Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Ask HN: Do you have any evidence that agentic coding works?

news.ycombinator.com

Post reply on HN
Print view View original

311–320 of 478 posts

Re: Ask HN: Do you have any evidence that agentic coding works?

#311

Post by zvqcMMV6Zcr »

julianeon wrote: post #117

As far as I can tell, there are exactly 3 use cases that have demonstrably worked with AI, in the sense that their stakeholders (not the AI companies, the users) swear it works. 1. training a RAG on support questions for chat or documentation, w/good material 2. people doing GTM work in marketing, for things like email automation 3. people using a combination of expensive tools - Claude + Cursor + something else (may…

I know of 4th - line level interference. I love how it works in IDEA, 30% of time I just slap TAB to accept generated code, if it is wrong I just continue writing as usual. Avoids most drawbacks: you don't pay for it(it is running on local CPU), the "review" is instant while still providing boost to productivity.

Re: Ask HN: Do you have any evidence that agentic coding works?

#312

Post by J8K357R »

My personal experience, agentic coding produces brittle code. Often it works but it violates every opinionated fiber of coding style in me as a developer, even when tuned to my style. It does not give me code that I want to maintain long term. But did I mention it often works. If you treat it like code, written by a petulant toddler, that needs review and refactoring then it works well. It’s also pretty good at code reviewing your code, finding subtle mistakes.

Re: Ask HN: Do you have any evidence that agentic coding works?

#313

Post by fatherwavelet »

closewith wrote: post #239
dust42 wrote: post #216

Earlier quoted context omitted.

> This stuff is relatively new, I don't think anyone has truly figured out how to best approach LLM assisted development yet. Exactly. But as you say, there are so many people riding the hype wave that it is difficult to come to a sober discussion. LLMs are a new tool that is a quantum leap but they are not a silver bullet for fully autonomous development. It can be a joy to work with LLMs if you have to write the um…

> But as you say, there are so many people riding the hype wave that it is difficult to come to a sober discussion. LLMs are a new tool that is a quantum leap but they are not a silver bullet for fully autonomous development. While I agree with the latter, I actually think on former point - that hype is making sober discussion impossible - is actually directionally incorrect. Like a lot of people I speak to privately…

It is delusional to believe the current frontier models can only write CRUD apps.

I would think someone would have to only write CRUD apps themselves to believe this.

It doesn't matter anyway what a person "believes". If anything, I am having the opposite experience that conversing with people is becoming a bigger and bigger waste of time instead of just talking to Gemini. It is not Gemini that is hallucinating all kinds of nonsense vs the average person. It is the opposite.

Re: Ask HN: Do you have any evidence that agentic coding works?

#314

Post by njhnjh »

dagss wrote: post #193

I've been programming for 20 years, and I've always been under-estimating how long things will take (no, not pressured by anyone to give firm estimates, just talking about informally when prioritizing work order together). The other day I gave an estimate to my co-worker and he said "but how long is it really going to take, because you always finish a lot quicker than you say, you say two weeks and then it takes two…

[dead]

Re: Ask HN: Do you have any evidence that agentic coding works?

#315

Post by jrjeksjd8d »

A key thing I noticed in sincere anecdotes about LLM code is that is always seems to be outside of the author's area of expertise.

I work with an infrastructure team that are old school sysadmins, not really into coding. They are now prodigiously churning out apps that "work" for a given task. It is producing a ton of technical debt and slowing down new feature development, but this team doesn't really get it because they don't know enough software engineering to understand.

Likewise the recent example of an LLM "coding a browser" where the result didn't compile and wasn't useful. If you took it at face value you'd think "wow that's a hard task I couldn't do, and an LLM did it alone". In fact they spent a ton of effort on manually herding the LLM only for it to produce something pretty useless.

Re: Ask HN: Do you have any evidence that agentic coding works?

#316

Post by keyle »

Just watch the first 2 mins of this and try to keep a straight face.

https://www.youtube.com/watch?v=4OmlGpVrVtM

You'll have your answer.

PS: don't try to use his website/apps, half of it is broken... and he has generated a 'jobs' page on the main app's website which made me laugh so hard I got a coughing fit.

Re: Ask HN: Do you have any evidence that agentic coding works?

#317

Post by coryrc »

cocoto wrote: post #188
dysleixc wrote: post #158

Earlier quoted context omitted.

[flagged]

From the very beginning everyone tells us “you are using the wrong model”. Fast forward a year, the free models become as good as last year premium models and the result is still bad but you still hear the same message “you are not using the last model”… I just stopped caring to try the new shiny model each month and simply reevaluate the state of the art once a year for my sanity. Or maybe my expectation is clearly…

Are you sure you haven't moved the goalposts? The context here is "agentic coding" i.e. it does it all, while in the past the context was, to me anyway, "you describe the code you want and it writes it and you check it's what you asked for". The latter does work on free models now.

Re: Ask HN: Do you have any evidence that agentic coding works?

#318

Post by solaris2007 »

christophilus wrote: post #306
solaris2007 wrote: post #276

Earlier quoted context omitted.

[flagged]

> If you are any good with those four languages, you are leagues ahead of anyone who does Javascript full time. That is a priggish statement, and comes across as ignorant. I’ve been paid to program in many different languages over the years. Typescript is what I choose for most tasks these days. I haven’t noticed any real difference between my past C#, C++, C, Java, Ruby, etc programming peers and my current JavaScri…

> That is a priggish statement

A cursory glance at the definition of "prig" shows that what I wrote there is categorically not. You should at least try to look up that word and if you look it up and still don't get it then what you have is a reading comprehension issue.

> Typescript is what I choose for most tasks these days.

So you're smart on this, at least. Cantrill said it really well, Typescript brought "fresh water" to Javascript.

> haven’t noticed any real difference between my past C#, C++, C, Java, Ruby, etc programming peers and my current JavaScript ones.

You might still be on their level. I see that you didn't mention Rust or at least GoLang. Given the totality of your responses, you're certainly not writing any safe C (not ever).

Re: Ask HN: Do you have any evidence that agentic coding works?

#319

Post by Cthulhu_ »

> with the claim that we should move from “validating architecture” to “validating behavior.” In practice, this seems to mean: don’t look at the code; if tests and CI pass, ship it.

The thing that people don't seem to understand is that these are two separate processes with separate goals. You don't do code reviews to validate behaviour, nor do you test to validate code.

Code reviews are for maintainability and non-functional requirements. Maintainability is something that every longer term software project has run into, to the point where applications have been rewritten from scratch because the old code was unmaintainable.

In theory you can say "let the LLM handle it", but how much do you trust it? It's practially equivalent to using a 3rd party library, most people treat them as a black box with an API - the code details don't matter. And it can work, I'm sure, but do you trust it?

Re: Ask HN: Do you have any evidence that agentic coding works?

#320

Post by maerch »

embedding-shape wrote: post #277
zh3 wrote: post #274

Earlier quoted context omitted.

NP, and the exact definition of vibe-coding is, I think, yet to be determined. This wasn't a yolo, it was read all the prompts and generally accept them. Overall I'd say the code and web page are at least of a quality I've seen in many commercial settings; the code itself looks reasonable and if I was to do anything to it for a real 'release', I'd update the documentation which has suffered due to the extensive scope…

> the exact definition of vibe-coding is, I think, yet to be determined Huh? No, that's been established since Karpathy coined the term; you don't review the code, only use the agent and don't care about how it was done, just about the results. The actual interesting stuff is how to use LLMs together with a human, to build high quality code. More "augmenting the human intellect" rather than "autonomous robots buildin…

> Huh? No, that's been established since Karpathy coined the term; you don't review the code, only use the agent and don't care about how it was done, just about the results.

However, nowadays it is used as a synonym for everything that is somehow generated by an LLM. Regardless of whether it is a spec-driven, carefully reviewed and iterative piece of software or some yolo-style one-prompter with no idea how it was done.

Post reply on HN