Async Tool Calls: Stop Your Agent Blocking on Slow APIs

OpenAI shipped GPT-6 Astra on September 3, and most of the noise since has been about benchmarks. The Hacker News threads argued about whether it's really a generational jump or "more like 5.7." That argument buried the change that actually rewrites how you build agents: async tool calling.

Here is the problem it solves. Every agent you have written until now has a dead spot in it. The model asks for a tool, your code runs that tool, and the model sits idle until the result comes back. If that tool is a 4-second vector search or a 10-second CI job, the model is doing nothing for 4 or 10 seconds — even when the user also asked something the model could answer immediately, from memory, with zero dependencies.