Any of you using Claude?

bartosimpsonio

Elite Member
Executive VIP
Jr. VIP
Joined
Mar 21, 2013
Messages
24,792
Reaction score
31,569
Before anything else, read this. Read it now.


Alright, now onto my post.

The new Claude model isn't just a 0.1 version upgrade. It is ridiculous. You tell it to write a browser extension to auto browse and mine informations from a social network and it gives you the finished bot running on your browser. You tell it to do all your work, go for a walk return and it's done.

Whatever you're doing in IM you should be using Claude. It is plain simply ridiculous.

And as the text you just read has it, if you don't adapt you will perish within the next few months. This is the most important year of your digital life. Seize the opportunity.
 
And as the text you just read has it, if you don't adapt you will perish within the next few months. This is the most important year of your digital life. Seize the opportunity.
The problem is that every single man and his dog ate seizing their opportunity. There are as many opportunities being seized as slot in the casino wheel. In which will the ball drop?
 
Claude has become my daily driver for coding. It’s incredibly solid for refactoring, debugging, and working through complex logic without going off the rails. I still use other tools here and there, but for day-to-day programming, Claude has been consistently reliable for me.
 
The new Claude model isn't just a 0.1 version upgrade. It is ridiculous. You tell it to write a browser extension to auto browse and mine informations from a social network and it gives you the finished bot running on your browser. You tell it to do all your work, go for a walk return and it's done.
Sure, as long as the final product doesn't handle sensitive information, it's alright.

For production, its good to create demos and manually fix all possible bugs/exploits
 
Sure, as long as the final product doesn't handle sensitive information, it's alright.

For production, its good to create demos and manually fix all possible bugs/exploits
You just outlined how it should be done especially for the vibe coder who thinks it's all done already with a claud prompt
 
Sign me up! Which plan does that max or pro?
Pro is fine for 99% use cases. If you start getting very slow responses and run out of tokens then you may need max. I haven't needed it yet.
The problem is that every single man and his dog ate seizing their opportunity. There are as many opportunities being seized as slot in the casino wheel. In which will the ball drop?
Exactly. But if you don't, you'll be even farther behind than if you do. I'm an old fart at this game and I'm working hard to keep up with the times.
Claude has become my daily driver for coding. It’s incredibly solid for refactoring, debugging, and working through complex logic without going off the rails. I still use other tools here and there, but for day-to-day programming, Claude has been consistently reliable for me.
It's amazing for agentic tasks too. You tell it what to do and it's your worker doing things for you.
Sure, as long as the final product doesn't handle sensitive information, it's alright.

For production, its good to create demos and manually fix all possible bugs/exploits
That was my thought early on. Now there's no longer a choice, you gotta dive in head first. Either you train your own AI (be prepared to invest mid 5 figs at least) or buy a subscription
You just outlined how it should be done especially for the vibe coder who thinks it's all done already with a claud prompt
Now it's going beyond prompts, you actually create an employee that you store data for, it has brains, it works for you. You tell it what you want, how you want it, then go out and let it work.
 
That was my thought early on. Now there's no longer a choice, you gotta dive in head first. Either you train your own AI (be prepared to invest mid 5 figs at least) or buy a subscription
you can also invest in your own rig and handle offline AI models, but if you want convenience, then just buying a susbcription is miles easier
 
Pro is fine for 99% use cases. If you start getting very slow responses and run out of tokens then you may need max. I haven't needed it yet.
Do I need pro account or do I need api access? Let's say for creating a browser bot.
 
Show us an app that you created with only one prompt. Fully functional.

Yes, AI is very nice in helping with coding. But when you need to actually ship to production and have a useful app, you need a lot of manual work. Don't sell the dream that with just one prompt you are done.
 
I built my adult tube site using Claude Code and Opus 4.5. It was functional but I ran into some pretty serious bugs and it's still helpful to think. I'm thinking a large part of why I run into bugs is that these tools don't maintain context between chats.

If I have a deploy script that only updates a Docker container if it's dependencies change and I modify a dependency, I'm at the mercy of AI figuring out to update the deploy script also.
 
Show us an app that you created with only one prompt. Fully functional.

Yes, AI is very nice in helping with coding. But when you need to actually ship to production and have a useful app, you need a lot of manual work. Don't sell the dream that with just one prompt you are done.

On vscode I simply typed in a prompt like "generate a chrome extension for me that does A, B, C and D " and so on. It generated the whole thing in a couple minutes then provided me detailed instructions on how to use the extension locally with automatic reloading on changes. I vibe coded a chrome extension that can scrape a major social media platform using my current login



I built my adult tube site using Claude Code and Opus 4.5. It was functional but I ran into some pretty serious bugs and it's still helpful to think. I'm thinking a large part of why I run into bugs is that these tools don't maintain context between chats.

They do if you do it right, I'm reading up on that as we speak, check out claude subagents and harnesses

If I have a deploy script that only updates a Docker container if it's dependencies change and I modify a dependency, I'm at the mercy of AI figuring out to update the deploy script also.

First generate one using claude, test it extensively, then after it's done tell a subagent to run it periodically. Don't blindly trust it to develop and run, develop first, check it out and test, then later automate it
 
I love it but I think in 10-15 years the world is going to be unrecognisable. I’m not sure if I should be excited or scared.
 
Before anything else, read this. Read it now.


Alright, now onto my post.

The new Claude model isn't just a 0.1 version upgrade. It is ridiculous. You tell it to write a browser extension to auto browse and mine informations from a social network and it gives you the finished bot running on your browser. You tell it to do all your work, go for a walk return and it's done.

Whatever you're doing in IM you should be using Claude. It is plain simply ridiculous.

And as the text you just read has it, if you don't adapt you will perish within the next few months. This is the most important year of your digital life. Seize the opportunity.
Expect , it won't work for any serious production site lol.

The autscraper statement is also total fucking bullshit. Lmao.

Like the browser they made, doesn't work at all, unmaintained mess, c or cpp compiler doesn't even run a hello world, call gcc wherever possible.

I work professionally as a dev with some ai tools, we embed some ai logic to the app too, actually with calls to Claude api, to generate conversations.

It hallucinates table names frequently, and that's the agentic ai, if you're not setting that up very correctly and do not hand hold it along and watch over its shoulder knowing what to do and what it is doing, you'll end up building hallucination city, even goa would be jealous of how many hallucinations are generated.

You must establish and control the database and back end, else it will very very quickly fail on you and run you in circles. It consistently destroys previously working components when adding new features etc etc. It's pretty okay, but one prompt go out to the beach and come back find a finished app? Hahaha not a fucking way in he'll. Try and see for yourselves how that goes, try it.

Not "read this bs article" , go, do just that, a reasonably complicated app with auth, roles, freemium, upgrade features, roles , tiers , payments , persistent data and deploy on a server on your domain, with secured server and database..
Ah yeah you do not know how to do that, you didn't even know these things exist, right? If you don't instruct ai to do so and double check , tripple check and test, it'll never, ever happen.

So try that, something real, something bit more sophisticated than a todo or crud app. Something with restful api, perhaps throw real time features in there with web sockets, a little Saas , don't forget all the onboarding journey logic.

I will be waiting, and you will never do anything, but if you try, come back and show the app, then we can laugh together.

You're in for a rude awakeneing
 
Expect , it won't work for any serious production site lol.

The autscraper statement is also total fucking bullshit. Lmao.

Like the browser they made, doesn't work at all, unmaintained mess, c or cpp compiler doesn't even run a hello world, call gcc wherever possible.

I work professionally as a dev with some ai tools, we embed some ai logic to the app too, actually with calls to Claude api, to generate conversations.

It hallucinates table names frequently, and that's the agentic ai, if you're not setting that up very correctly and do not hand hold it along and watch over its shoulder knowing what to do and what it is doing, you'll end up building hallucination city, even goa would be jealous of how many hallucinations are generated.

You must establish and control the database and back end, else it will very very quickly fail on you and run you in circles. It consistently destroys previously working components when adding new features etc etc. It's pretty okay, but one prompt go out to the beach and come back find a finished app? Hahaha not a fucking way in he'll. Try and see for yourselves how that goes, try it.

Not "read this bs article" , go, do just that, a reasonably complicated app with auth, roles, freemium, upgrade features, roles , tiers , payments , persistent data and deploy on a server on your domain, with secured server and database..
Ah yeah you do not know how to do that, you didn't even know these things exist, right? If you don't instruct ai to do so and double check , tripple check and test, it'll never, ever happen.

So try that, something real, something bit more sophisticated than a todo or crud app. Something with restful api, perhaps throw real time features in there with web sockets, a little Saas , don't forget all the onboarding journey logic.

I will be waiting, and you will never do anything, but if you try, come back and show the app, then we can laugh together.

You're in for a rude awakeneing
Exactly, I asked him for the app and he hasnt shown it
 
Expect , it won't work for any serious production site lol.

The autscraper statement is also total fucking bullshit. Lmao.

Like the browser they made, doesn't work at all, unmaintained mess, c or cpp compiler doesn't even run a hello world, call gcc wherever possible.

I work professionally as a dev with some ai tools, we embed some ai logic to the app too, actually with calls to Claude api, to generate conversations.

It hallucinates table names frequently, and that's the agentic ai, if you're not setting that up very correctly and do not hand hold it along and watch over its shoulder knowing what to do and what it is doing, you'll end up building hallucination city, even goa would be jealous of how many hallucinations are generated.

You must establish and control the database and back end, else it will very very quickly fail on you and run you in circles. It consistently destroys previously working components when adding new features etc etc. It's pretty okay, but one prompt go out to the beach and come back find a finished app? Hahaha not a fucking way in he'll. Try and see for yourselves how that goes, try it.

Not "read this bs article" , go, do just that, a reasonably complicated app with auth, roles, freemium, upgrade features, roles , tiers , payments , persistent data and deploy on a server on your domain, with secured server and database..
Ah yeah you do not know how to do that, you didn't even know these things exist, right? If you don't instruct ai to do so and double check , tripple check and test, it'll never, ever happen.

So try that, something real, something bit more sophisticated than a todo or crud app. Something with restful api, perhaps throw real time features in there with web sockets, a little Saas , don't forget all the onboarding journey logic.

I will be waiting, and you will never do anything, but if you try, come back and show the app, then we can laugh together.

You're in for a rude awakeneing

Exactly, I asked him for the app and he hasnt shown it

Bummer I guess you're right. Forget all about claude!
 
Exactly, I asked him for the app and he hasnt shown it
Can we have different opinions on things without being so rude about it? I realize AI and its ability to possibly replace people at jobs doesn't sit well with people, but I think we can have an open discussion about it, as that is what makes BHW a great place in my opinion.
 
I didnt want to be rude, is just this is the 100th post like this without showing anything for it, I agree that AI has and will keep changing all jobs/roles/programming as we know, but as of now, it is not as simple as a single prompt to create full fledged apps
 
Back
Top