I experienced a real-world example today that demonstrates exactly why this capability is needed.
I am a paid ChatGPT user and had a lengthy text conversation about a specific issue I was experiencing with ChatGPT Voice. ChatGPT had enough context to understand the problem, help me reproduce it, review screenshots of the behaviour, distinguish the expected experience from what actually occurred, and formulate a clear product issue.
At that point, I explicitly wanted the issue escalated to OpenAI.
ChatGPT could do everything except complete the final transaction.
Instead, I had to leave the conversation, locate the appropriate support channel, interact with another AI support system, answer additional intake questions, explain what I was trying to report, and then be told that the matter had been escalated to a specialist who would contact me later.
That is unnecessary friction for both the customer and OpenAI.
I lead global enterprise support/service-delivery functions and work with AI-enabled transformation, so I look at this not only as a frustrated customer but also through a service-management lens. The missing capability is not intelligence. ChatGPT had already understood and structured the issue. What was missing was permissioned end-to-end process integration.
A better workflow would be:
User: “Submit this as product feedback.”
ChatGPT:
- Determines whether the issue is appropriate for product feedback/support.
- Creates a structured submission using the conversation context.
- Shows the user exactly what will be transmitted, including any diagnostic or account information.
- Requires explicit user approval.
- Routes the submission to the appropriate internal system.
- Returns a confirmation/reference number.
Nothing should be submitted autonomously, and the entire conversation should not automatically be shared. The user should explicitly authorize the exact payload.
This would reduce customer effort, duplicate explanations, abandonment, misrouting, and unnecessary support interactions while potentially improving the quality of information reaching product and engineering teams.
The irony of my experience was striking: ChatGPT was sophisticated enough to diagnose and articulate the product problem with me, but I then had to leave ChatGPT and manually navigate another AI system simply to tell OpenAI what ChatGPT had already understood.
The intelligence is already at the point of need. Give it a controlled, consent-based mechanism to complete the transaction.