Hacker News Reader

Comment by mjr00 | Hacker News Reader

Maybe. I'm actually a big fan of Claude/Codex and use them extensively. The author of the article says the same.

> To be clear: I like and use AI when it comes to coding, and even for other tasks. I think it’s been very effective at increasing my productivity—not as effective as the influencers claim it should be, but effective nonetheless.

It's hard to get measured opinions. The most vocal opinions online are either "I used 15 AI agents to vibe code my startup, developers are obsolete" or "AI is completely useless."

My guess is that most developers (who have tried AI) have an opinion somewhere between these two extremes, you just don't hear them because that's not how the social media world works.

View on HN

Well I've just watched two major projects fail which were running mostly on faith because someone read too many "I used 15 AI agents to vibe code..." blog posts and sold it to management. The promoters have a deep technical understanding of the problem domain we have but little understanding of what an LLM can achieve or what it can understand relating to the problem at hand.

Yes you can indeed vibe code a startup. But try building on that or doing anything relatively complicated and you're up shit creek. There's literally no one out there doing that in the influencer-sphere. It's all about the initial cut and MVP of a project, not the ongoing story.

The next failure is replacing a 20 year old legacy subsystem with 3MLOC with a new React / microservices thing. This has been sold to the directors as something we can do in 3 months with Claude. Project failure number three.

The only reality is no one learns or is accountable for their mistakes.

> I've just watched two major projects fail

This is an opportunity. You can have a good long career consulting/contracting for these types of companies.

Why do you think I work there!

Emergency clean up work is ridiculous money!

My experience has been a mixed bag.

AI has led us into a deep spaghetti hole in one product where it was allowed free rein. But when applied to localised contexts. Sort of a class at a time it’s really excellent and productivity explodes.

I mostly use it to type out implementations of individual methods after it has suggested interfaces that I modify by hand. Then it writes the tests for me too very quickly.

As soon as you let it do more though, it will invariably tie itself into a knot - all the while confidently ascertaining that it knows what it’s doing.

On localised context stuff, yeah no. I spent a couple of hours rewriting something Claude did terribly a couple of weeks back. Sure it solved the problem, a relatively simple regression analysis, but it was so slow that it crapped out under load. Cue emergency rewrite by hand. 20s latency down to 18ms. Yeah it was that bad.

For me it's just wildly unpredictable. Sometimes it gets a small task perfectly right in one shot, sometimes it invents an absurd new way to be completely wrong.

Anyone trusting it to just "do its own thing" is out if their mind

For me I would ask it to do a simple thing and it would give me the tutorial code you could find anywhere on the Internet. Then you ask it to modify it in a way that you can't find in any example online, it will tell you it's fixed everything, but actually nothing has changed at all or it's completely broken.

I think if someone's goal was just the tutorial code, it would have been very impressive to them the AI can summon it.

this is what I've been using freebie gemini chat for mostly, example code, like reminding me of c stdlib stuff, javascript, a bit of web server stuff here and there. I think it would be fun to give googles agent or cli stuff a spin but when I read up here and there about antigravity, I'm reading that people are getting their accounts shutdown for stuff I would have thought was ok, even if they paid for it (well actually as usual the actual reasons for accounts getting zapped remain unknown as is today's trend for cloud accounts).

I'm too poor for local llms, I think there might be a 2 or 4gb graphics card in one of my junk pcs but thats about it lol

It only takes a cursory knowledge of what LLMs really are to understand why recreating tutorials is easy, but making actual new stuff that is well engineered (takes way way more than "passes the test suite") is difficult.

Actual novel stuff is so far out on the long tail of iterations that it's a gamble: it might pop up in an early run, or might take 2000 prompts and $20,000 worth of tokens. And it's still not really engineered, it's 10,000 monkeys with typewriters copying random shakespeare snippets off the chalkboard. At some point you'll get all of Hamlet, but most of the time you'll get garbage, and sometimes you'll get Romeo & The Taming of The Tempest.

I found that unpredictability to be interesting. I'm doing super simple projects with these models and a year, or even six months ago, it would give me a block of code and as soon as you ran it, it would fail. And you'd have to paste the error in and keep going until it was smoothed out.

The other day though I asked for something simple and it one-shotted the problem. To me, that's new.

I know this success was a statistical outlier, however. I grok how to use it and to not trust it. I'm just shocked so many people smart people fail to understand it.

Rather than making a good product that’s useful to the world, the goal of current startups seems to be milking VCs who are desperately searching for the new version of the mobile phone revolution that will make this all ok… so it seems like they’re accomplishing their goal?

I reckon the reason the VC rhetoric has reached running-hair-dye-Giuliani-speech level absurdity isn’t because they’re trying to convince other people— it’s because they’re trying to convince themselves. I’d think it was funny as hell if my IRA wasn’t on the line.

I think no one cares about the truth or building something good any more. It's a meme economy. Tell a story and the numbers go up. Until they don't.

Yes my pension is probably going down the same sinkhole with your IRA. Good luck. We need it.

Its always been like this but for awhile you had people like Steve Jobs to hold people like Bill Gates accountable. He long referred to MSFT as being mcdonalds in relation to the stuff they produced - very pedestrian.

Yep, all that accountability gates has faced in his life.

Is he facing charges yet for sneaking drugs into his wife's food, or did he only ever discuss that with his buddy Jeff E and never actually follow through with it

It's worse than a meme economy, it's a gambling economy. The entire VC business model is gambling (fund 10 hope 1 pays for the losers). Crypto is all gambling. The stock market is gambling. TV ads are all gambling ads. Even dating is gambling.

Now programming and art are both gambling.

> "AI is completely useless."

This is a straw man. I don't know anybody who sincerely claims this, even online. However if you dare question people claiming to be solving impossible problems with 15 AI agents (they just can't show you what they're building quite yet, but soon, soon you'll see!), then you will be treated as if you said this.

AI is a superior solution to the problem Stack Overflow attempted to solve, and really great at quickly building bespoke, but fragile, tools for some niche problem you solve. However I have yet to see a single instance of it being used to sustainably maintain a production code base in any truly automated fashion. I have however, personally seen my team slowed down because code review is clogged with terribly long, often incorrect, PRs that are largely AI generated.

I use both Claude and Codex (Claude at work, Codex at home).

They are fine, moderately useful here and there in terms of speeding up some of my tasks.

I wouldn't pay much more than 20 bucks for it though.

Reply on news.ycombinator.com