Two things that have quietly bottlenecked production AI systems are finally getting dedicated tooling: fast, reliable decision-making inside agent loops, and retrieval quality in RAG pipelines. This week’s cluster of releases from Cloudflare, AWS, Cohere, and Perplexity/turbopuffer makes that shift concrete. Instead of asking a general-purpose LLM to decide whether to call a tool or how to route a request, teams are increasingly plugging in small, typed decision models built for exactly that job. For anyone doing serious LLM application engineering, this is a meaningful architectural signal, not just another model drop.
Decision Models Are Becoming the New Control Layer for AI Agents
