The waiting sucks! It gets me out the zone too. I noticed it helps me to use the fast Claude models when not doing complex or planning stuff, as it’s soo much faster. It got better with 4.x luckily.
I recently largely vibe coded a product, too. The 30 second downtime is just enough to get distracted. I try to force myself to wait. When I realize a task is taking even longer than that, I try to focus on exactly that hard part you identified - looking on reddit for others with a need for the thing I'm working on. I'll try to spend the time figuring out where and how else could I get the thing I'm working on into the right people's hands. I usually build for myself first (most recently a git TUI), and then try and find other people who might benefit. Now I'm kinda sorting doing both at the same time. The big problem is walking away from the computer entirely. If I browse reddit for a minute and come back to claude, the context switching is rough, but if I go for a walk, come back and find a message from a client, move the claude tmux session to the background and forget about it for a day, when I do remember that I was doing something, I have a bad mental habit of procrastinating. When the feature to implement is new, whether AI's going to do it or I am, I want to sit down and get at it. When the feature was started a while ago, and I know I'll need to sit down and re-up on context, my head wants to put it off. I try to avoid letting things get into that place (maybe I need a notification about my backgrounded tmux sessions at the end of each day... I hadn't thought of that until now). And when things do get into that place, it helps to remind myself that I don't _have_ to re-up on the context. I can just start over. If you're working in small iterative chunks with AI there's not much lost other than tokens if you just start over. That "you can just start over" mind shift helps a bit with my "oh no, I left something mid development and I don't want to pick it back up" brain baggage. Usually that gets me back at the computer and looking at it - and usually I don't have too much trouble picking things back up where I left them.
"it helps to remind myself that I don't _have_ to re-up on the context"
I think that's the key. We are still so used to coding taking so long that we still focus on optimizing every second. But when you can spin up new code so quickly, its less about keeping the AI "busy" and more about planning to build the RIGHT code.
Great article! Yeah the waiting is rough, especially when you have a 20 stories of task running in a ralph loop. And switching between projects works sometimes but on a long enough running task I almost forget where I was heading with those tasks. It's definitely something I think is going to take time to figure out.
It's wild… I'm moving SO MUCH faster than I was before. And yes, I still feel the need to optimize every second of down time. Like, just slow down brain sheesh!
Thanks for this post, very interesting. Glad to read that I'm not alone being bored about the waiting. More than a year of Cursor usage, I'm kinda see some serious downside to this amazing productivity: it erodes my focus. This waiting have something to do with it imho, because it urge me to do something else, and it often (always?) doesn't have to do with code. I can't focus on code as I focused back in the day, struggling on bugs with tens of Stack Overflow tabs. Did you find a balance on what to code "by hand", and what to do with AI?
The loss of focus is real. And something I don't have an answer to. I think the big next step for me will be finding flow state while working with AI. It happened so often when handwriting code but is now rare to nonexistent.
The waiting sucks! It gets me out the zone too. I noticed it helps me to use the fast Claude models when not doing complex or planning stuff, as it’s soo much faster. It got better with 4.x luckily.
I recently largely vibe coded a product, too. The 30 second downtime is just enough to get distracted. I try to force myself to wait. When I realize a task is taking even longer than that, I try to focus on exactly that hard part you identified - looking on reddit for others with a need for the thing I'm working on. I'll try to spend the time figuring out where and how else could I get the thing I'm working on into the right people's hands. I usually build for myself first (most recently a git TUI), and then try and find other people who might benefit. Now I'm kinda sorting doing both at the same time. The big problem is walking away from the computer entirely. If I browse reddit for a minute and come back to claude, the context switching is rough, but if I go for a walk, come back and find a message from a client, move the claude tmux session to the background and forget about it for a day, when I do remember that I was doing something, I have a bad mental habit of procrastinating. When the feature to implement is new, whether AI's going to do it or I am, I want to sit down and get at it. When the feature was started a while ago, and I know I'll need to sit down and re-up on context, my head wants to put it off. I try to avoid letting things get into that place (maybe I need a notification about my backgrounded tmux sessions at the end of each day... I hadn't thought of that until now). And when things do get into that place, it helps to remind myself that I don't _have_ to re-up on the context. I can just start over. If you're working in small iterative chunks with AI there's not much lost other than tokens if you just start over. That "you can just start over" mind shift helps a bit with my "oh no, I left something mid development and I don't want to pick it back up" brain baggage. Usually that gets me back at the computer and looking at it - and usually I don't have too much trouble picking things back up where I left them.
"it helps to remind myself that I don't _have_ to re-up on the context"
I think that's the key. We are still so used to coding taking so long that we still focus on optimizing every second. But when you can spin up new code so quickly, its less about keeping the AI "busy" and more about planning to build the RIGHT code.
Great article! Yeah the waiting is rough, especially when you have a 20 stories of task running in a ralph loop. And switching between projects works sometimes but on a long enough running task I almost forget where I was heading with those tasks. It's definitely something I think is going to take time to figure out.
It's wild… I'm moving SO MUCH faster than I was before. And yes, I still feel the need to optimize every second of down time. Like, just slow down brain sheesh!
Good article! I haven’t figured out the downtime part of this either. My problem has been YouTube… 😳
Well at least you can pretend it's productive by learning from YouTube! 😂
Couldn't agree more. The way AI handles the boilerplate is just astonishing and a little terrifiyng, isn't it? So fasinating.
So true. I've even seen it be successful in non-trivial implementations, like user authentication or payment processing!
Thanks for this post, very interesting. Glad to read that I'm not alone being bored about the waiting. More than a year of Cursor usage, I'm kinda see some serious downside to this amazing productivity: it erodes my focus. This waiting have something to do with it imho, because it urge me to do something else, and it often (always?) doesn't have to do with code. I can't focus on code as I focused back in the day, struggling on bugs with tens of Stack Overflow tabs. Did you find a balance on what to code "by hand", and what to do with AI?
The loss of focus is real. And something I don't have an answer to. I think the big next step for me will be finding flow state while working with AI. It happened so often when handwriting code but is now rare to nonexistent.