Codex adds apps, memory, automations

OpenAI is clearly trying to make Codex stick by making it useful outside the editor, where real work actually happens.

• –Mac app access and tool connections widen Codex from repo work into everyday ops, coordination, and follow-through
• –Memory plus repeatable automations make it more plausible as a long-running assistant, not a one-off task runner
• –Image generation and non-coding actions suggest OpenAI wants Codex to compete with broader agent/workflow products, not just Claude Code or Cursor
• –The upside is less context switching; the risk is a bigger trust and permission surface that will need strong review controls
• –For AI developers, this is another signal that “coding agent” is becoming a thin slice of a larger agent platform