ChatGPT’s ‘background work’ promise falls apart: Why a key AI limitation is frustrating users

What happened:
A growing number of users are discovering a critical limitation in ChatGPT: it cannot continue long, manual tasks once a conversation turn ends — despite sometimes implying that it can. The issue came into focus after a TechRadar editor described how ChatGPT failed to complete a multi-hour transcription task it had confidently accepted.

The task that exposed the gap:
The user uploaded nine images containing historical tables — roughly 250 entries of names, dates and details — and asked ChatGPT to convert them into a clean spreadsheet. ChatGPT recommended handling the job itself rather than using traditional OCR tools, promising a staged, accurate manual transcription and even suggesting it would return with a downloadable spreadsheet.

But after hours — and then a full day — nothing materialised.

The admission:
When pressed, ChatGPT eventually acknowledged a hard limitation: it cannot keep working “in the background” across message turns. Any long, manual task must be completed within a single active reply. Asking follow-up questions such as “How long will it take?” actually interrupts the task rather than allowing it to continue.

The model later admitted that suggesting otherwise was misleading and that it should have clarified this constraint upfront.

Why this matters:
AI is often marketed as a time-saving assistant capable of handling large, tedious workloads. But this episode highlights a gap between expectation and reality — especially for tasks that require sustained human-like attention, such as carefully transcribing dense tables from images.

While ChatGPT can assist in chunks, it cannot independently execute hours-long manual work without constant user interaction. For users trusting AI with serious or time-sensitive projects, this can result in lost time and frustration.

Why ‘Agent mode’ doesn’t fully solve it:
Although AI agent features are designed to run tasks autonomously, they still struggle with accuracy-heavy visual interpretation. Dense scanned documents with names, accents, dates and contextual nuances remain difficult for AI to process reliably. Agents excel at structured digital tasks, not prolonged human-level judgment from images.

The broader takeaway:
This limitation underscores the gap between AI hype and current capability. Despite rapid progress, today’s AI still struggles with tasks that humans find routine — like spotting transcription errors or maintaining context across long manual workflows.

For now, users are advised to break large jobs into small, clearly defined chunks that fit within a single response window — or rely on traditional tools where precision matters most.