Read your code (etsd.tech)

[dead]

Title made me think this was about writing code for vibrators.

Part of the definition of vibe coding is not looking at the code. If you read code you are not "vibing" anymore, you are just using AI tools to write code.

I'm not going to read code created by an AI. The AI exists to prevent me from having to deal with the complexity of a task. The absolute most I want to read and write are type signatures. I'll set those up, let the AI go, see if it works. If it doesn't, maybe retry with a better prompt. If it still doesn't work, then I'll have to get involved. Start implementing from the top down, and once there is enough architectural structure, the AI can usually finish up.

This still happens quite a bit, and it's just like taking away a hard task from someone less experienced. The difference is there is no point in investing your time teaching or explaining anything to the AI. It can't learn in that way, and it's not a person.

New vibe coding definition just dropped! "Vibe-Coding is a dialogue-based coding process between a human and an AI where the human guides and the AI implements."

Reminds me of Steve Yegge's short-lived CHOP - Chat Oriented Programming: https://sourcegraph.com/blog/chat-oriented-programming-in-ac...

I remain a Karpathy originalist: I still define vibe coding as where you don't care about the code being produced at all. The moment you start reviewing the code you're not vibe coding any more, by the definition I like.

"You cannot delegate the act of thinking"

This is very dumb. Of course you can.

That article is like advising a blind person to thoroughly look at the road before crossing it.

You can be an architect without reading the code. My process involves building a detailed plan before starting, at which point I ask lots of questions about architecture. After implementation I have a custom agent for reviewing the architecture and I usually ask a few questions at this point as well.

In what other area is it fashionable to use your own shortcomings and lack of knowledge as evidence that something is a bad idea?

Personally, I find that if a model can vibe-code the functionality I'm working on then it's not very high-value functionality. Perhaps (a) it's boilerplate (fine), (b) I'm not creating enough/the right abstraction, or (c) the code could easily be written by a junior dev. If I'm working on truly new functionality, modeling complex states and assumptions, or producing something that generalizes to many settings, the model does poorly beyond being a souped up auto-complete.

That's not to say that these models don't provide value, especially when writing code that is straightforward but can't be easily generalized/abstracted (e.g., some test-case writing, lots of boilerplate idioms in Go, and basic CRUD).

In terms of labor I potentially see this increasing the value (and therefore cost) of actual experienced developers who can approach novel and challenging problems, because their productivity can be dramatically amplified through proper use of AI tooling. At the other end of spectrum, someone who just writes CRUD all day is going to be less and less valuable.

These tools continue to increase in power. It’s incredible and there’s no use denying it. Asking people to read ai generated code will soon make as much sense as asking people to read compiler generated ASTs.

> Not reviewing AI-generated code will lead to serious problems.

A requirement to do so might lead to more. Like loss of job for the illiterate "programmer".

Related reading: Programming as theory building ~ Peter Naur

[PDF] https://pages.cs.wisc.edu/~remzi/Naur.pdf

If you write code yourself your productivity is limited by how much code you can write. If you read/review code generated by AI, then it's limited by how much you can read/review. The latter is not that much bigger than the former. I believe at some point we'll have to let go.

Are LMM compilers that are generating computer understandable instructions. I think of back in the day people use to chisel assembly code by hand. then came a system that all people to code more in english-like commands(prompts).

I've found that there's a pretty consistent relationship between how clearly I can imagine what the code should look like, and how effective vibe coding is. Part of the reason for that is that it means I'll be more opinionated about the output of the model, and can more quickly tell whether it's done something reasonable.

I'm waiting for LLMs to vibe-code new algorithms.

Currently, the only way to understand code is to read it. You no longer need to understand code to produce it (maybe in some pre-AI cases that was also true).

So no, you don’t _need_ to read code anymore. But not reading code is a risk.

That risk is proportional to characteristics that are very difficult, and in many cases impossible, to measure.

So currently best practice would be to continue reading code. Sigh.

This comment section is unbelievable lol. I'm genuinely struggling to believe this many people would let code into a codebase without review.

Coding with AI agents is very similar to working with junior developers (or even other developers) in a team. Someone needs to give the project direction, keep the code organized, suggest refactorings, and many more things.

I guess it's really hard to work with AI agents, if you don't have real project experience in a more senior position.

we can arrange them, not arranged by them

Reading code is just as fun as writing it, to be honest