I eat junk food and give my brain calories to fuel up for the next round. I find claude coding sessions to be cognitively intense! For me it will take awhile to build optimal work habits.
AndyL
Vidar
More sessions…?
Usually two or three with stepwise milestone plans or smaller single steps. So when one works I check on the output of what another session has done since the last step and test or check on it. Then write a new prompt for it. Depending on the project this keeps me busy, and Claude has to wait for me.
The notable exception being some very data intensive ones were it can take a very long time for each step. If I know it will take an hour or three I just do some other kind of work entirely, or try do it in the background while I do research and planning.
I also have this ongoing skill development, so I will ask the sessions to give feedback of the skills used during or after use, and then spend time trying to evaluate and make improvements accordingly.
Sometimes I roam around this forum 
4
sodapopcan
This is actually one of the more interesting threads on LLMs I’ve read in a while.
My experience so far as been far has been working far more exploratory and chaotic. Most of my prompts are very precise and usually don’t take very long at all.I’ve only had a couple of times where something has taken… I’m not even sure, maybe 30 minutes? But after reading Kip’s answer I want to try something like that!
In any event, I’m usually thinking, “Is this really what I want to do?” and then when it’s done I say, “Wait, no, let’s try implementing it this way” and it says, “Yes, you are very smart, that is a much better idea!” and then when it’s done I say, “Oh wait no, I tried the new way and the old way is actually better, let’s revert back to it” to which it says, “Of course! You are very smart and the old way so obviously better, I should have seen that!” I keep forgetting to enable the Caveman plugin.
Otherwise, my answer is pretty simple: YouTube, which still includes videos of people talking about how much they hate AI even though I’m fully bought in at this point 
4
Aiken00
Usually will prepare the further instruction to be typed. And review the code generated from Claude Code last session.
Sometime working on frontend code, will do some regression testing manually.
2
egeersoz
jam
DaAnalyst
arcanemachine
In the past week, I have had a couple tasks where the agent is “thinking” for 5 solid minutes on a single reply prompt. No back-and-forth with tool calls, like one single reply. I even thought that one of them timed out, so I cancelled it and re-ran the query, and the response took another 5 minutes. This was with Claude Opus 4.6.
2
mudasobwa
Creator of Cure
I have a rule “Watch the prompt response and treat no outcome for more than 30 secs as a timeout, report it and ask what to do next.”
The 5-mins thinking is a definite hint that the outcome will be not ideal, to put it gently. It’s like a junior developer constantly saying at standups “I have some progress,” meaning there is no progress whatsoever.
1
arcanemachine
I hear where you’re coming from, but my experience does not (necessarily) match that. When this is happening, it’s usually comes after a pretty demanding prompt, and correlates roughly with how much “thinking” would be expected.
I will say that Claude has some bad days, though. Due to Anthropic’s compute constraint, they seem to be (tinfoil hat time…) running quantized models during peak usage hours, which would probably also correlate with long response times.
3