OpenAI developing AI agents for persistent, autonomous operation
OpenAI is reportedly developing a "persistent mode" for its Codex tool, aiming to enable AI agents to work proactively and for extended periods without constant user input.
San Francisco-based AI research firm OpenAI is developing a new "persistent mode" for its Codex tool, a move that could allow AI agents to operate proactively and for extended durations without requiring continuous user commands. The aim is to enable the AI models to work autonomously, rather than waiting for new instructions.
Code changes in Codex's command-line version indicate the introduction of this "persistent mode," which is currently in a testing phase. While not yet widely available or officially released, OpenAI has confirmed the feature's development. This mode allows users to define the computational resources, token count, and time allocated to a model before it begins a task.
Once activated, persistent mode enables Codex to continue its work until deliberately put into a dormant state. This contrasts with current modes that typically cease operation after a task is completed or after a set period. A new "proactivity" feature within this mode grants the agent the ability to autonomously plan and execute follow-up tasks based on past interactions and its understanding of the user.
OpenAI CEO Sam Altman has previously expressed a vision for transforming ChatGPT into a proactive AI agent. The development of persistent mode and its proactivity features may represent a step toward this goal, potentially increasing the utilization of advanced AI models. The company has encountered challenges with previous attempts at proactive AI products, such as the Pulse service launched and later discontinued last year.