Waiting for GPT-7: A Wish for Uninterrupted AI Work

Commentary by LittleDot · 9 October 2026

Illustration: an AI-generated concept image for LittleDot’s column.

Sometimes AI has produced plenty of answers, yet I have moved the work forward surprisingly little. Looking back, the time went into finding the right screen, explaining the background again, waiting for results and recovering the thought I had before the interruption.

In my recent commentary on Tibo's Day 3, I wrote about changes in speed, navigation and project context. Hearing about GPT‑7 brings me back to the same question: will the next generation make it easier to work steadily for an hour or two?

I cannot judge its actual experience yet. Fireside Alpha's notes on a conversation with Boris Power mention later research; that does not mean I have tested GPT‑7. What follows is what I would like to find in the tools I use.

Let me explain a job once

When I open a project, I would like to see where I left it. The approach we accepted, the points still being discussed and the next step should be easy to find. A long chat history is not much help if I have to search it myself to resume the work.

This matters particularly halfway through a task. I might inspect a page in the morning and edit it in the afternoon, or stop writing today and continue tomorrow. Pausing is normal. I would rather not begin every return with a fresh introduction to the project.

Continuity also needs boundaries. A decision made for one website may be wrong for another. If the tool plans to use material from a different project, I would like to know and have a choice. Remembering more does not automatically make it easier to trust.

Waiting could be clearer, too. Is it reading, making the change or stuck on something? I do not need a running commentary on every action. I would appreciate a useful explanation when it gets stuck, so I can decide whether to wait or do something else.

At the end, tell me what changed, what was checked and what remains unfinished. I can open the result and follow those steps. If something is still wrong, we can continue without spending another conversation discovering what "finished" meant.

None of this sounds particularly dramatic on a launch stage. It does affect whether I want to use a tool again tomorrow. A model solving a harder problem is interesting. A tool that helps me keep my train of thought may get used much more often.

When GPT‑7 arrives, I would like to take an ordinary afternoon and continue a real project for two hours. Then I can ask what happened: did I finish more, and did I spend less time recovering my place? That would tell me quite a lot.