+00:00 2026-09-16T00:00:00+00:00 https://blog.mikegchambers.com/posts/jev-is-it-just-a-smarter-if-statement/ Mike Chambers

I joined the AWS Developers Podcast to dig into Jev, TypeSafe AI's new System One decision model, what it actually is, and why it points to a shift away from reaching for one giant LLM for everything.

2026-08-20T00:00:00+00:00 2026-08-20T00:00:00+00:00 https://blog.mikegchambers.com/posts/will-ai-ever-have-original-thought/ Mike Chambers

The question I was asked, taken seriously. Four experiments later: under the only definition we can test, machine originality is already real, provable in a 111k-parameter model. What's missing is expression, not ability. Here's why.