Codex, OpenAI’s software engineering agent, is expanding beyond coding into broader task execution and decision-making, signaling a move toward more general intelligent agent capabilities.

Codex now works natively with the web via an in-app browser, enabling users to comment on pages to guide the agent with precise instructions.

New enhancements include dedicated GitHub review comment support, multi-tab terminal access, alpha SSH remote devbox access, a new summary pane for agent plans, and a sidebar file preview, improving developer workflows.

Analyst Kieron Allen of Cloud Wars discusses the strategic significance and potential industry impact of these developments.

OpenAI is pursuing a unified super app where AI agents assist across tasks and tools, a vision leadership says is actively being built.

Codex memory, currently in preview, lets the agent remember past interactions, preferences, and corrections, and suggest optimal starting points based on accumulated context.

A background computer use feature lets Codex interact with any app on a user’s computer by seeing, clicking, and typing with its own cursor, enabling parallel operation by multiple agents without interrupting human work.

OpenAI is releasing over 90 plugins (including GitLab Issues, Microsoft Suite, and Neon by Databricks) to expand Codex’s skills, integrations, and MCP server support.

Codex adds expanded automations to reuse conversation threads with preserved context and to schedule and perform work autonomously.

New image generation is supported via gpt-image-1.5, broadening Codex’s creative capabilities.

The article frames OpenAI’s work as redefining the super app concept in the AI era by enabling agents that can answer, act, learn, and decide for users.