Code is cheap now, but software isn't | Hacker News Reader

Code is cheap now, but software isn't

chrisgregori.dev

159 pointsfs_software74 comments

I saw someone use the term "orchestration", which seems to be the word for building the software using LLM tools.

It made me think of the conductor, seemingly the most skillless job in the orchestra. All you do is wave the batton, no need to ever play a instrument. If LLMs are doing the hard part (writing code) then we can be the conductor waving the batton.

But of course the visuals are misleading. Being conductor doesn't take the least skill, it takes the most. He hears every instrument individually, he knows the piece intimately, and through his conducting brings a unique expression to a familiar work.

LLMs have made the musician part automated. They'll play whatever you want. No doubt a powerful tool in the hands of a skilled conductor. And a incredible tool for someone who can't play to generate music for themselves.

There's no shortage of "I built it and they won't come" posts here on HN, predating LLMs by decades. Because code has never been the hard part of "software as a business ". LLMs have driven this point home. Code has never been cheaper. Business has never been harder.

Vibecoding is the feeling of coding. It's the same feeling people have when they say they can see the picture in their head, but can't quite draw it.

In other words the blind leading the blind

Or, if you focus on the "slop" aspect of AI, the bland leading the bland ;-)

If you're talking about the original vibe coding, sure.

But there are many ways to apply LLMs in the development flow.

Only specifying features broadly is like a product manager might is definitely highly luck dependent wrt how buggy it will turn out.

But understanding the feature and determining what needs to be done broadly, then ask the LLM to do so and verify after if the resulting change makes sense according to your mental model of the software is definitely not that.

Also, I disagree with your implied message. I frequently struggle to articulate solutions even if I know how they'd work

This should apply to art even more, because art is strongly supported by emotions - and people may know the feeling of the emotion (of the image), but not have an explicit framework for it yet

People are trying to use Vibe Engineering for when you know what needs to be done and how, but are using an LLM as a tool to write the code faster and more efficiently.

The OG definition of Vibe Coding is just playing a client who wants $thing, but doesn't need how to write a line of code.

This is the "og definition" of vibe coding

https://x.com/karpathy/status/1886192184808149383?lang=en

He knows how to code, as such your personal definition does not agree with how the term was coined.

Now, 1 year later and people interpret a lot into the term.

in that context I could accept your perspective to that term, but it's certainly not the original meaning

True, the original-original by karpathy was more Vibe Engineering, but that term ran away from his definition real fast. =)

This "orchestration" software is about people trying to increase productivity by running many instances of a coding agent on the same project, without stepping on each other too much. It doesn't seem to be fully baked yet. A "shared nothing" architecture where you work have each instance work on a distinct project seems simpler if you want to spin more plates.

> LLMs have made the musician part automated. They'll play whatever you want

I like your metaphor even as someone who can be a bit skeptical of the overly broad promises of LLM’s/AI. But I do think this statement is too generous. It implies way too much actual musical ability. It also means that everything I can imagine musically is possible which it just isn’t, as there are limitations just like with real musicians.

If we want to really make the metaphor work, it’s an orchestra full of very informed people who have read a lot about music and have an idea of what their instrument should sound like and can even make whatever they’re holding sound like the appropriate instrument most of the time sort of. With our direction, our “conducting,” their success goes up.

But ultimately: they aren’t real musicians, they aren’t holding the right instruments, and they haven’t actually been taught how to read music. They are just often good at sort of making it work in a way that approximates what we want.

Yes, I agree, the musicians aren't mystros. And their technique could use improvement.

But I think the analogy holds (from an output point of view), the musicians will continue to improve, and some sections play better than others. The overall effect is "pleasing" although perhaps not concert quality.

Yeah that I definitely agree with. Appropriately emphasizes how important your input is.

But also the market for conductors is very small. There are 100 musicians but only one conductor in an orchestra.

So what you wrote does not bode well for the profession.

True. The market for "coders" will likely go down. The market for developers will remain the same (which means go up, the market for developers has always gone up.)

One likely impact of LLM coding is a huge increase in the amount of custom software as it will become cheaper. That could lead to more work, not less, just with a different skill set.

So then why did MIDI not replace musicians and conductors many decades ago? Why do we even bother thinking in terms of sheet music, or programs in terms of code?

It kinda sorta did. Decades ago, all music was played by live players. Today, there are lots of albums, lots of background music on television, radio, etc., that is made mostly or entirely using MIDI-controlled virtual instruments. No longer do you need to book an actual chamber orchestra for a little 30-second spot on some cooking show.

So those musicians are no longer getting booked for that bit of music. Instead, one person produces it in their home studio. But, there’s now an industry for creating software tools that support that workflow, and there are a lot more opportunities for such music than there used to be. The amount of music used in background spots on television is astounding.

Things changed. Some jobs diminished (studio players?) or went away altogether (music copyists?). But new work came into existence.

Yeah my point was that there's not much existing software within a business that's the equivalent of an ad jingle, unless you really split hairs and start counting excel macros or something.

Will there be new software like that? Maybe, but you'll never hear about it. Not only because it's throwaway code, but because the best interface is probably no code at all. The chatbot will instead spin up a VM behind the scenes and never even show the code it generated unless you dig for it.

On the other hand, if there is budget available, like on real movies and bigger television projects, real musicians are still used. And across the board, except for musical styles that explicitly call for electronic sounds, most people agree that using live players would be preferred if only they had the time and money.

I wonder if there’s any parallel to that in software?

I think the equivalent to "live players" are frontend app devs.

It's a deeply unpopular opinion around here, but if a human has to interact with anything that's where most of the effort and budget is going to go. They're still the "rock stars".

That skill set is not merely writing code. It's more about collaboration with all the stakeholders and making a ton of deliberate decisions and compromises. It doesn't matter how "good" an LLM is at writing code for the web. That's subjective, and that's my point. We've had all kinds of no-code solutions for a very long time.

An experienced frontend dev is necessary when the project isn't just for other devs or internal use.

I agree, and the corollary of this would be that the most senior engineers (who might be at a staff/principal level at a given company) who have the most amount of domain knowledge, deep understanding of various software architectures, and a product/customer oriented mindset may stand to benefit the most from AI-assisted coding, despite some narratives being peddled around by executives that they could do without senior engineers.

Unfortunately, for junior engineers the CS path has likely become more arduous, and we'd probably see something more of a doctor-like career path for CS students, where they specialize to obtain deeper architectural knowledge, before receiving employment.

> Being conductor doesn't take the least skill, it takes the most.

I agree with the gist of your comment, but I have to push back on the above statement. Conducting an orchestra is a different skillset than playing virtuoso violin, but it is not more difficult or more important. Its just different. The same applies to any leadership or management position. A very skilled orchestra can in even hide the fact that the conductor is a bit crap. Same with a company or sports team performing so well that they overcome the weaknesses of a lackluster manager. Even though they will still often get the credit.

Conductor of an orchestra and current skillet of managing a coding agent feels intuitive to me which addresses understanding the differences in management and skill set. Great analogy.

> Business has never been harder.

Really though? That seems completely wrong.

Conducting an orchestra is not a mechanical activity. In many/most cases, the conductor is doing a live mix of the piece. They control the tempo of the whole orchestea, the volume and accent of different sections and players. They cue in percussion and embellishments. A conductor must know and fully understand the entire score being played. That's upward of a dozen musical threads peing played by 20, 30 people all at once. A conductor's job is to hear each one of these threads individually and simultaneously to shape the music into the final performance they want.

If a conductor's job can be reduced to a metronome, why hasn't it? I've had a credit card sized metronome in my instrument case for 15 years. Most professional musicians carry metronomes. We've had perfectly accurate metronomes for something like 500 years, so why is "conductor" a profession at all?

Can't read this every paragraph ends with it's not x it's y. Just give me the prompt so I can read the real insights you have and not the generated fluff.

I randomly skipped to five different paragraphs and each one ended with a "!x but y" logical statement, just formatted differently most of the time. Crazy how you can't unsee it.

A sibling [dead] comment to mine is a rebuttal to "just post the prompt", where it itself was expanded to several paragraphs that each say nearly nothing, including this gem:

> "That’s not a critique of the writing. It’s a diagnosis"

I miss when people just typed their thoughts concisely and hit send without passing it to an inflater. I'd maybe have a chance of understanding the sibling comment's point.

Now it is a tell but eventually people may natutally start speaking like this!!

This isn't mind control, just language evolution quiety nudged by AI. ;)

I've rather suspected that people are subconsciously adapting their language patterns to those that they hear over and over, and with AI content so prolific online now, it's natural that people are being programmed more and more with those patterns.

We trained a model on human language which is now in the business of itself retaining human language.

Yeah it's becoming increasingly obvious now. The moment I see this "contrast framing" I stop reading.

I read this as 'contrast farming' and like the term better.

That's not just contrast framing. It's contrast farming.

Thanks for saving me from reading it myself.

we just need to send the article back to the LLM to get it synthesized /s

This made me think of:

- use an LLM to compress a blog article into a singular prompt

- Run it through against all the major LLMs to have them expand it back out again

- Diff the original against the generated versions in terms of content/ideas

- Spit out an "entropy ranking".

Again I'm the author.. and I actually wrote this.

Maybe I need to stop reading AI posts and thinking that it's good writing? But will take the feedback on board and try not to over analyse each paragraph - it was a lot messier upon first draft but maybe ready better I guess.

I appreciated the essay, GPTisms aside. The core concept is one that I've felt for a long while, but you articulated it more cleanly than I've been able to. I'd explained it as into "the tasks are replaced, the job is not", but I like your distinction better.

Appreciate it! Glad you enjoyed the read.

Planning on writing a lot more this year.

You really learn how bad everyone's taste is, just how low the bar can be when it comes to the written word. LLMs really can teach us something about ourselves!

The "barrier to entry for building software" has not collapsed, as it was never about "where engineering shifts from writing code to shaping systems". It has always been about understanding the problem to solve and doing so in a provably correct manner.

Another way to reify this is:

  When making software, remember that it is a snapshot of 
  your understanding of the problem.  It states to all, 
  including your future-self, your approach, clarity, and 
  appropriateness of the solution for the problem at hand.  
  Choose your statements wisely.

> People are increasingly building tools to solve a single, specific problem exactly once—and then discarding them. It is software as a disposable utility, designed for the immediate "now" rather than the distant "later."

Yes! This is 100% it.

This is a net good for everyone because it brings basic programming literacy to the masses and culls a lot of junk projects that are littering github or SaaS scams.

It means people can focus on the problems that actually matter.

AI doesn't have any impact on the need for accountable humans to write code.

The scratchpad analogy is so good. Most mature business software is almost literally like a tome of legal documents that have to be edited carefully, but that doesn't have anything to do with the napkin in your pocket.

In a way it's good but as far as energy usage goes, it sucks.

Not only is it taking way more energy to write software now with LLMS than by "hand", now everyone is repeating work many times over to write the same tools.

From a freedom standpoint one could argue is gives the user the most freedom to have what they want and need. But its very bad from an energy efficiency point of view.

We are all fashion designers now. LLM can code the cloth, do the seams, put in zippers, and sow buttons. You pick what creature you are designing for. You study how it moves, where the limits of movement lie, where it needs ventilation, flexibility, extra reinforced knees, fire resistance, a cape for flare (or no cape for safety). We are all Edna Mode. And the best of us can turn the problems we are working on into superheroes.

You model inference provider and any intermediaries get to watch what you’re designing from behind the curtain and copy, train on, or sell the insights if you’re not paying attention.

> ...if you’re not paying attention.

Even if, I would only trust local models.

So many nice blogs showing on HN, and no RSS feed. Seems like most are on github pages, that should be a feature over there.

Author here - I added an RSS feed :) https://www.chrisgregori.dev/rss.xml

I had this in the old version of my site that I refreshed in the last 2 months and forgot to take the RSS feed with me, thanks for the reminder!

If your job is to write code, you are being replaced. If your job is to use technology to solve problems, your job just got a lot more interesting.

And if you make this distinction, you don't understand the operations of the business.

Man. This post reminded me I wanted a Firefox extension for switching between tabs using Q and E. I got it done in like 15 min and moved on.

In other words, yes we have CNC machines and electric saws and whatnot, reliable to a certain degree (you can still injure yourself badly), but it doesn't remove the need of a carpenter, because a carpenter also knows how to make a hammer from scratch even if he doesn't make one in his entire life.

eh, rather, a carpenter knows to swing a hammer in a pinch, doing say framing, even if they mostly use a nailgun.

When you pay for anything you basically exchange money so that someone else take care of a problem you have. Obviously if you are paying you expect the result to be of good quality. Software is no different, AI won't change that fact and engineering is about creating robust solution at the cheapest price. Just my 2 cents.

solving problem should be an obsession rather than building. AI have fueled way too many builders while edge cases and lifecycle maintainence of the code is more of an afterthought

If building is easy enough, it's often OK that it's not perfect because you can just as easily fix problems as they arise if it's just for your own personal use. And making software just for your own personal use can now actually make sense wheras it would more often be a complete waste of time to do by hand.

> The contrast with the traditional SaaS model is stark.

It looks like LLM-supported coding becomes the new SaaS. The more things change, the more they stay the same.

The cost of reviewing code is not cheap now.

Turns out taste and judgement still matter. Weird.

This is a massive shift.

For the love of all that is holy, I cannot read another 5 page AI post that could've been like 200 words. Just make it a paragraph or two and write using your brain, people. Does everything have to be ran through an AI? I'm sure there's some decent ideas in here, but I'm not wasting my time reading this slop.

you won't get clicks and you can't build brand with it. sad but true.

but i don't understand, how is more words better? people already barely read anything. how is writing 4 pages of slop a better brand builder? wouldn't succinct articles be better?

Keep in mind a gaggle of CC/Opus 4.5 agents is the worst this tech will be. How long ago did we hail 3.5 as the best? ATM, we are counting in months and weeks and the frogs are still leaping, and for all the naysayers, this has not yet slowed down.

Will the bubble pop and wipe out progress? Will this be a boon or bust for humanity? Who knows. In any case, the future's so bright, I gotta wear shades ;).

This post is AI generated slop.

You're absolutely right!

I thought it seemed like an especially good read!

What makes you think it is?

A title per paragraph (slight exaggeration), half of the form The X, The Y, The Z. Every section ends with "it's not x; it's y" contrast framing.

But really the only issue is it's monotone linkedin still insight fluff and you can't tell where the prompt ends and the LLM crap begins. I expect something interesting was put into the LLM, but the LLM has destroyed the author's ability to communicate it with me effectively. Everything is overinflated to the same level of importance and I can't tell what the author actually cared about expressing.

I'm the author.. and I actually wrote this so maybe I need to stop reading AI posts and thinking that it's good writing?

> The barrier to entry has effectively collapsed.

Google, Apple, Meta, X, Bluesky, Shopify, Stripe and all the big software companies must be really shaking in their boots for disruption against the army of vibe coders. /s

(They are actually laughing at all of them)

Why would any big software company need to care? There are so many small businesses with unique problems with no current off-the-shelf software solutions because they've always been too niche to justify the time and expense of bespoke development. Now that door is open. Big software companies can keep servicing big businesses and mass markets, while opportunities abound for anyone else willing to innovate on smaller problems. Not everything needs to be built to scale.

What a random set of companies to choose. You'd probably need to think critically about each one of those when assessing the accuracy of your statements.

> What a random set of companies to choose.

All of the mentioned named companies have network effects, distribution and trust.

Not quite easy to copy. Disposable LLM gen'd code without users is cheap, which is the point of the article.

This is my 3rd gold rush in 6 years.

> They are laughing at them.

Yes but not for the reason you think - more that those are the future customers. If you look closely most are pivoting slowly away from software and shifting more to AI + hardware. The slow layoffs and pivoting that capital to infra shows this. All that "vibed" software needs to run somewhere. Also the models that generate and also power all that software need compute which comes from somewhere.

If I can:

- Have large margin compute since GPU's, power, data centre, etc setup is expensive AND

- Models that outperform models you can have at home.

- Vibed software that derives a lot of functionality from the AI compute and wants to be hosted on compute.

The big companies are pivoting away from software to being more infrastructure like for the democratized software that is projected to be made. They will be fine but in 10 years they will be more cloud hyperscalers, AI compute agents, etc than software businesses. Any software they write will be more to package up their compute as higher margin products.

None of this IMV gives any hope to current SWE's.

Reply on news.ycombinator.com