OpenAI's Codex, the company's coding and software assistant, has gained a new automation capability, according to The Decoder. The outlet reports that Codex can now "watch you work once and repeat the task forever" — observing a user perform a task a single time, then reproducing that same workflow on demand afterward.
The framing here is what makes the update notable. Most automation tools require people to spell out instructions step by step, write scripts, or configure rules in advance. A system that learns by observation flips that model: instead of describing what you want done, you simply do it once while the tool records the process, and it handles the repetition from then on.
The Decoder's headline points to the core promise — a one-time demonstration translating into indefinite, repeatable execution. That positions Codex less as a tool you program and more as one you teach by example.
It's worth being precise about what the source actually establishes. The available reporting confirms the new capability and its watch-once, repeat-forever framing, but does not, in this item, detail exactly which kinds of tasks are supported, how reliably the repetition works, what limits apply, or when and to whom the feature is rolling out. Those specifics aren't present in the source provided here.
Why it matters: if software can reliably learn a task from a single demonstration and then carry it out repeatedly, it lowers the technical bar for everyday automation — letting people hand off routine, repetitive work without writing a line of code.