Vibecoding: a beginner's guide

Fulcrum29 said:
https://twitter.com/x/status/2017082134717223008

This might be a Rabbit device that sells, could be interesting. They have since given attention to the R1, and it has received another big update. I only hope that you can run a localised model (either on device or network) without a subscription.
Click to expand...
I can’t resist a good signup form:
IMG_9008.jpeg

Off-topic rant: who creates a signup form where the text box doesn’t disappear when the signup button is clicked???***

***Nobody say AI, I’m warning you.
 
I tried vibe coding + mixed with hand coding and after the AI always said, you're right! let me fix that -- git reset --hard i tossed that ****, even after I gave it explicit commands it was not allowed to run the fker still managed to run it.

My github graph - no vibe coding:

1770186659330.png

Openclaw developer slop ai graph:

1770186838253.png

There is literally zero ****ing chance that guy has reviewed 89K commits, ZERO.
 
semaphore said:
I tried vibe coding + mixed with hand coding and after the AI always said, you're right! let me fix that -- git reset --hard i tossed that ****, even after I gave it explicit commands it was not allowed to run the fker still managed to run it.

My github graph - no vibe coding:

View attachment 1883728

Openclaw developer slop ai graph:

View attachment 1883732

There is literally zero ****ing chance that guy has reviewed 89K commits, ZERO.
Click to expand...
Yeah I believe it makes a big difference which AI model and version of that model you select: some are supposed to be better than others.

I’ve personally found that Gemini is responsive but haven’t gotten into anything too in-depth yet, I’ve been focused on setting up the environment I want to use and the workflow to backup files I’ll be editing with AI.

The issue I ran into Cursor: my free trial ended abruptly and I’m now being prompted to upgrade to the premium version. Of course.

Now I’m trying to re-adjust to get Gemini to look at code I want it to edit and then paste that into VS Code … at least I’ve got Gemini advanced thrown into my Google One subscription (which I need for data storage).
 
Foxhound5366 said:
Yeah I believe it makes a big difference which AI model and version of that model you select: some are supposed to be better than others.
Click to expand...
I've tried them all, they all have their pro's and cons.

Foxhound5366 said:
I’ve personally found that Gemini is responsive but haven’t gotten into anything too in-depth yet, I’ve been focused on setting up the environment I want to use and the workflow to backup files I’ll be editing with AI.
Click to expand...
Gemini is pure trash, out of all the models it has performed the worst, for me at least.

Foxhound5366 said:
The issue I ran into Cursor: my free trial ended abruptly and I’m now being prompted to upgrade to the premium version. Of course.
Click to expand...
Cursor is ****, the developers are sketchy as **** for that platform. Just use raw cli, cursor is redundant these days.

Foxhound5366 said:
Now I’m trying to re-adjust to get Gemini to look at code I want it to edit and then paste that into VS Code … at least I’ve got Gemini advanced thrown into my Google One subscription (which I need for data storage).
Click to expand...
Just use anti gravity.
 
semaphore said:
I've tried them all, they all have their pro's and cons.


Gemini is pure trash, out of all the models it has performed the worst, for me at least.


Cursor is ****, the developers are sketchy as **** for that platform. Just use raw cli, cursor is redundant these days.


Just use anti gravity.
Click to expand...
Yeah I don’t have complex needs, so quite willing to accept that Claude (which I’ve heard praised by other developers) might be better for more complex coding. Again that requires a subscription for the better models.

Why do you recommend Anti Gravity over VS Code? I’ve never worked on Anti Gravity before, but I don’t care much about which IDE I use … I just treat them like glorified versions of notepad.
 
Foxhound5366 said:
Yeah I don’t have complex needs, so quite willing to accept that Claude (which I’ve heard praised by other developers) might be better for more complex coding. Again that requires a subscription for the better models.

Why do you recommend Anti Gravity over VS Code? I’ve never worked on Anti Gravity before, but I don’t care much about which IDE I use … I just treat them like glorified versions of notepad.
Click to expand...
Because antigravity has gemini baked in.
 
Foxhound5366 said:
Yay! I’ll definitely give that a look then.
Click to expand...
FYI Gemini has a cli as well that you can run in the terminal.

Ironically out of all the models I’ve tested the only one that didn’t produce abhorrent slop was gpt 5.2. I only tested it because I was given a month of pro for free.

Claude has become pretty terrible of late, coupled with the horrific rare limits.

Gemini has great context, but that’s also bad as it tends to hallucinate more and fk things up.

Ultimately you’ll find that using Gemini for planning, Claude for coding and gpt for review would be a good approach, but costly.
 
Right now if you are not using Claude with opus 4.5 you are living what last year's AI could do. If you are in anyway serouis about using AI to assist with code or to fix old code. There others are simply not there yet.

I have the same project exact wording to both. Claude finished it in about 15 minutes gemini spun up a testdocker and ate it's own tokens till I had none left because it went into some loop.

Spend $20 get claude with opus 4.5 and re write what you wanted to see the difference.
 
Nevar said:
Right now if you are not using Claude with opus 4.5 you are living what last year's AI could do. If you are in anyway serouis about using AI to assist with code or to fix old code. There others are simply not there yet.

I have the same project exact wording to both. Claude finished it in about 15 minutes gemini spun up a testdocker and ate it's own tokens till I had none left because it went into some loop.

Spend $20 get claude with opus 4.5 and re write what you wanted to see the difference.
Click to expand...
Nah, Opus is over hyped. It’s just as crap as sonnet and makes the same mistakes, not to mention you’ll blow your limit before you’ve even blinked. I had Max sub for a few months and tested it extensively.
 
semaphore said:
Nah, Opus is over hyped. It’s just as crap as sonnet and makes the same mistakes, not to mention you’ll blow your limit before you’ve even blinked. I had Max sub for a few months and tested it extensively.
Click to expand...
When Opus (and sonnet) actually comes through its good, but they tend to go ****edup from time to time and nowadays more than often. Pretty much same for claude interactive - however, that said, claude interactive can be nice to do some initial analysis paralysis.
 
Herr der Verboten said:
When Opus (and sonnet) actually comes through its good, but they tend to go ****edup from time to time and nowadays more than often. Pretty much same for claude interactive - however, that said, claude interactive can be nice to do some initial analysis paralysis.
Click to expand...
Yeah, when I tried gpt 5.2 I gave it the same problem statement. I was very surprised by its output, Gemini wasn’t far behind, Claude couldn’t even fix the issue it created.
 
semaphore said:
Yeah, when I tried gpt 5.2 I gave it the same problem statement. I was very surprised by its output, Gemini wasn’t far behind, Claude couldn’t even fix the issue it created.
Click to expand...
The downhill imho started somewhere at the end of the year. Coming back to claude in Jan it was already worse.

Chatgpt somehow seems to have improved and improved. I do find it better for long debugging assist whereas claude will just give up because now it is hard or require effort.

Haven't used Gemini at all. Grok, well mid last year was actually quite good before it went downhill. Nowadays grok is just google bitch.

... another thing with claude it seems they have internally added fast mode you can't switch off. In most cases it goes fast auto on you and don't even read or digest what you give it. This is very much inline with how chat and grok also handle fast: kak.
 
Herr der Verboten said:
The downhill imho started somewhere at the end of the year. Coming back to claude in Jan it was already worse.

Chatgpt somehow seems to have improved and improved. I do find it better for long debugging assist whereas claude will just give up because now it is hard or require effort.

Haven't used Gemini at all. Grok, well mid last year was actually quite good before it went downhill. Nowadays grok is just google bitch.

... another thing with claude it seems they have internally added fast mode you can't switch off. In most cases it goes fast auto on you and don't even read or digest what you give it. This is very much inline with how chat and grok also handle fast: kak.
Click to expand...
Never tried Grok, personally, Elon is a choad so don't care much for his shitty products.
 
semaphore said:
I tried vibe coding + mixed with hand coding and after the AI always said, you're right! let me fix that -- git reset --hard i tossed that ****, even after I gave it explicit commands it was not allowed to run the fker still managed to run it.

My github graph - no vibe coding:

View attachment 1883728

Openclaw developer slop ai graph:

View attachment 1883732

There is literally zero ****ing chance that guy has reviewed 89K commits, ZERO.
Click to expand...
Lol this is mine, need to vibecode harder :D
1770205726755.png
 
You must log in or register to reply here.
Back
Top
Sign up to the MyBroadband newsletter
X