ChatGPT Queue vs Steer: Send Messages While It’s Thinking Without Changing the Current Task

August 25, 2026 Human Experience Reform

Human Experience Reform public Evaluation

Stable Evaluation ID: HXR-EVAL-0032

ChatGPT Active-Task Message Routing: Update, Queue, or Side Conversation

Users should choose whether a new message changes current work, waits for later, or stays conversationally separate.

Positive foundation: ChatGPT now supports mid-run Update or steering in some long-running tasks. This Evaluation preserves that improvement while evaluating the remaining routing choices.

In plain English

During a long-running task, the contributor could send additional input that changed work already underway and considered that useful. But there was no separate way to discuss the project without altering the active task, so the contributor opened another ChatGPT thread.

Why this matters

“I need to correct the task,” “remember this for afterward,” and “let's talk about the project without changing what is running” are different intentions. When every new message implicitly steers active execution, useful conversation can accidentally alter work or force the user to manage parallel threads manually.

Current findings

Positive practice: mid-run Update reduces correction delay and can avoid restarting long work.

Remaining gap: the observed interface did not expose an explicit semantic destination between updating the active task, queuing a later instruction, and continuing a side conversation.

Limit: community requests and Codex comparators provide language and feasibility context, not proof of one shared ChatGPT implementation or a product roadmap commitment.

Evidence and limits

Direct contributor testimony supports the active-task interaction and second-thread workaround.
OpenAI release notes document current mid-run Update/interrupt behavior.
OpenAI-hosted community requests use phrases such as “temporary side threads,” “message queue while GPT is generating,” and “do not affect the main chat context.”
No population prevalence, technical cause, or delivery commitment is established.

Reform requested

Preserve Update/steering. Add an explicit destination for messages sent while work is active: Update current task, Queue for after current task, or Side conversation that leaves active execution unchanged until the user deliberately applies or merges something from it. Show the routing effect before sending and preserve provenance when side content is later applied.

Acceptance test

Start a long-running Text Mode task. The user can intentionally send one message to steer current work, one to wait until the task finishes, and one to a side conversation that does not alter active execution. Each destination and effect is understandable before sending, recoverable afterward, and available without opening an unrelated thread.

What could change HXR's view?

A current ChatGPT routing control or documented behavior HXR did not observe.
Design evidence explaining context, safety, or memory constraints.
Implementation and test evidence for queue and side-conversation modes.
User research showing a different control achieves the same agency reliably.

OpenAI: respond or document reform

An authorized representative may clarify current behavior, supply positive or challenging evidence, propose a routing model, document implementation, or request verification.

Start a private organization response

Have you wanted to message ChatGPT while it was thinking?

Describe whether you wanted to steer current work, queue something for later, or talk alongside it; what the interface did; and what workaround you used. Remove private conversation content and account details.

Back to blog