Clear the session after 300k tokens and force your workflow to adapt to it. The rabbit holes come when you’re at context rot and blindly following Claude debugging some problem that you shouldn’t even be trying to solve.
Other than that, set an alarm when you want to stop working. When it rings, physically unplug your work laptop and move it somewhere out of sight.
Is this work related or addiction related? You are repeatedly pulling a slot machine in hopes for a jackpot of working usable code, something which does not come up just often enough to require another pull of the conversation turn.
Claude does not have a sense of time. What is stopping you from not talking to it until the next day?
Vibe coding is addictive for the same reasons slot machines are.
Sometimes the output is great, sometimes not, but it always feels like a change to the prompt will get you closer to the prize. Just like pulling the handle on a slot machine gets you one more chance at a jackpot.
Today I opened Twitter. I had 1.5K followers back in the day. I think it had similar dynamics: an immensely powerful tool, but users had no discipline and those running it had really bad incentives.
I know this feeling. Before AI you had to evaluate an idea, plan its inplementation, judge if it is worth it or even doable. Now, you just prompt and see if you keep the output and think of the next thing to do. This can feel addictive. When i get lost in this i try to step away and think about what the bigger goal is and whether i am just distracting myself.
I guess this helps me not get drained. But there is something which also helps tremendously: my coworkers do not work on the same things as i do. if they did, i would feel a big pressure to finish first.
Are you competing with anyone?
Pure vibe-coding isn’t that satisfying (as you’ve discovered) the problem is you’re constantly reviewing its huge output, and it probably goes wild in directions making assumptions.
It’s much more fulfilling to give the LLM guardrails by engineering the solution.
This means coming up with the high level architecture, dependencies, breaking down the system into deliverable chunks, and implementing those chunks. The LLM can help in all those steps of the process. And it will be much easier to review and keep an understanding of the code.
Okay, I think the first problem is calling it an addiction because it is a similar feeling if you want to keep coding in machine code instead of moving higher in the hierarchy and start working with C++. It is an extremely handy tool to be able to produce some work, so you shouldn't take it off your belt. However, I will talk about Gemini addiction, where I can't even decide which color of shoes I should buy. I'm asking it to pick for me, which is terrible. So there is that.
Interesting points, thank you. To explore your idea: what if I was addicted to using C++? I don't mind using the new, high-level tool all day (be it Claude Code or C++), but it's continuing to vibe code with it for several hours AFTER the workday is OVER that is bad. To your point, I guess when one learns a more powerful programming language, there is an initial period of enthusiasm and then it might wear off and become more mundane. Maybe I just haven't reached that with Claude Code yet. It feels like Claude Code has kept improving quickly, which may have kept this workaholic binge cycle going for longer than usual. Regarding the non-code decision stuff - yes, it's definitely related. I have always been someone who researches, seeks reassurance, and wants expertise and rules to apply to my decisions, instead of making them from the gut. I suppose this "Ask HN" post is an example of that, in a way - I'm crowdsourcing solutions to a personal problem. But it feels different (and better) to seek opinions from human beings - even though they will be represented to me as text, just like "opinion-like" strings of words from Claude... when I was younger I would describe myself as addicted to online forums. So I guess the artificiality of the intelligence isn't the problem. It's just the speed and always-on connectedness - similar to the internet itself.
I probably would try to delegate to Claude or any other agent quite long and hard task and then log off for the day living my offline life. Thus definitely remove my "guilt" for not being productive and let the useful work being produced. Not ideal,I know, but this is what I was doing lately
Clear the session after 300k tokens and force your workflow to adapt to it. The rabbit holes come when you’re at context rot and blindly following Claude debugging some problem that you shouldn’t even be trying to solve.
Other than that, set an alarm when you want to stop working. When it rings, physically unplug your work laptop and move it somewhere out of sight.
Is this work related or addiction related? You are repeatedly pulling a slot machine in hopes for a jackpot of working usable code, something which does not come up just often enough to require another pull of the conversation turn.
Claude does not have a sense of time. What is stopping you from not talking to it until the next day?
Vibe coding is addictive for the same reasons slot machines are.
Sometimes the output is great, sometimes not, but it always feels like a change to the prompt will get you closer to the prize. Just like pulling the handle on a slot machine gets you one more chance at a jackpot.
Today I opened Twitter. I had 1.5K followers back in the day. I think it had similar dynamics: an immensely powerful tool, but users had no discipline and those running it had really bad incentives.
I know this feeling. Before AI you had to evaluate an idea, plan its inplementation, judge if it is worth it or even doable. Now, you just prompt and see if you keep the output and think of the next thing to do. This can feel addictive. When i get lost in this i try to step away and think about what the bigger goal is and whether i am just distracting myself. I guess this helps me not get drained. But there is something which also helps tremendously: my coworkers do not work on the same things as i do. if they did, i would feel a big pressure to finish first. Are you competing with anyone?
Pure vibe-coding isn’t that satisfying (as you’ve discovered) the problem is you’re constantly reviewing its huge output, and it probably goes wild in directions making assumptions.
It’s much more fulfilling to give the LLM guardrails by engineering the solution.
This means coming up with the high level architecture, dependencies, breaking down the system into deliverable chunks, and implementing those chunks. The LLM can help in all those steps of the process. And it will be much easier to review and keep an understanding of the code.
Okay, I think the first problem is calling it an addiction because it is a similar feeling if you want to keep coding in machine code instead of moving higher in the hierarchy and start working with C++. It is an extremely handy tool to be able to produce some work, so you shouldn't take it off your belt. However, I will talk about Gemini addiction, where I can't even decide which color of shoes I should buy. I'm asking it to pick for me, which is terrible. So there is that.
Interesting points, thank you. To explore your idea: what if I was addicted to using C++? I don't mind using the new, high-level tool all day (be it Claude Code or C++), but it's continuing to vibe code with it for several hours AFTER the workday is OVER that is bad. To your point, I guess when one learns a more powerful programming language, there is an initial period of enthusiasm and then it might wear off and become more mundane. Maybe I just haven't reached that with Claude Code yet. It feels like Claude Code has kept improving quickly, which may have kept this workaholic binge cycle going for longer than usual. Regarding the non-code decision stuff - yes, it's definitely related. I have always been someone who researches, seeks reassurance, and wants expertise and rules to apply to my decisions, instead of making them from the gut. I suppose this "Ask HN" post is an example of that, in a way - I'm crowdsourcing solutions to a personal problem. But it feels different (and better) to seek opinions from human beings - even though they will be represented to me as text, just like "opinion-like" strings of words from Claude... when I was younger I would describe myself as addicted to online forums. So I guess the artificiality of the intelligence isn't the problem. It's just the speed and always-on connectedness - similar to the internet itself.
I probably would try to delegate to Claude or any other agent quite long and hard task and then log off for the day living my offline life. Thus definitely remove my "guilt" for not being productive and let the useful work being produced. Not ideal,I know, but this is what I was doing lately
If you could that reliably, you wouldn’t have a job
"this is a well-structured prompt. i'll pull up the relevant research and return with suggestions."