OpenAI wants to teach Codex not to stop after completing one task. The company is testing Persistent mode. This is a mode in which the AI agent will be able to continue working, find the next useful steps on its own, and return to tasks even after a pause.
OpenAI wants to teach Codex not to stop after completing one task. The company is testing Persistent mode. This is a mode in which the AI agent will be able to continue working, find the next useful steps on its own, and return to tasks even after a pause.
WIRED drew attention to a new feature in the public Codex code.
An OpenAI representative confirmed to the publication that the company is indeed experimenting with such a mode, but does not plan to launch it for everyone in the near future.
Persistent mode has appeared among the reasoning effort settings — parameters that determine how much time and computing resources the model can spend on answering. In the code, it is described as “Continue working until put to sleep.”
And the completion of the response to the user will not necessarily mean the end of the agent's work. Codex will be able to decide for itself whether it makes sense to do something else. For example, wait for the result of a running process, check whether the change made worked, or continue the work started earlier. To do this, it will be able to take into account previous interactions and "knowledge about the user."
According to WIRED, the AI assistant will also have the ability to write to the user itself, although the instructions call for doing so only when a truly important result, obstacle, or human intervention is required.
Previously, dev.ua wrote that the company is acquiring startup Ona, whose cloud environments allow AI agents to continue working even after the user closes the laptop.
