The fix that makes long AI conversations faster — and what it unlocks

Right now, when you build an AI app — a support tool, a coding assistant, an agent — something wasteful happens each time a user sends a new message: your app resends the entire conversation from scratch. Every previous message, every AI response, all over again. The longer the session, the worse the wait.

Anthropic's Message Threads, now in public beta on the Messages API, fixes this. Your app sends the full conversation history once. After that, you only send the new messages — a tiny fraction of the data. The server holds the thread in memory, encrypted, and continues from where it stopped.

Anthropic says this cuts wait time by about 35% for long sessions — conversations with 500,000 to 1 million tokens. Tokens are the small chunks of text an AI reads and writes — in practice, that is a conversation many pages long. In a customer support queue or a multi-step agent workflow, users feel that difference immediately.

One thing to know before you build on this: threads expire five minutes after the last message and are never written to disk. That keeps them private — but it also means they are not eligible for HIPAA workflows. Healthcare and other regulated industries will need to wait. For everyone else building agents and assistants, the right move is to try this now — it is free to test and takes a single line of code to add.