Coffee & Code #3: The industry is doing all kinds of acrobatics to build in governance, observability, auditability, and determinism into complex AI workflows. Usually these 4 things require 6 vendors. It really should be just one tool.
I'm surprised how it hasn't really sunk in that MCP isn't just a plug in that let's you talk to some app. It's a protocol that you can use for your own purposes, like doing all the things I described in the first para.
Take observability. Just built and shipped (MCP for X) a feature that visually describes the path traversed along a complex decision-logic, so that for every agent run you can see exactly what path was taken by an LLM, and have full traceability. The example video is for a sample healthcare prior auth example.