I use ChatGPT-5 Thinking a lot for day-to-day work. I prefer the response quality over speed, so I always pick the longer-thinking model. It takes ~1–5 minutes (often 1–2) to respond, but those short waits are getting increasingly distracting.
I find myself doing something else, and it eventually takes more than the thinking time. Such as right now, it takes me 5 minutes to write this post, and ChatGPT responded like 3 minutes ago.
Does anyone else have the same problem? What do you do during these gaps? :D
I often use that time to spec out a future task. Either by going through Github issues, doing some research and adding details, or by spinning up another codex/claude session to create a detailed design document for a future task and iterating on that. So one agent is coding while another is helping me to spec out future work. So when the coding agent is done I can immediately start on the next task with a proper spec, reducing margin for error.
I stare are the prompt impatiently, thinking to myself how annoying the wait is and how it used to be faster.
It really depends, sometimes I am just doing other tasks that aren't directly coding related. But often I also start with a plan of what I will prompt the LLM with, as I have a rough idea of the work that I want to get done in a few hours. So I start with the much larger tasks, and then while the LLM is thinking, I am kind of planning through my approach with the next task(s). Sometimes I am also wireframing, or drawing things which need to be improved on-top of a screenshot for the next round of UI/UX improvements.
Anki. Or pushups.
Anki and AnkiDroid are perfect for filling in those few minutes throughout the day. And if you're in a private environment, a small set of pushups a few times a day keep you awake and make you feel great.
This is a timely and interesting topic, I was just thinking about this the other day. What I've been doing lately is context switching between two different tasks, which is not ideal and you do forget what's going on between the two. The alternative seems to be to just be idle. I'm trying to prepare myself for a future where agents are much more autonomous, and I can give them tasks and just have them go run with them. This seems like the path to that, but if anyone has better ideas, I'm all for hearing them.
I have 4 computers setup with different monitors with different projects at different states and when I need to wait for one (llm or build deployment or strategy game) I flip to the next. Each with a separate keyboard. Mouses are on different levels.. two main keyboards one above for straight setup, laptop to the right and a fourth keyboard knee level that goes into the lap.
You only get so much time in a day do everything at once.
When using codex-cli / Claude Code, I look for opportunities to improve either the AGENTS.md file based on the path that the LLM is taking or to improve repo structure, docs, var names, etc that might help make things more clear in the future. Since this is still somewhat task related, I can usually stay focused.
I haven't had much luck doing two completely different tasks at the same time. At net I think it slows me down due to the context switching.
Aider can generate sounds and notifications after it complete its output. Not that it makes me more disciplined. Still waiting for LLMs with ultra fast output.
Use Black Screen app (https://blackscreen1.com/) to stare at the backness of all your monitors. Or look at some random photos with it. :-)
I cut and paste the chats into other GPTs to see what they have to say.
If I can context switch between two prompts I will.
If the task is hard and I can't easily break context I'll do exercises, chores, or comment on HN.
Obviously I move the mouse pointer so as not to freeze the machine and speed up the process.
You're looking at it chief.
I have multiple things going on at once. Doing like 3-5 things at a time.
Have developers forgotten what long compile times could be like?
I'm always prompting. I let Gemini calendar my time (Google Calendar, of course) so that there are zero gaps where unwanted human thought could sneak in. I use an agent manager called Pelican to check in with all of my agents every second and have them tell me what I should be working on next using meta-agents, which are agents running tools in a loop to run tools in a loop. It's super effective!