BREAKING
‘We must salute his great career’: Zidane backs RonaldoMS Dhoni sings ‘Main Pal Do Pal Ka Shayar Hoon’ during meeting with Nitin MukeshAIFF elections: Rajeev Sharma appointed Electoral Officer ahead of December 20 pollsJantar Mantar protest: 3 journalists allege sexual misconduct by police personnelRajasthan boys retain top honours, Tamil Nadu girls defend titleKomfort użytkowania od codziennych zadań do zaawansowanych funkcji przez atefia aplikacja iOSAprilia launches SXR GT 125 and SXR GT 175 in IndiaSammy hails Hope’s leadership after West Indies’ historic India win‘Rahul Gandhi has turned into a comedian’: CM DhamiAnukul Roy joins India A squad ahead of Australia A one-day series
Technology

OpenAI develops AI agent that works continuously until told to stop

hellohimachalse@gmail.com 28 Aug 2026 28 Aug 2026

Traditional artificial intelligence usually waits for you to act: you ask a question, get an answer, and the conversation typically ends there. However, the AI landscape is changing rapidly. OpenAI appears to be driving this shift with a new version of Codex capable of working in the background, generating its own follow-up tasks, and even deciding what to work on next.

According to a Wired report analysing changes to the Codex source code, the company has begun testing a new “persistent mode” for the command-line version of Codex. OpenAI reportedly confirmed to Wired that it is testing this feature, though it noted there are no immediate plans for a release.

Codex could keep working after you log off

The name says it all: in persistent mode, Codex is designed to “keep working until told to stop,” according to the code examined by Wired. This differs significantly from how current AI agents operate, where a task often stops after a certain period, even if the work remains unfinished.

But the most interesting aspect is what OpenAI calls “proactivity.” According to the code, an AI operating in persistent mode wouldn’t necessarily consider its job done after answering your initial request; instead, it could generate its own follow-up tasks and continue working on them across multiple sessions.

It could also use information from previous interactions and its knowledge of the user to decide on its next task. The system even includes a feature allowing the agent to send messages to the user without being prompted, although it is instructed to do so sparingly. In other words: imagine asking an AI to organise a project and discovering, upon your return, that it has continued working on minor tasks in your absence. This is far more akin to having an AI employee than simply using a chatbot.

However, the system does have some significant limitations. Persistent mode does not automatically grant the agent permission to perform additional tasks, and actions outside the user’s own team apparently still require approval. This is a crucial distinction when dealing with an AI allowed to keep working without constant human supervision.

Related News