OpenAI Pushes Ahead with New Always‑On Agent
agents openai
| Source: Mastodon | Original article
OpenAI continues to develop a new always‑on AI agent, signaling its commitment to persistent, real‑time assistance.
OpenAI is quietly adding a “persistent mode” to its Codex coding assistant, a development first spotted in the tool’s public repository. The new code, flagged by Wired and echoed by Gizmodo, enables the command‑line agent to stay active until a user explicitly puts it to sleep, spawning its own follow‑up tasks and adjusting its reasoning effort on the fly. In practice, Codex could continue working on a project, generate intermediate steps, and react to new inputs without being re‑invoked for each command.
The move marks a shift from the on‑demand model that has defined most consumer AI tools toward a more autonomous, always‑on paradigm. Persistent agents promise to streamline complex development pipelines, allowing developers to delegate longer‑running or multi‑step coding chores to the AI. At the same time, the capability raises fresh safety questions: an agent that can act proactively and generate tasks independently may drift from user intent, consume resources unchecked, or expose new attack surfaces. OpenAI’s internal controls for “proactivity” and the ability to pause the agent hint at an awareness of these risks, but the details remain undisclosed.
OpenAI’s push follows its earlier announcement in April that it was testing a ChatGPT‑based “Hermes” platform for custom, always‑on agents with bespoke skills and workflows. The Codex persistent mode appears to be the next iteration of that strategy, extending the concept to code generation. Observers will be watching for a formal rollout, documentation of safety guardrails, and how the feature integrates with OpenAI’s broader agent ecosystem. The timing of a public release, pricing model and any regulatory scrutiny will shape whether persistent AI assistants become a mainstream productivity tool or remain a niche experiment.
Sources
Back to AIPULSEN