TLDRocket
1 October 2026
AI’s center of gravity is shifting from talking to deciding, and from model demos to production bottlenecks. AWS shipped Strands Decider 2B—an open-source, 2B-parameter “Jev clone” that returns calibrated choices instead of text, targeting under-150-millisecond, local-friendly latency for routing, tool selection, and guardrails. The same week OpenAI surfaced a similar direction, and together these launches point to a broader pattern: agent systems are being assembled like workflows, where lightweight decision steps keep LLMs focused on the messy reasoning and cheap calls handle the repetitive execution.
Read the full briefing →