OpenAI speeds up AI agent workflows using WebSockets, cutting API overhead and latency in the Responses API

OpenAI published technical improvements to its Responses API that use WebSockets (a persistent two-way connection between client and server) and connection-scoped caching to reduce how much data travels back and forth during agentic workflows (tasks where an AI makes decisions and takes actions on its own).

The change eliminates redundant API calls by caching information tied to a single connection, so repeated requests don't force the AI to reprocess the same data—think of it like remembering context across a conversation instead of starting fresh each time. This reduces latency (the delay before an AI produces a response) and lets the API handle more concurrent requests without strain.

Developers building AI agents—chatbots that automate customer service, code review systems, or data analysis pipelines—can now deploy these workflows with lower costs and faster response times, making real-time agent interactions more practical for production applications where speed and cost directly affect user experience.