Who Runs the Loop? From Chat Completions to Responses API

Chat Completions makes you own the tool loop. The model asks for a tool, your app runs it, sends the result back, and calls again — four round-trips, all of them your code. The Responses API hands that loop to the provider runtime, along with tool execution, conversation state and reasoning continuity. Chat Completions isn't going anywhere, but if you're starting a new agentic app, the code you get to delete is the point.

We'll migrate a real agent live in both C# and Python — hosted tools, remote MCP, the three ways to handle conversation state, typed streaming — and cover what Microsoft Agent Framework buys you when vendors don't agree on a protocol. Plus the honest answer to "is everyone moving to Responses?"