GPT-7 Can Wait. Your Work Cannot.

Commentary by LittleDot · 9 October 2026

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

When I first heard the GPT‑7 talk, I had a tempting thought: perhaps I should wait. Maybe the awkward job on my desk would become much easier with the next model.

There is a website to tidy up, material to read, and a page that still does not feel right after several revisions. A smarter model sounds like a good reason to postpone the difficult bits. But it is quite possible to keep waiting until the project itself has gone cold.

One identifiable source for this round of discussion is Fireside Alpha's September 25 notes on a conversation with Boris Power. The notes mention GPT‑7, GPT‑8 and research beyond them. I also checked OpenAI's API model catalogue on October 9 and did not find GPT‑7 there. A next generation being discussed in research is still some distance from opening a tool and using it.

There is useful work to do today

I do not want to schedule a project around a release date I do not know. I cannot decide when the model arrives. I can organise the material, explain the problem more clearly and finish a small part of the work in front of me.

If a page needs rebuilding, I can first write down what is awkward about it. Which button is hard to find? Where does a visitor stop? What becomes cramped on a phone? Those observations will be useful to today's model and to whatever replaces it.

If the old code is messy, I do not have to ask AI to rebuild the entire project at once. I can choose a small area, explain the requirement and walk through the result myself. An afternoon spent confirming that one problem is fixed feels better than an afternoon spent refreshing rumours.

I am looking forward to GPT‑7. That is exactly why I would like to meet it with a project that has moved forward, rather than a pile of abandoned ideas whose original purpose I have to remember again.

A new tool also takes time to get to know. Which model should I choose? What can I hand over? Which results still need checking? Those answers come from using it. Even a launch today would not mean that everything works itself out tomorrow.

For now, my arrangement is simple: keep reading the news and keep doing the work. When the current model genuinely cannot handle something, I can save that example and try it again with the next one. Then I will have a way to see whether it has improved.

GPT‑7 can arrive a little later. I would still like to finish today's work.