If you only use chat-style AI and you're just starting to follow new AI tools, Quoting OpenAI can look like a tiny formatting problem. That is where people waste time. You see the update, almost scroll past, then wonder if you're missing the one thing that changes your next move.
The useful call is bigger than footnotes: OpenAI is not a chat box for every job. For complex work, treat it like a job queue. Queue it, poll it, then finish the quoting cleanly. Complex tasks should be queued and polled, not forced through a single live response.
If you only chase surface features, you spend time, budget, and attention in the wrong place. The hidden cost is worse: you keep polishing the last visible step, like quote formatting, while missing the workflow change underneath.
The official pattern in the brief points the same way from two sides. Background mode is for harder jobs that can take minutes, with states like queued and in_progress, plus cancellation [S002]. The Batch API packages requests into a .jsonl async job when you do not need an immediate response [S003]. That is not chat-box thinking. That is job-queue thinking.
So for Quoting OpenAI, my practical rule is simple: use 1 stable source ID per quote block after the job is done, instead of pretending every longer task should behave like an instant chat reply. When you're judging an update, don't ask how many features it lists. Ask whether it changes your next move.
One boundary matters: do not make everything async and ruin real-time experience. Use live replies when speed matters. Use queue-and-poll when the task is long, multi-step, or needs traceability. If this sharpens the decision for you, share it with the person still treating every AI task like a chat bubble.