OpenAI WebSocket Mode for Responses API

Alex Zielinski, Andrew Codispoti, Peter Bakkum, Ben Bartlett, and David Carr just made building persistent AI agents a hell of a lot faster. Forty percent faster, to be exact. WebSocket Mode for the Responses API is OpenAI's answer to the biggest pain point in agent development—that clunky request-response cycle that makes AI feel like a telegraph operator from 1844. Now you get a persistent connection. Your agent stays alive, thinks in real-time, and doesn't have to reconnect every single time it needs to say something. For anyone building AI agents that actually need to hold context and act continuously, this is the difference between a chatbot and something that feels like a real assistant.

The team here is quietly stacked. Ben Bartlett is a physics PhD from Stanford who was doing photonics and quantum research before falling into AI—your classic "let me just solve the hard problems" type. Andrew Codispoti is described in his own GitHub bio as a "Vim wielding pythonista," which tells you everything you need to know about the kind of developer-friendly engineering that went into this. These are the people who actually use the APIs, building tools for themselves and everyone else who's trying to ship agents that don't suck.

This is OpenAI doing what OpenAI does best—making the underlying infrastructure invisible so developers can focus on what their agents actually do. The 40% speed gain isn't a gimmick, it's the difference between an agent that feels responsive and one that makes users want to throw their laptop out the window. If you're building anything with persistent state—customer service agents, coding assistants, autonomous workflows—your life just got significantly easier.