I’ve started asking AI to draw out the design before writing code. It’s much easier to find a bad approach in a diagram than after it dumped hundreds of lines of code. It’s also a great way to learn: sometimes it suggests an approach I never thought of, and discussing the tradeoffs helps me understand it. As long as I’m actively involved in the design, it feels more like collaborating with someone than outsourcing my thinking.
You got a point. I have tendency to 'talk it through' with gpt beforehand and it prepares initial prompt based on that conversation. It is only lately that it provides pseudo-design notes on its w/o asking.
I try to sketch out my own component design & architecture ahead of time. Then I ask AI to generate code one class at a time, or even one method at a time. That makes code review much easier for me. It also allows me to understand the code, so that when issues come up in production, I know what to look at and how the code actually works.
I always have two projects going at once, one for Claude and one for me. I'll set Claude off on a long autonomous turn, move that window to its own monitor so i can check on it as needed, and I get to work on my project. It's kinda like having a colleague sitting next to you who is working on a separate project. You can talk about each other's projects and ask for help or a fresh set of eyes when necessary. There are actually several projects that i won't let Claude touch simply because I don't want any drift in my own understanding of what has been done.
No, it hasn't been an issue. Claude will stay within its project boundaries and is generally very cautious, maybe sometimes even overly-cautious, about anything it does that might effect other projects. It knows about all the projects on my dev machine and is good at staying in its lane.
I’ve started asking AI to draw out the design before writing code. It’s much easier to find a bad approach in a diagram than after it dumped hundreds of lines of code. It’s also a great way to learn: sometimes it suggests an approach I never thought of, and discussing the tradeoffs helps me understand it. As long as I’m actively involved in the design, it feels more like collaborating with someone than outsourcing my thinking.
You got a point. I have tendency to 'talk it through' with gpt beforehand and it prepares initial prompt based on that conversation. It is only lately that it provides pseudo-design notes on its w/o asking.
I try to sketch out my own component design & architecture ahead of time. Then I ask AI to generate code one class at a time, or even one method at a time. That makes code review much easier for me. It also allows me to understand the code, so that when issues come up in production, I know what to look at and how the code actually works.
Ok. That is actually pretty hardcore in the sense that it almost forgoes the speed benefit, but I get what you are going after.
I always have two projects going at once, one for Claude and one for me. I'll set Claude off on a long autonomous turn, move that window to its own monitor so i can check on it as needed, and I get to work on my project. It's kinda like having a colleague sitting next to you who is working on a separate project. You can talk about each other's projects and ask for help or a fresh set of eyes when necessary. There are actually several projects that i won't let Claude touch simply because I don't want any drift in my own understanding of what has been done.
Interesting. Do you separate environments in any way ( VM, toolbox or anything like that ) to prevent contamination?
No, it hasn't been an issue. Claude will stay within its project boundaries and is generally very cautious, maybe sometimes even overly-cautious, about anything it does that might effect other projects. It knows about all the projects on my dev machine and is good at staying in its lane.