Discord: https://discord.gg/J7qsYmDg
When I see people complain about the issues they have with AI I know exactly what they mean because I ran into all of those headaches as well. However, using good prompts and project management literally makes things 1000x better and less frustrating. I get no hallucinations, no remodeling of structure without asking, no destroying code because it gets too complicated, etc.
I don't want to go over every single detail because a lot you can read about or figure out easily enough but I will tell you how I, a non coding idiot, am able to get anything I want out of AI quickly and effectively.
First things first,
Projects on Claude Web
Claude allows you to create a project for a specific thing you are developing and save all of your scripts there as they are generated for easy recall.
You can also give instructions for the overall project so you don't have to keep repeating yourself.
I don't really fill these out with extreme details (why later) but I do give some rules that matter for me. As a simple example I absolutely loathe using cmd prompt to browse to folders and execute scripts. I also don't like editing code snippets and I hate when scripts crash over something minor then I have to chase down the bug.
So my instructions are usually very simple like
Always make my python scripts executable by double click.
Always write error / status logs on scripts unless I say otherwise.
Always give me the full rewritten code instead of telling me to replace or edit functions.
These instructions alone have saved me so much busy time and annoyances it's crazy.
General Tips
I have seen so many people on these forums and IRL say that AI gets terrible when your project gets too big. I know, I've seen it, but you can avoid it completely.
For the last massive thing I built I put the project instructions and in the initial prompt for Claude to modularize everything and make it organized AF with easy to decipher folder names. I am super specific in my first prompt but the result is a massive codebase that has things like my image processor in a single small file. It's connected to 7-15 other scripts that are also modularized and task specific which makes changing things a breeze. One of my workflows is grab data from site a in bulk with a crawler and database it, crawl site b to pull in additional data, crawl site c to pull in even more data, use script d to standardize one part of the data based on functions and script e to calculate my prices based on set rules and script f to generate text files with all of that data in a set format and script g to adjust that content for multiple sites with their own unique needs and script h to take all of the images and overlay a watermark for each unique site, upload those to a host via API and pull back in the watermarked urls for production use and script i to generate data files for each website or implementation. For instance I could have an import file made for WordPress or another CMS or both. It can go on and on but each script is it's own thing, not a wall of code that I struggle with later. If I need to edit a crawler or data layout it's a single small file and Claude can fix it in seconds.
Just to visualize my folder structure might be
Main > Crawlers > Site 1, Site 2
Main > Scripts > image-processor, input-data-file-generator, import-generator
Then in Configs I would have a file of my image host info like api key, and a site-specifics file with all of the dynamic things that would change per website so the other scripts can easily pull in correct unique info per site and I can add more with an extra line of variables.
Run Things Local
I've had so many people ask me what else I'm hooking into to work with databases, code, etc. and the answer is NOTHING. I run python and databases locally. If building a website I build locally then after I'm happy ask Claude to convert it to whatever code / format / specifications are needed to host somewhere.
Always Make A Startup Launcher
Super easy to end up with tons of scripts, backup scripts, etc. which turns into a complicated mess. To avoid this I always have Claude write me a startup launcher that when run shows all of my correct important scripts in the order I normally use them and I normally name it 000-start-here.py in root so it's right at the top. Open this, press 1 to do part 1, 2 to continue to next, etc. If you have a very standard flow you can make a script to execute things in order without prompts so it just does what you want with no input.
Safeguards For Oddities
When I make scripts or run into weird quirks of something I always give myself an out. Quick example is that when crawling thousands of pages on one site from time to time it will run into bad pages or malformed data. Instead of skipping all of these I often have a "if this happens prompt me for ____" built in. One site lands on good page or a multi result page and on that I made the trigger for "done" to be when it got to a standardized part of the right page. If the bad page is encountered I can tell it to save all for later and worry then or simple stop processing until I answer a prompt which in my case is either click on right item from the bad page or hit S for skip.
Visual Viewer
This one is magical. Every single time I'm working with anything I ask for a visual viewer that will show me in a pretty format what exactly I have in my database. Say you're fixing titles for images my viewer would show the image and title and anything else important so I can see that it's good on a page with 100 or 1000 results. I also make this viewer interact with the database and in the quickest way possible so I can click BAD IMAGE and it will tag that entry with that and it disappears from view (and into the bad image view.) Now I can work on those later all at once. You can even have a replace image, edit text or add data on the viewer to quickly go through these things.
Request Entire Project Up Front!
Claude is way better at making a functioning site or application if it knows what you want from the start. The last attempt I told it to make me a linktree clone with all of my specific changes, anything it thought my be important and in a way that adding more features later would be easy.
It gave me back a complete and working website with 90% of the features I'd want, several awesome additions I didn't ask for and cleaning up the remainder was a breeze. I said I'd eventually want an affiliate program and it built the whole thing complete with stats tracking, referrals and very pretty charts in the affiliate login. It will give you just the core files until you say good and what about x but since it designed it with all in mind those all work amazingly.
Ask For A Project Overview
You know how annoying it is when a chat fills up which is why I always ask for a project overview file that lists all my paths, scripts created, to do list, etc. and update with new information on each change.
Claude Code Specifically
Give Claude access only to project folders and if modularized it will find it's way around easily.
Include the project overview / flow file which it will save as MD so if you restart later you can say "read the MD to pick up where we left off."
Tell it to make backups when x happens. Like I have a directive in project instructions to always back up my database when using scripts to directly change things in them in case I don't like the result or something happens.
Tell it to update your instructions file after any major change and include relevant notes.
Tell it to put any 1 time use or temp scripts in some folder and even to delete after use. I might say I need to take all x out of my database and it will write a script, I do not want this in root or cluttering up folders I use so I have them all go into 'codework' or 'temp-scripts' so I don't even see them.
I normally run my first prompt on Claude web and then use Claude Code to work with it after. Once on Claude Code I get everything done from a single command prompt which is insane to me. It's like directing employees that run around being awesome while I do other things.
=============
That's all I can think of off the top of my head but in general I promise you that anything you have wanted to build in the past you can build today. I have yet to run into a single thing that I couldn't get done and with preplanning and a good initial prompt it is insanely good. I got lazy on my last attempt and the mess / organizing the mess took up 20x the time so I just redid it and it's blazing fast and accurate.
If you want to yap about AI casually or ask questions I'm on Discord always. https://discord.gg/J7qsYmDg
When I see people complain about the issues they have with AI I know exactly what they mean because I ran into all of those headaches as well. However, using good prompts and project management literally makes things 1000x better and less frustrating. I get no hallucinations, no remodeling of structure without asking, no destroying code because it gets too complicated, etc.
I don't want to go over every single detail because a lot you can read about or figure out easily enough but I will tell you how I, a non coding idiot, am able to get anything I want out of AI quickly and effectively.
First things first,
Projects on Claude Web
Claude allows you to create a project for a specific thing you are developing and save all of your scripts there as they are generated for easy recall.
You can also give instructions for the overall project so you don't have to keep repeating yourself.
I don't really fill these out with extreme details (why later) but I do give some rules that matter for me. As a simple example I absolutely loathe using cmd prompt to browse to folders and execute scripts. I also don't like editing code snippets and I hate when scripts crash over something minor then I have to chase down the bug.
So my instructions are usually very simple like
Always make my python scripts executable by double click.
Always write error / status logs on scripts unless I say otherwise.
Always give me the full rewritten code instead of telling me to replace or edit functions.
These instructions alone have saved me so much busy time and annoyances it's crazy.
General Tips
I have seen so many people on these forums and IRL say that AI gets terrible when your project gets too big. I know, I've seen it, but you can avoid it completely.
For the last massive thing I built I put the project instructions and in the initial prompt for Claude to modularize everything and make it organized AF with easy to decipher folder names. I am super specific in my first prompt but the result is a massive codebase that has things like my image processor in a single small file. It's connected to 7-15 other scripts that are also modularized and task specific which makes changing things a breeze. One of my workflows is grab data from site a in bulk with a crawler and database it, crawl site b to pull in additional data, crawl site c to pull in even more data, use script d to standardize one part of the data based on functions and script e to calculate my prices based on set rules and script f to generate text files with all of that data in a set format and script g to adjust that content for multiple sites with their own unique needs and script h to take all of the images and overlay a watermark for each unique site, upload those to a host via API and pull back in the watermarked urls for production use and script i to generate data files for each website or implementation. For instance I could have an import file made for WordPress or another CMS or both. It can go on and on but each script is it's own thing, not a wall of code that I struggle with later. If I need to edit a crawler or data layout it's a single small file and Claude can fix it in seconds.
Just to visualize my folder structure might be
Main > Crawlers > Site 1, Site 2
Main > Scripts > image-processor, input-data-file-generator, import-generator
Then in Configs I would have a file of my image host info like api key, and a site-specifics file with all of the dynamic things that would change per website so the other scripts can easily pull in correct unique info per site and I can add more with an extra line of variables.
Run Things Local
I've had so many people ask me what else I'm hooking into to work with databases, code, etc. and the answer is NOTHING. I run python and databases locally. If building a website I build locally then after I'm happy ask Claude to convert it to whatever code / format / specifications are needed to host somewhere.
Always Make A Startup Launcher
Super easy to end up with tons of scripts, backup scripts, etc. which turns into a complicated mess. To avoid this I always have Claude write me a startup launcher that when run shows all of my correct important scripts in the order I normally use them and I normally name it 000-start-here.py in root so it's right at the top. Open this, press 1 to do part 1, 2 to continue to next, etc. If you have a very standard flow you can make a script to execute things in order without prompts so it just does what you want with no input.
Safeguards For Oddities
When I make scripts or run into weird quirks of something I always give myself an out. Quick example is that when crawling thousands of pages on one site from time to time it will run into bad pages or malformed data. Instead of skipping all of these I often have a "if this happens prompt me for ____" built in. One site lands on good page or a multi result page and on that I made the trigger for "done" to be when it got to a standardized part of the right page. If the bad page is encountered I can tell it to save all for later and worry then or simple stop processing until I answer a prompt which in my case is either click on right item from the bad page or hit S for skip.
Visual Viewer
This one is magical. Every single time I'm working with anything I ask for a visual viewer that will show me in a pretty format what exactly I have in my database. Say you're fixing titles for images my viewer would show the image and title and anything else important so I can see that it's good on a page with 100 or 1000 results. I also make this viewer interact with the database and in the quickest way possible so I can click BAD IMAGE and it will tag that entry with that and it disappears from view (and into the bad image view.) Now I can work on those later all at once. You can even have a replace image, edit text or add data on the viewer to quickly go through these things.
Request Entire Project Up Front!
Claude is way better at making a functioning site or application if it knows what you want from the start. The last attempt I told it to make me a linktree clone with all of my specific changes, anything it thought my be important and in a way that adding more features later would be easy.
It gave me back a complete and working website with 90% of the features I'd want, several awesome additions I didn't ask for and cleaning up the remainder was a breeze. I said I'd eventually want an affiliate program and it built the whole thing complete with stats tracking, referrals and very pretty charts in the affiliate login. It will give you just the core files until you say good and what about x but since it designed it with all in mind those all work amazingly.
Ask For A Project Overview
You know how annoying it is when a chat fills up which is why I always ask for a project overview file that lists all my paths, scripts created, to do list, etc. and update with new information on each change.
Claude Code Specifically
Give Claude access only to project folders and if modularized it will find it's way around easily.
Include the project overview / flow file which it will save as MD so if you restart later you can say "read the MD to pick up where we left off."
Tell it to make backups when x happens. Like I have a directive in project instructions to always back up my database when using scripts to directly change things in them in case I don't like the result or something happens.
Tell it to update your instructions file after any major change and include relevant notes.
Tell it to put any 1 time use or temp scripts in some folder and even to delete after use. I might say I need to take all x out of my database and it will write a script, I do not want this in root or cluttering up folders I use so I have them all go into 'codework' or 'temp-scripts' so I don't even see them.
I normally run my first prompt on Claude web and then use Claude Code to work with it after. Once on Claude Code I get everything done from a single command prompt which is insane to me. It's like directing employees that run around being awesome while I do other things.
=============
That's all I can think of off the top of my head but in general I promise you that anything you have wanted to build in the past you can build today. I have yet to run into a single thing that I couldn't get done and with preplanning and a good initial prompt it is insanely good. I got lazy on my last attempt and the mess / organizing the mess took up 20x the time so I just redid it and it's blazing fast and accurate.
If you want to yap about AI casually or ask questions I'm on Discord always. https://discord.gg/J7qsYmDg
GFY Hall of Famer 
Comment