Personal agents learn to watch the clock: Agents API adds time awareness so briefings land before the deadline
Synopsis
OpenAI added time awareness to the Agents API, giving the model periodic reminders of the current time during a task, paired with asynchronous function tools that let it keep working while waiting for a user's answer. In a meeting-preparation example with a meeting 15 minutes away and a briefing due in 12, the agent can judge when to keep searching and when to verify findings and start writing; if the user never answers, the application closes the question two minutes before delivery and returns a no-answer status, and the agent still delivers a briefing by the deadline.
Interpretation
A personal agent given a task with a deadline can now pace its work against the time remaining instead of searching until something outside stops it. Previously the application watched the clock on its own, and the model did not know how much time was left before delivery, so it could not choose between opening another line of research and starting to write the briefing. The meeting-preparation example sets the meeting 15 minutes out and briefing delivery 12 minutes out, and the agent prioritizes checking its strongest findings and writing the briefing over opening another line of research.
The implementation separates the clock reminder from the question: the model receives the current time periodically, and the question goes out as an asynchronous function tool. Time awareness is enabled with the experimental.time_awareness flag, so the model keeps getting time updates while searches and user input are pending; the asynchronous function tool is marked async, and the application keeps receiving the agent's updates while it collects the reply. In the example the agent receives the task at 4 p.m. UTC, must deliver by 4:12 ahead of a 4:15 meeting, and the application closes unanswered questions at 4:10, returning no_answer as the tool result.
Perspective
The demonstrated scope is one meeting-preparation task: the agent researches a company, asks about the meeting's focus, and delivers a briefing before the deadline; the application must convert relative times such as "in 15 minutes" into absolute timestamps so the model and application share the same time data.
The material states that if the question goes unanswered, the agent should still have enough context to write a useful briefing; if the deadline arrives first, it returns the latest completed briefing or reports that no verified findings are ready.
