Platform Squeeze, Agent Governance, and the AI Tool That Finally Got Real Microphone Support

Three signals worth sitting with today: AWS published a concrete four-step governance model for how enterprises should grant agents tool access, LangChain released a post on 100x cheaper LLM evaluation that has real cost implications for anyone running production evals, and Zachary Lipton dropped a sharp observation about platform companies undercutting their application-layer partners — which is the most honest framing of the current AI stack risk I've seen this week.