Vibe coding in a nutshell
-
But seriously, what IS vibe coding?
It's just prompt engineering for coding. Let an AI dump a bunch of code for you, debug until it no longer errors, pull request and repeat next sprint.
5% of the time, it works every time
-
Disclaimer: these tweets aren't real.
Vulnerable, Insecure, Broken Entry
-
But seriously, what IS vibe coding?
Using chatgpt and not checking its inevitably incorrect output and sending it
-
Disclaimer: these tweets aren't real.
Man, I'm getting old. I don't understand why all jokes have to be fake twitter screenshots now.
-
But seriously, what IS vibe coding?
These days it more or less explicitly refers to asking an LLM to write your code for you based on prompts.
But on a broader spectrum it is just the idea of (I forget the buzz word) Ticket Driven Development. A manager defines software based on a series of (jira, gitlab, kanban, whatever) tickets/issues and someone below them (in this case, an LLM) implements it.
Done properly? It is incredibly effective as it allows designers and "idea people" to work to their strengths and junior developers to work to theirs. The problem being that, much like when it is a junior dev under them, the person making the tickets likely has no idea what they are doing.
Which is the big problem. Someone who has been writing scripts for decades? Using chatgpt to get the syntax of a function or even to write a utility script is great. They can focus their brainpower on the harder/more fun stuff. Someone who has been writing code for, at most, a year or two? They never learn those foundations and never have a way to do anything the LLM can't (or verify if the LLM is correct).
-
It's just prompt engineering for coding. Let an AI dump a bunch of code for you, debug until it no longer errors, pull request and repeat next sprint.
5% of the time, it works every time
Is it truly a deviation from outsourcing code to a 3rd party?
-
It's just prompt engineering for coding. Let an AI dump a bunch of code for you, debug until it no longer errors, pull request and repeat next sprint.
5% of the time, it works every time
Brought to you by (us) security researchers who will happily come in and sort out your security issues later. For a very hefty hourly fee.
-
Man, I'm getting old. I don't understand why all jokes have to be fake twitter screenshots now.
It is just a format for the joke, like a web comic or video of a skit.
-
Is it truly a deviation from outsourcing code to a 3rd party?
Yeah, because at least a decent 3rd party might hand you documentation and have the sense to build something consistent or maintainable. AI has a limited context scope and frequently suffers a type of short term memory loss that results in repeated work or variations in work that confuse the end result.
-
It is just a format for the joke, like a web comic or video of a skit.
wrote on last edited by [email protected]Yes, as I've said, this is just me getting old rather than any issue with the joke format
-
Man, I'm getting old. I don't understand why all jokes have to be fake twitter screenshots now.
Initially, I had the same thought. However, given that it was clearly fake and meant to be humorous without causing any harm, I believe it's acceptable.
-
Brought to you by (us) security researchers who will happily come in and sort out your security issues later. For a very hefty hourly fee.
Deserved. These companies need to find it out the hard way.
-
But seriously, what IS vibe coding?
It's magic
-
These days it more or less explicitly refers to asking an LLM to write your code for you based on prompts.
But on a broader spectrum it is just the idea of (I forget the buzz word) Ticket Driven Development. A manager defines software based on a series of (jira, gitlab, kanban, whatever) tickets/issues and someone below them (in this case, an LLM) implements it.
Done properly? It is incredibly effective as it allows designers and "idea people" to work to their strengths and junior developers to work to theirs. The problem being that, much like when it is a junior dev under them, the person making the tickets likely has no idea what they are doing.
Which is the big problem. Someone who has been writing scripts for decades? Using chatgpt to get the syntax of a function or even to write a utility script is great. They can focus their brainpower on the harder/more fun stuff. Someone who has been writing code for, at most, a year or two? They never learn those foundations and never have a way to do anything the LLM can't (or verify if the LLM is correct).
I’ve been programming for over half my life now, I actually like vibe coding with Claude these days.
It basically gets me through the hump of “ugh this task is going to be annoying as fuck to do” which is where I personally lose most of my efficiency (I have a lot of difficulty forcing myself to do something I don’t want to).
It’s like when I had interns and I’d give them tasks. Describe the work, scope it, add some guard rails to keep it directionally okay, and send it off to get reviewed later. And that works great with modern agents.
I will say vibe coding is damn good at debugging, way better than I am, so I use it for that a lot now.
-
Disclaimer: these tweets aren't real.
Villainously, I'm billing eight-hours
-
who will happily come in and sort out your security issues later
I really doubt anybody will be happy about it, even after considering the size of the fees. And also, you have a very high estimation of the capacity of those people to notice they have to call you, I really doubt it's deserved.
-
Disclaimer: these tweets aren't real.
The tweet is fake.
I don't really think it being edited adds anything to the joke. -
It's just prompt engineering for coding. Let an AI dump a bunch of code for you, debug until it no longer errors, pull request and repeat next sprint.
5% of the time, it works every time
The critical detail being that you don't actually know what's inside (and it's definitely bad). Just using LLM assistance for a your boilerplate code doesn't count.
-
I’ve been programming for over half my life now, I actually like vibe coding with Claude these days.
It basically gets me through the hump of “ugh this task is going to be annoying as fuck to do” which is where I personally lose most of my efficiency (I have a lot of difficulty forcing myself to do something I don’t want to).
It’s like when I had interns and I’d give them tasks. Describe the work, scope it, add some guard rails to keep it directionally okay, and send it off to get reviewed later. And that works great with modern agents.
I will say vibe coding is damn good at debugging, way better than I am, so I use it for that a lot now.
I've been a Jr coming up to two years. When working on tasks I have a rough idea of what I want to achieve and some steps on the way there, but don't know how to actually implement it. I've found using copilot useful to fill in some of the gaps and give me ideas and direction.
I'm concerned that there are skills I am missing out on developing, but at the same time if AI is being pushed so heavily is it not something I should lean into to be better equipped in working with it?
-
Yeah, unlike an outsourcing outfit an AI company won't take the fall when given shit requirement and shit pay they deliver shit work.
