Revenge of the Junior Developer | Hacker News Reader

Revenge of the Junior Developer

sourcegraph.com

68 pointsado__dev89 comments

I’m usually one of the people complaining about hype cycles, and it’s usually been correct to be pessimistic about them.

But in this particular case I have to think a lot of people just haven’t tried it in its best form. No, not a local model on your MacBook. No, not the web interface on the free plan. Go lay down $300 into API credits, spend a weekend (or maybe two) fully setting up aider, really give it a shot. It’s ultimately a pretty small amount of money when it comes to figuring out whether the people who are telling you there’s an existential risk to your livelihood on the horizon are onto something, don’t you think?

$300? That much sounds like it would last a year. You don't need to spend anywhere near $300 just to try things out

RTFA, agents churn through credits. Further enforcing my belief that the vast majority of people have not actually tried it yet.

> RTFA, agents churn through credits.

I literally use Claude Code as part of my job, so either the "FA" is wrong, or my costs are only low because work just happens to have a codebase that reduces costs (lol)

We're using Cline, and it's very easy to blow through $20-40 if you get going. The value proposition is absolutely there so we eat the cost, but OP is correct in that agentic coding eats tokens like there's no tomorrow.

I burned through $50 credits in a sitting the first time I used Claude Code to see what I could vibe-code from scratch. You're pushing whole files of tokens through it every time you interact with it.

Huh, maybe the reason it doesn't use this much for me is because my work has mostly source files that are far too large for Claude Code to read. It always has to ask for permission to use grep because it can't figure out how to use the read tool properly. I've done entire new features on probably around $15 of credit max.

for better or worse, Claude code is less parsimonious with tokens compared to aider for the same thing.

> It’s ultimately a pretty small amount of money when it comes to figuring out whether the people who are telling you there’s an existential risk to your livelihood on the horizon are onto something, don’t you think?

Nope. I'd rather buy some books or a Jetbrains subscription.

I find myself much the opposite - I don't usually complain about hype cycles, thinking we should wait and see before reserving judgement. In this case I feel like we've seen enough to know LLMs are not capable of performing anyone's job.

How much does the Deepseek's equivalent cost?

I know this isn't what you were asking, but the answer is $300. And if I wrote $3000 in the original comment, the answer would be the same. You cannot have enough money.

> I have bad news: Code completions were very popular a year ago, a time that now feels like a distant prequel. But they are now the AI equivalent of “dead man walking.”

I disagree. I view this as a "machine guns versus heat sinking missiles from the 70s" dichotomy. Sure, using missiles is faster. However, sometimes you're too close for missiles. Also, machine gun rounds are way cheaper than missiles. However, when they first came out, missiles were viewed as the future. For a while, fighter jets were made without machine guns, but they added them back later because they decided they needed both.

Sometimes I find I want to drill down and edit what Claude generated. In that case, copilot is still really nice.

With regard to ai assisted coding: the more you know what you're doing, the more you know the code base, the better result you'll get. To me it feels like a rototiller or some other power tool. It plows soil way faster than you can and is self propelled, but it isn't self directed. Using it still requires planning and it's expensive to run. While using the tool, you must micromanage its direction, constantly giving it haptic feedback from the hands, or it goes off course.

A rototiller could be compared to a hired hand plowing himself, I guess, but there's way less micromanagement with a hired hand vs a rototiller.

Kind of like horses and cars. Horses can get you home if you're drunk. Cars can't.

The proper use of AI agentic tools is like operating heavy machinery. Juniors can really hurt themselves with it but seniors can do a lot of good. The analogy goes further: sometimes you need to get out of the backhoe and dig with smaller tools like jackhammers or just shovels. The jackhammer is like copilot -- a mid-grade power tool -- and Claude code is like the backhoe. Clunky, crude, but can get massive amounts done quickly, if that's what's needed.

> Clunky, crude, but can get massive amounts done quickly, if that's what's needed.

You know what's quicker in your analogy? A spell. Or in the coding world. Template, snippets, code generators, framework, and metaprogramming. Where you abstract all the boilerplate behind a few commands. You already know the blast radius of your brute modification tools, so you no longer have to micromanage them. And it's reliable

I love how new technology becomes like religion. It develops both cult followers and critics.

In that lens think the AI cult is more right than the Crypto cult. At least I can use it to do something tangible right now while crypto is still pretty useless after many years.

In some sense I think these technologies need the cults and the critics though. It’s good to have people push new things forwards even if everyone isn’t along for the ride. It’s also good to have a counter side poke holes. I think the world is better with both optimists charting new paths forwards and pessimists making sure they don’t walk right off a cliff.

Cryptocurrency allows people who are otherwise limited by draconian payment platform limitations, puritanical moralists who own credit card companies and such to buy and sell things online without being stopped. It's obviously not a good system but it provides a high effort release valve, a fallback mechanism that hopefully will undermine some of these draconian measures. Lots of people have been helped by the existence of cryptocurrency because of this, especially since lately a lot of bad actors control payment platforms and often just shut down businesses on whims.

Whether more have been helped or hurt is debatable but it certainly has a tangible, if niche, use case with real value. It certainly has no value as a store of value, though.

Evidence suggests it's the exact opposite. The only thing crypto has really been good for is a store of value. Sure, it's a volatile commodity, but over any 5 year average window, Bitcoin definitely beats inflation. That makes it a safe store of value.

Huh. While I agree that Bitcoin specifically has been a successful speculative investment in the past, "a safe store of value" is a massive stretch. It's a safe store of value in the same way a S&P 500 index fund is, except more volatile. What's that phrase... "Past performance is not indicative of future gains?"

What evidence? People use it to bypass the payment processor cartels all the time.

Define all the time. I haven’t seen a shop that accepts crypto. There’s nothing I could buy with it.

A lot of bad actors control most Crypto platforms. Some of them are even justly rotting in jail (until Trump pardons them), and a hell of a lot more of them deserve to be in jail, while most AI users aren't actively committing crimes, laundering and embezzling money, dodging taxes, pumping and dumping rug pulls and pyramid schemes, and committing fraud.

It's been such a relief since the Crypto scammers finally shut the fuck up with their incessant ShitCoin and NFT shilling and get-rich-quick pyramid schemes, so please don't any of you start up again, for the love of God.

AI is NOT like Crypto in any way shape or form, and this is an AI discussion, not a Crypto discussion. And I'm sick and tired of hearing from Crypto shills yapping HODL and FUD while I'm actually getting productive work done and making real money while creating tangible value and delivering useful products with AI, without even having to continuously recruit greater fools and rip off senior citizens and naive suckers of their life savings by incessantly shilling and pumping and dumping and pulling rugs out from under people.

other countries (e.g. India), just have... built better ordinary payment networks. Or laws for them (e.g., Europe). The US... has regulatory capture by draconian platforms.

Crypto is still pretty valuable relative to how useless it remains.

Unless you are imagining a world in which there's a global conflict and crypto isn't shut down in the first 12 months.

Even if I would agree with everything the article says, I have no idea how the author gets to the conclusion that junior developers will prevail because they are faster at adopting LLMs.

Didn't he just made a point about how fast the situation is evolving? I had some FOMO about ai last year, not anymore. I don't care that I don't have time to fully explore the current LLM state of the art, because in a month it will be obsolete. I'm happy waiting until it settles down.

And if their scenario ends up happening, and you can basically multiply a dev's productivity by N by paying N x K dollarinos, why would you chose a junior dev? It's cheaper, but sometimes a junior dev doesn't take longer to arrive at a solution, it never does (same for senior devs, don't get me wrong, but it happens less often).

It's a counterpoint to Yegge's original post https://sourcegraph.com/blog/the-death-of-the-junior-develop...

And it's not saying his original post is wrong, they should be taken together. He's saying those who adapt to the new paradigm will "win", whether senior or junior.

the only good part was the joke about "vibecoding" (shudder what a stupid term) being like a fart and attracting flies... ok investors

still, this "ai code tools will deprecate real programming" bullshit will one day be laughed at just like how most of us laugh at shitcoin maniacs

it just takes a lot of people way too long to learn

Maybe there's a different universe out there, where the code you write is not expected to work, so you can poke the LLM for a whole day to see if it barfs something out.

I spend much of the day reading and thinking and only a small portion actually writing code, because when I'm typing, I usually have a hypothetical solution that is 99% correct and I'm just bringing it to life. Or I'm refactoring. You can interrupt me at any time and I could give you the complete recipe of what I'm doing.

Which is why I don't use LLMs, because it's actually twice the work for me. Typing out the specs, then verifying and editing the given result, while I could type the code in the first place. And they suck at prototyping. Sometimes I may want to leave something in the bare state where only one incantation works, because I'm not sure of the design yet, and have a TODO comment, but they go to generate a more complicated code. Which is a pain to refactor later.

I agree with the spirit of the argument, but I don’t think you’re taking into account the scale of “typing” we’re talking about now.

For example, yesterday I needed a parser for a mini-language. I wrote a grammar — actually not even the formal grammar, just some examples — and what I wanted the AST to look like. I said “write the tokenizer”, and it did. I told it to tweak a few things and write tests. It did. I told it to “write a recursive descent parser”, and it did. Add tests and do some tweaks, done.

The whole thing works and it took less than an hour.

So yeah, I had to know I needed a parser, but at that point I could pretty much hand off the details. (Which I still checked over, so I’m not even in full vibe mode, guess.)

This is a cool use case that probably saved you some time, but writing a recursive descent parser is something freshman CS students do for a lab assignment.

It isn’t exactly breaking new ground or doing anything you couldn’t find with a quick google search.

You want me to do a Google search for a parser for the language that I just invented?

If you have a parser generator tool that can go from zero to parser based on a few examples rather than a grammar, I’d love to hear about it.

> If you have a parser generator tool that can go from zero to parser based on a few examples rather than a grammar, I’d love to hear about it.

That's the part where the LLM would be useful. Turn the examples into a grammar, verify the grammar, and use a parser generator for the rest.

You know the code does what it should and you get there in minutes.

I'm pretty sure it could have done that equally easily, but this grammar is small enough that a parser generator is overkill. (I made that decision before I asked the LLM to write the parser.)

Isn't that what tools like antlr [0], bison[1] do?

[0]: https://www.antlr.org/

[1]: https://www.gnu.org/software/bison/

They typically generate generic LALR parsers, not custom recursive descent parsers. I'm no expert, but my understanding is that custom recursive descent parsers have advantages in terms of readability and error message generation.

"add tests and do some tweaks"

there's the snake in the grass - you're asking the LLM to write tests after the code is generated, so naturally, it will write passing tests - including tests which enforce buggy behavior

if not now, then you're lucky; watch out for next time.

I disagree, agentic coding is amazing for prototyping. It removes a ton of friction and inertia and allows me to try a bunch of ideas to see what works (from a delivery and/or UX perspective). Of course, you should have the systems thinking and experience in place to know what it is you're doing with it at any given point , but for demos and prototypes it has been an absolute boon.

Generally, you wouldn't type out the spec either, either you provide an existing spec to the model (in form of white board notes, meeting notes, etc.) or you iterate conversationally until you arrive at an initial implementation plan.

It's a different way of working for sure, and it has distinct draw backs and requires different mental modes depending if you're doing green field, demo/prototype, existing large app feature development etc. but it's been a massive productivity enhancement, especially when tackling multiple projects in quick succession.

A good example of this at the extreme is game design.

When you're making a game, it's really expensive to try out different ideas, like a few different implementations of a mechanic. It could be hours of work to make a change. You tend to have to do a lot of thinking to tweak a mechanic in a fundamental way just to see what it feels like, knowing that you're probably going to throw it away.

LLMs are really good at this. Make a git branch, ask Claude Code to tweak the physics so that it does X, see what it feels like. Rollback the change or continue tweaking.

Same with branching the UI to see what a different idea would feel like. Simple changes to explain could result in hours of refactoring UI code (chore work) just to see if you like the result. Or just ask the LLM to do it, see what it feels like, roll it back.

> LLMs are really good at this. Make a git branch, ask Claude Code to tweak the physics so that it does X, see what it feels like

How do you verify that Claude did what you wanted?

Maybe more importantly, how do you verify that Claude didn't make a change that you didn't want?

Trying to verify the code changes just by testing the running project seems incredibly careless

You READ what it did and REVIEW every change and ASK it to explain and LEARN from it so you can do it yourself too: the OPPOSITE of "vibe coding", whose goal is to avoid learning and reading and reviewing and understanding at all costs, to live a careless effortless TLDR life of self imposed ignorance.

There's a responsible enlightened self enriching way to use AI tools, and there's "vibe coding".

> You READ what it did and REVIEW every change and ASK it to explain and LEARN from it so you can do it yourself too

I don't believe anyone will actually use AI this way long-term

Even if we start out this way, eventually people will trust it enough that they get complacent and mistakes will start to slip through

Besides which, this doesn't actually sound faster than just writing the code yourself

Speak for yourself. That's the way I use it, so I get a lot more out of it and get a lot more done than a "vibe coder" who's lazy and doesn't want to put in any effort or learn anything new.

You simply can't learn nearly as fast with vi or Emacs or VSCode, and they can't explain unknown code and APIs and algorithms and bug fixes to you, suggest best practices, review code, write documentation for existing code, check and align code with updated documentation, or automatically update existing documentation and comments to reflect the actual code.

Right now I'm using Cursor to refactor and overhaul a huge complex multi-repo codebase of Python, TypeScript, Bash, SQL, JSON, YAML, Markdown, and ML models, from multiple repos with Google Cloud Build to a monorepo with GitHub actions, and it's no cake walk, a huge amount of work, I'm learning and applying a whole lot of new stuff, meticulously documenting every part of the system, making it rock solid, DRY, with extensive logging and error handling, and there's no way I would have been able to do 5% of this work in the same amount of time without using AI. You don't HAVE to be lazy, instead you can work as hard as you've ever worked without AI and get a hell of a lot more done while learning new things, if you so choose to. It's all up to you.

If you insist on going into it refusing to use it as a learning tool, and don't want to work hard, then you be you, and I'm going to have a huge advantage over you or any "vibe coder".

You asked "How do you verify that Claude did what you wanted?" and I told you. Just because you're too lazy and incurious to drink the water I led you to doesn't mean I or anyone else can't.

Yup, this is gonna be like autopilots and aeroplanes all over again.

It's a shame that we'll invest so much money and compute into this, and continue to ignore all the ergonomics research that could help.

Usually my own way of working is to use Balsamiq[0] to have a visual prototype to test out flows, Figma|Sketch for the UI specs, then to just code it. Kinda the same when drawing where you just doodle until you have a few workable ideas, iterate of these to judge colors and other things, and then commit to one for the final result.

[0]: https://balsamiq.com/

Yeah, call me a cynic or conservative or whatever - I'll believe it when I see it. I give very little weight to predictions about the future from AI shills, especially when they include some variant of "we're 90% there already" or "an exponential shift is imminent, if things keep improving at this rate, which they Will." Opinion discarded, create your thing and come back if/when it works.

Everything is shifting so fast right now that it hardly matters anyways. Whatever I spend time learning will be outdated in a few years (when things are predicted to get good). It does matter if you're trying to sell AI products, though. Then you gotta convince people they're missing out, their livelihood is at stake if they don't use your new thing now now now.

I had this rather comical picture where developers finally get to experience what it is like to have someone write software for you. You get sort of what you asked for but it is obviously wrong to even the most novice user. You then change the requirements and get something entirely different but equally wrong or worse... and a new invoice. hahaha It gets more funny the more I think about it.

Everyone: telling me how great AI is.

No one: making anything great with AI.

Sourcegraph: an AI company, routinely promoting their LLM-optimism blogposts to HN, perpetuating the hype cycle their business model depends on.

This sapling is twice as large as it was a week ago, which was twice again as large as it was the week before. Why, at this rate, it'll be bigger than the whole world in but a month.

I remembered the dialogue from the movie Snatch:

    T: What's happening with this sausages, Charlie?
    C: 2 minutes, Turkish.
    -- 5 Minutes later ---
    T: How long for the sausages?
    C: 5 minutes, Turkish.
    T: It was 2 minutes, 5 minutes ago.

I don't know why I remembered it. Is it AI, or self driving cars, or both. Huh.

Don't forget the electric air taxis!

Drone-bot delivering my new dishwasher-top-rack-rails today; waiting 4 days for a human; SMH

Where's my TacoCopter??!

https://tacocopter.com/

self driving cars are here, just unevenly distributed. Waymo operates in several cities already, providing millions of rides to people. Just because you haven't seen them doesn't mean they don't exist.

Waymo operates in tiny parts of several cities when you look at the maps. They didn't finish the first 90% when you look from that perspective.

Waymo provides a tech demonstrator in a closed circuit, and you can pay to ride their demonstrators in select areas of select cities. It's not half bad, let's be honest, but having these limits puts them to a point very far from "self driving cars are here". Even Tesla isn't there with their camera only subsystems and "if you crash you keep both pieces" policy. It's another tech demonstrator you can buy.

If we accept your point that "self driving cars are here", we should accept that Fusion energy is a solved problem because WEST tokamak held plasma for 22 minutes, electric propulsion is a solved problem since we have electric trains for a long time, and Hyperloop is fully operational because I have scale model in my backyard which carries thousands of ants every day.

People open their phone, order a car, take a ride from an arbitrary point within the geofenced area, to another arbitrary point inside that area. Within that area. they're handling real life conditions. dogs and children running into the street, bicyclists, moving trucks blocking streets, construction. Other cars. Maybe that's not enough for you, but that's not a carefully staged tech demo with a car driving a loop around a fixed track at Disneyland that's gonna break the moment it has to deviate beyond what you're allowed to touch. Unless you think Waymo is paying all the drivers in LA drivers to drive different around their cars, I guess.

If the tokomak was energy positive for those 22 minutes, there was more than one of them, and they were running it multiple times a day then I would say it is. But it's not. I'm not sure why you have this need to lump them together. Fusion is quite a different problem, with yet to be solved materials science problems and control systems issues. I'm not claiming that fusion's here, and also I don't think quantum computers are gonna be here anytime soon.

Oh good. When I saw the graph I started wondering what I was being sold.

> We’re talking about each developer gradually boosting their productivity by a multiplier of ~5x by Q4 2025 (allowing for ramp-up time), for an additional amortized cost of only maybe $50k/year the first year. Who wouldn’t go for that deal?

OK, I'll take the other side of that bet. If in Q4 '25 devs using cursor or whatever are 5x as productive as me using emacs, I'll give this AI stuff another chance. But I'm pretty sure it won't happen.

You should probably compare the same dev using AI vs not using AI. Otherwise you can use that argument for anything.

Notepad is a better IDE than emacs if we compare a really good dev using notepad vs a shitty dev using emacs.

Ok, if I notice other devs suddenly getting 5x as productive I'll give it a try, but so far no such effect has been demonstrated. It seems like a pretty straightforward research question, and you'd think if there was any demonstrable effect the companies selling these things would use such research to market their products. So where is it?

There's a dead body lying there but you won't believe they're actually dead until someone creates a video reenactment of the killer stabbing them and shoves that video in your face?

Speaking of videos, https://youtu.be/opi1s_5Dm-c is probably below your level and it's kinda long.

I don't get it. What is the dead body thing about?

There are people building things and making money using these new techniques, but because they haven't stopped to produce research studies and white papers about their work, you want to presume it doesn't exist.

Yes, without evidence there's no effect. It's telling that nowhere on cursor.ai's website do they appear to claim using their tool will actually make you, or crucially your team, more efficient.

evidencthe absence of evidence doesn't mean there's no effect

There is no trustworthy evidence that I'm aware of to support the claim that "using an LLM IDE makes developers 5x more effective". Do you disagree? Testimonials don't cut it, because individuals are prone to hysteria, delusions, and just making stuff up.

All I'm asking for is a few trustworthy case studies showing unambiguously that it's worth my time and money to pay attention to this genai stuff. That shouldn't be hard for a trillion dollar industry, right? Like show me this isn't just hysterical hype. Don't tell me, show me.

What does it even mean to be 5x more effective? Because that's the thing. Testimonials is where we're at because how do you measure "more effective"? I'm not making any claim on a number, 5x 10x? .2x because you have to go back and fix the shitty code it wrote? We don't have a good way to measure programmer productivity. We all know lines of code is ridiculous. And how do you account for how much brain space it takes up. If the LLM lets someone bang out useful code in between meetings all day, vs needing some truly quiet time for some deep deep work to properly get into the right zone... how do you metricize that and show, what, lines of code and interruptions per hour on a graph to have a line that goes up and to the right?

It turns out that it's actually a really hard problem to measure, so, even for a trillion dollar industry, it is hard.

What did you think of that video?

This is the true litmus test IMO. If LLMs are so great and make everyone so productive, then where are the results? Where are all of the amazing products being released that otherwise would have required 10x the investment? Shouldn't there be _anything_ we can point to that shows that the "productivity needle" is being moved?

The demo is the work. Social media views is where the productivity is being moved. And VCs are paying for this ball to keep rolling hoping to eventually cash out big some day.

AI is an irrational market at the moment and this is not going to change anytime soon.

I don't think it makes everyone so productive, tbh! If you really know what you are doing and are willing to burn tokens, it will really take your work to the next level—provided you don’t work with a niche product or language that the models weren’t trained on.

If I may use an analogy, it's like what sampling is for music producers. The sample is out there—it’s already a beautiful sample, full of strings and keys, a glorious 8-bar loop. Anyone can grab it, but not every producer can sell it or turn it into a hit.

In the end, every hype train has some truth to it. I highly suggest you give it a shot, if you haven't already. I’m glad I did—it really helped me a lot, and I am (unfortunately, financially) hooked on it.

then find a "great" artist willing to cosign that they made something using that AI, vs their own talents.

What are you on about? It's happening everywhere, in all companies. It's literally impossible to miss and how to handle the massive increase in code output is becoming a well-known problem. I think it's interesting how the naysayers claim that it's not happening, yet simultaneously worry about the large amount of low-quality code being produced.

lines of code is not equal to features released. especially when you have to diagnose faulty code in production later. So... low quality code is worse than no code.

It's the opposite. Most people are not are boasting about their productivity improvements but it's everywhere. Unless you work at a company where you're not allowed to use these tools it should be impossible to miss. Even the most hardcore naysayers I know are now using AI tools. The new discourse is whether the massive increase in code output leads to issues or not (I think it does), but claiming it doesn't happen is not a serious take anymore.

I read the whole thing, top to bottom, and by the 80% mark, I still wasn’t sure whether the piece was written in earnest or the sharpest piece of satire I’ve seen in a decade.

On a corporate blog, though, for a product that is competing in this space?

That was the only thing that conclusively gave it away!

Much of this post was spot on — but the blind spots are highly problematic.

In this agentic AI utopia of six months from now:

* Why would developers — especially junior developers — be assigned oversight of the AI clusters? This sounds more like an engineering management role that’s very hands on. This makes sense because the skill set required for the desired outcomes is no longer “how do I write code that makes these computers work correcty” and rather “what’s the best solution for our customers and/or business in this problem space.” Higher order thinking, expertise in the domain, and dare I say wisdom are more valuable than knowing the intricacies of React hooks.

* Economically speaking what are all these companies doing with all this code? Code is still a liability, not an asset. Mere humans writing code faster than they comprehend the problem space is already a problem and the brave new world described here makes this problem worse not better. In particular here, there’s no longer an economic “moat” to build a business off of if everything can be “solved” in a day with a swarm of AI agents.

* I wonder about the ongoing term scaling of these approaches. The trade off seems to be extremely fast productivity at the start which falls off a cliff as the product matures and grows. It’s like a building that can be constructed in a day up to a few floors but quickly hits an upper limit as your ability to build _on top of_ the foundational layer of poorly understood garbage.

* Heaven help the ops / infrastructure folks who have to run this garbage and deal with issues at scale.

Btw I don’t reject everything in this post — these tools are indeed powerful and compelling and the trendlines are undeniable.

I think a lot of people are going to be surprised at how fast "vibe coding" with agents replaces a lot of traditional software engineering. Especially for non-critical software, but eventually where safety matters too since AI can generate tons of test cases.

There is so much hype around agents but I am still thoroughly unimpressed.

They’re fine at basic tasks, but nothing more.

And without any traditional software engineers, who's going to check that those test cases actually do anything useful and verify the important properties of the system? It doesn't matter how many unit tests your Therac-25 has if none of them test the thing that matters.

Well theoretically you can also do E2E testing where an AI agent also clicks around a user-interface and takes screenshots, which are then fed back into it.

That's just another type of test, the question was who's going to check that the tests are actually testing the correct things. Maybe look up the Therac-25 that was mentioned.

I still have to get used to the fact that a (sourcegraph.com) /item may contain (steve-yegge.blogspot.com) content

I'm going to guess this is satire unless sourcegraph only has "junior" developers, on "junior" pay.

The article tells me something unfortunate about the appropriateness about ever buying software from this person; based on how they write.

As always, citation needed.

(Also, grain of salt required, because this is a blatant marketing post.)

Look, I've been hearing "the models will get better and make these core problems go away" since it become common to talk about "the models" at all. Maybe they will some day! But also, and critically, maybe they won't.

You also have to consider the future where some companies spend an additional $50-100k per developer and they DON'T see any of this supposed increase in performance, if these "trust me, it'll happen this time" promises don't come true. This is the kind of bet that can CRATER companies, so it's not surprising to see some hesitation here, a desire to see if the football will be again yanked away.

Plus, and I believe most damningly, this article appears to be engaging in the classic technocratic failure mode: mistaking social problems for technical ones.

Obviously, yes, developers engage in solving technical problems, but that is not all they do, and at the higher level, that becomes the least of what they do. More and more, a good developer ensures that they are solving the RIGHT problem in the RIGHT WAY. They're consulting with managers, (ideally) users, other teams, a whole host of people to ensure the right thing is built at the right time with the right features and that the right sacrifices are being made. LLMs are classically bad at this.

The author dismissively calls this "getting stuck", and handwaves the importance of it away, saying that the engineer will be able to unstuck the model at first (which, if we're putting armies of "vibe coding" junior engineers in charge of the LLMS, who've not had time enough in their career to develop this skill, HOW?), and then makes the classic claim "but the models will get better", and predicts the models will eventually be able to do it (which, if this is an intractable problem with LLMS -- and so far evidence has been leaning this way -- again, HOW?).

Forgive that apalling grammar. I am het up. But note well what I'm doing: I'm asking "should we even be doing this?" Which is something these models a) will have to do well to accomplish what the author insinuates they will, and b) have been persistently terrible at.

I'm going to remain skeptical for now, since it seems that's my one remaining superpower versus these LLMs, and I guess I'm going to need to keep that skill sharp if I want to avoid the breadline in this author's future. =)

LOL, pretty good, he had me going a few times. I'm sure a surprising amount of people will actually take this seriously, showing how ludicrous the situation is right now.

One of the funniest things I’ve read in a while. Also full of some truths. I think learning how to use AI will become a core part of being a dev but I seriously doubt they’ll have anywhere near the competency of solving a problem that a junior engineer has. They can certainly write code like one though.

I really recommend this to anyone reading - if you haven’t tried using cursor or copilot, check them out. It makes writing code less tedious.

> We think the flat hiring right now is just companies signalling that they don’t know what to do yet.

The whole article seems to be written disingenuously for the junior developer audience but this one kinda irked me: the flat hiring is because interest rates are high and has nothing to do with companies figuring out what to do with vibe coding.

On topic, nothing in this article suggests anything fundamentally useful about vibe coding other than it being an easier way to start for juniors and entry-levels. If you are a junior, go ahead and keep vibe coding but also do your best to understand the code you’re given. I strongly suspect that will (continue to) be something that makes people stand out.

Is it just me, or has anyone else noticed that:

1) Cursor has been crashing several times an hour for me recently.

2) Cursor seems to ignore .cursorrules files. I'm using the json format that's supposed to let you filter on file name patterns (although how that works for cross-cutting agent stuff I don't know).

3) Cursor is obsessed with making sketchy iffy defensive code checking for the most recent symptom and trying to guess and shart its way out of it instead of addressing the real problem? And it's extremely hard to talk it out of doing that, I have to keep reminding it and admonishing it to cut it the fuck out, fail instead of mitigate, address the root cause not the symptoms, and stop trying to close the barn door after all the horses have escaped. It's as of it was only trained on Stack Overflow and PHP manual page discussions.

Nothing smells like bullshit quite like a data-less "graph"

Reply on news.ycombinator.com