Hacker News Reader

Comment by TheCleric | Hacker News Reader

I am so glad someone else has this same experience as me because everyone else seems all in and I feel like I’m staring at an emperor without clothes.

View on HN

The truth often lies somewhere in between

My personal experience indicates this, AI enhances me but cannot replace me

Been doing something closer to pair programming to see what "vibe" coding is all about (they are not up to being left unattended)

See recent commits to this repo

https://github.com/blebbit/at-mirror/commits/main/

You are not alone. There are plenty of us, see here:

- Claude Code is a Slot Machine https://news.ycombinator.com/item?id=44702046

- GPTs and Feeling Left Behind: https://news.ycombinator.com/item?id=44851214

- I also used the Emperor/clothes metaphor: https://news.ycombinator.com/item?id=44854649

And just so we are clear, in the only current actual study measuring productivity of experienced developers so far, it actually led to 19% decline in productivity. https://metr.org/blog/2025-07-10-early-2025-ai-experienced-o...

So, if the study showed experienced developers had a decline in productivity, and some developers claim gains in theirs, there is high chance that the people reporting the gains are...less experienced developers.

See, some claim that we are not using LLMs right (skills issue on our part) and that's why we are not getting the gains they do, but maybe it's the other way around: they are getting gains from LLMs because they are not experienced developers (skills issue on their part).

I'll wait for more studies about productivity, one data point is not solid foundation, there are a lot of people who want this to be true, and the models and agent systems are still getting better

I'm an experience (20y) developer and these tools have saved me many hours on a regular basis, easily covering the monthly costs many times over

There’s not enough data…which somehow proves your pre-existing bias which goes against the only data we have.

I'm offering my anecdotal evidence as contrary to the loss of productivity piece that has been making the rounds, not as justification that they are wrong, but that we don't have enough data

It's also not the only data, here's one with results in the other direction

https://medium.com/@sahin.samia/can-ai-really-boost-develope...

Glad to see more data!

It’s interesting they showed the biggest gains for junior developers. The other study showing productivity losses for experienced developers. That suggests these tools are a lot more helpful for junior developers compared to senior developers at the moment.

From my experience, a senior dev who knows how to work these tools will be better than both. We need data on how long, and how much effort into developing the skill, working with AI a developer has put in.

Theore you use it, the more you get a feel for when and when not to use it

> Productivity was tracked through metrics such as completed tasks (pull requests), code commits, and successful builds.

Making untested garbage faster to check off tasks quicker. Reopen rate please? New bug task rate? Nobody looked...

> The study also monitored code quality via build success rates. Importantly, increased productivity did not come at the cost of more errors, showing that Copilot helped developers code faster and more accurately.

It builds therefore it works. And the test suite was also AI generated?

Feels like we're back in primary school learning programming.

Classic case of gaming the metrics.

You sound like you haven't given Copilot and friends a thorough chance or evaluation. If you had, you'd know that...

- people don't outsource, they pair-program, because these things cannot do complex tasks on their own

- they are quite good at running tests with coverage, inspecting the results, and fixing both the tests and code

- people make mistakes, expecting AI to be perfect is unreasonable, they are tools, not replacements

> Making untested garbage faster to check off tasks quicker

> Feels like we're back in primary school learning programming.

> Classic case of gaming the metrics.

plain view bias causes others to discount your opinion

> people make mistakes, expecting AI to be perfect is unreasonable, they are tools, not replacements

This is the key. These tools are an improvement for many people, but others pooh-pooh them for not being perfect. Working in a team with other programmers (or looking at my own older code) I often see mistakes, often obvious to me now.

> So, if the study showed experienced developers had a decline in productivity,

You forgot to add: first time users, and within their comfort zone. Because it would be completely different result if they were experienced with AI or outside of their usual domain.

Your comments are citing this blog post and arxiv preprint.

You are also misrepresenting the literature. There are many papers about LLMs and productivity. You can find them on Google Scholar and elsewhere.

The evidence is clear that LLMs make people more productive. Your one cherry picked preprint will get included in future review papers if it gets published.

Good thing you linked to these articles that support your claims, otherwise I would be left feeling like you are misrepresenting the literature to push your own agenda.

You have the ability to search Google Scholar, I'm quite sure of it

I do, and did, and found you were spreading misinformation.

This is false, but the nice thing is that anyone can go check it and don't have to rely on the claims of an account created specifically to reply to me.

This is false, this account was made specifically to inform people that a correct proof of P=NP is on Google Scholar.

Go check it!

I found nothing. Could you share the links you found, or are you lying

What were you using? Did you use it for a real project? I ask because you're going to have a vastly different experience with Cursor than with Claude Code, for example.

Didn’t take long for the “you’re using the wrong tool / holding the tool wrong” replies to appear.

Knowing how to search prompt Google became a skill among programmers, so has prompt/context engineering

Nah, Google made search unreasonably good at finding what users are looking for regardless of how poor the query was. Calling searching in Google a “skill” is similar to calling getting sucked into social media a skill.

It's different for programming, and especially evident when you are looking to solve a problem with an error message

You add “site: stack overflow.com”.

and then you don't also get GitHub, forums, reddit, etc..

I have very rarely needed the site: modifier, and even knowing this exists sets people apart, and reinforces there is skill to it

Recognition of that fact and then applying what you’ve learned about how to effectively wield the tool is a skill.

Unless the tool is trash and breaks, or otherwise unfit for the purpose.

You can hammer in nails with your knuckles. Do you want to do that daily? There are people who can, they're still worse at it than a plain hammer or someone even using a right rock. Much less an assembly line.

Sure, we're talking about how there's no skill use with Google. But there is. That doesn't mean that every tool is good for every purpose. I agree that not every tool is good for every purpose. That also doesn't mean that every time a tool isn't useful, it's because it's not fit for purpose. Sometimes, it's because the wielder doesn't have the appropriate skills.

It's really odd to me that pointing this out is used as some kind of counterargument. On its own that makes no sense.

Consider that the frequency of replies along those lines might be evidence that there's something to it. It's not necessarily true, of course, but if it's false then you need to explain why so many people believe otherwise.

My work has offered us various tools. Copilot, Claude, Cursor, ChatGPT. All of them had the same behavior for me. They would produce some code that looks like it would work but hallucinate a lot of things like what parameters a function takes or what libraries to import for functionality.

In the end, every tool I tried felt like I was spending a significant amount of time saying “no that won’t work” just to get a piece of code that would build, let alone fit for the task. There was never an instance where it took less time or produced a better solution than just building it myself, with the added bonus that building it myself meant I understood it better.

In addition to that I got into this line of work because I like solving problems. So even if it was as fast and as reliable as me I’ve changed my job from problem solver to manager, which is not a trade I would make.

Reply on news.ycombinator.com