On your phone you read the same task as on your laptop. So can the AI also evaluate the same local file again? The conclusion is tempting, but too hasty. A conversation can be visible across devices while a new step involving files still needs the original computer.
In the DevDay update of 29 September 2026, OpenAI described synchronised Work tasks across desktop, web and mobile devices. The current documentation names enabled synchronisation and a new, suitable task as prerequisites; existing tasks keep their original mode. Steps involving local files and tools require the connected computer.
Look at three levels separately
The first level is the conversation: question, answer and further instruction. The second is the source, for example a local events spreadsheet. The third is execution, that is, the step that actually reads or changes this source. Switching screens does not automatically make these levels available in identical form.
Our own example is a short club report built from two artificial files on the laptop. The first task is meant to deliver confirmed observations and open questions. Later, while you are out and about, you only want to make one statement more precise. This follow-up may need the existing interim result; a new calculation from the spreadsheet is a different step.
Set up a limited task on the computer
Name the authoritative files and the desired result. Record which details are confirmed. A manageable set of sources also helps on the small screen: you do not have to explain again which of several drafts is meant.
Create a short report from the two explicitly named sample files.
Separate confirmed observations from open questions.
If a source cannot be reached for a new step,
name the gap instead of guessing its content.
No external publication and do not include any further files.This is a learning task of our own, not a reported device session. For a real attempt, check in your own account beforehand whether the new task actually supports the intended cross-device workflow.
Ask a specific follow-up question on the go
On your phone you could write: “The statement on the trend in participant numbers should only cover the confirmed period. Mark missing comparison data.” That narrows the change and keeps the link to the interim result. It is not an invitation to invent new values.
Choose a task that can be judged sensibly on a small screen. You can often read a short piece of wording more carefully than several wide tables. For image details or extensive files, reviewing them later on a larger screen may still be the right step.
Treat the missing computer as a state in its own right
If a new step needs the local source and the computer is not available, this limit must become visible. An existing text and a newly calculated statement rest on different evidence. When you ask for a new analysis, ask explicitly about the source that was actually used.
A planned comparison can look at two states: computer connected and computer not connected. For each, note which parts of the conversation are visible and whether the specific file step can be carried out. Do not draw a blanket conclusion about the entire local folder from a visible chat.
Trace the return to the working state
On the laptop, check which change the mobile follow-up actually triggered. Was only the wording narrowed, or was a new analysis produced? Does the output still match the same sources? Keep one unambiguous version of the working file.
This workflow was not tested across devices here. Rollout and organisation settings can affect access. The news value lies in a conversation thread you can take with you; it does not imply a complete folder transfer or guaranteed offline continuation. The benefit becomes clear when the conversation, the source and the execution remain recognisable at every step.
Sources and documentation status: 25 October 2026.
