I'm not gonna say I made this post within hours of learning how to do this and realizing I'd been doing it wrong for two years, but … Infer from that what you will.
I was today years old when I decided to build out four features at once and thought telling them to make a new “branch” was enough. My goodness.
I created a whole certified mess. Including breaking login right before a meeting with a Beta Customer. Fixed it in the first 30 seconds of the meeting 😂, but it was not a fun day.
I’ve run many different projects side by side, I’ve worked on small features and bugs simultaneously, but this is the first time I tried to do multiple massive features at once. Lesson learned, this post is timely.
I TOO was in the middle of cleaning up a HUGE mess when this post landed. I kept overwriting and colliding with my code, redoing work. I chalk it up to the learning experience, I think I learn better by spending hours fixing mistakes :) (side note: Claude actually seems frustrated at me for a change I made without asking first, I felt like I had let it down)
I would also recommend adding explicit instructions for each agent to review the development process, including the git instructions, before starting any new code.
Since we are talking about multi-agent systems here, the instructions are saved in a central hub/brain/databse and the instruction "Review git instructions" are in the main claud.MD files to keep these lightweight. I have found that without this, Claude can and will skim and completely bypass instructions
Why you gotta call me out like this after two days cleaning up my “own goal?” 🤦🏻♀️🤡
I'm not gonna say I made this post within hours of learning how to do this and realizing I'd been doing it wrong for two years, but … Infer from that what you will.
❤️ the integrity of admitting that is awesome
It's my MO haha- even made it my profile description
Oh phew - I was only a few months in... I don't feel as bad :)
Missing Princess Donut on the thumbnails ;-)
Omg you’re right!
I was today years old when I decided to build out four features at once and thought telling them to make a new “branch” was enough. My goodness.
I created a whole certified mess. Including breaking login right before a meeting with a Beta Customer. Fixed it in the first 30 seconds of the meeting 😂, but it was not a fun day.
I’ve run many different projects side by side, I’ve worked on small features and bugs simultaneously, but this is the first time I tried to do multiple massive features at once. Lesson learned, this post is timely.
Thank you!
Made the post I wish I’d seen when I started.
And we arrived here almost simultaneously. 😂
I TOO was in the middle of cleaning up a HUGE mess when this post landed. I kept overwriting and colliding with my code, redoing work. I chalk it up to the learning experience, I think I learn better by spending hours fixing mistakes :) (side note: Claude actually seems frustrated at me for a change I made without asking first, I felt like I had let it down)
I would also recommend adding explicit instructions for each agent to review the development process, including the git instructions, before starting any new code.
Since we are talking about multi-agent systems here, the instructions are saved in a central hub/brain/databse and the instruction "Review git instructions" are in the main claud.MD files to keep these lightweight. I have found that without this, Claude can and will skim and completely bypass instructions
Thank you. You’re better than Tylenol.