“If it takes longer to explain to the system all the things you want to do and all the details of what you want to do, then all you have is just programming by another name,”
I think this is going to make the difference between junior and senior engineers even more drastic than it is today. It's really hard to know what/how to even describe real problems to these tools, and the people who invest the most in their tooling now, are going to be most successful. It's hard for someone who hasn't designed a large codebase already to do this in an ai native way where they don't have the experience of abstracting at the right level and things like that.
Today's equivalent, I've often found some of the best engineers I know have insane setups with nvim or emacs. They invest in their tool chain, and are now bringing AI into.
That quote really perfectly encapsulates the challenge with these tools. There is an assumption that inherently code is hard to write and so if you could code in natural language it would save time. But code isn’t actually that hard to write. Sure some people are genuinely bad at it just like I’m genuinely bad at drawing but a bit of practice and most people can be perfectly competent at it.
The hard part is the engineering. Understanding and breaking down the problem, and then actually solving it. If all we gain out of these tools is that we don’t have to write code by hand anymore they are moderately useful but they won’t really be a step change in software development speed.
It's not too different in my opinion from the skills need to build complicated machinery like Boeing 747s despite how much Wallstreet and PHBs want to believe it's fungible. Having competent experienced engineers on the ground level watching these processes and constantly revising and adapting to everything from personnel, material, or vendor changes is so far irreplaceable.
Maybe if we get super AGI one day. Even then I suspect that from a thermodynamics perspective that might not be cost effective as you often need localized on site intelligence.
It's an interesting question but I bet humans combined with AI tooling will remain cost competitive for a long time barring leaps in say quantum compute. After all organic brains operate at the atomic level already and were honed in an extremely competitive environment for billions of years. The calories and resources required to create highly efficient massively powerful neural compute had incredibly thin resource "margins" with huge advantages for species to utilize.
You hit the nail on the head too. Coding itself is very easy for anyone halfway decent in this career — and yet there were a ton of people in CS101 and even in later courses who struggled with things like for loops. It was very hard for them to succeed in this career.
What’s hard is coming up with the algorithm/system design, making the right choices that will scale and won’t become a maintenance nightmare, etc. And yeah, after almost a decade, I have picked up enough I can at least write an outline of a solution that will work alright. But there are still so many tricky edge cases and scaling problems that make it hard to turn “alright” into “really good!”
Sure, AI can help… but it mostly helps with greenfield projects. It doesn’t know about the conversations on slack & jira from a year ago. It doesn’t know about the dozens of other systems and ways the project interacts with other parts of the business. It doesn’t know why whatever regurgitated approach won’t be a good fit for our specific use case. And elaborating all of that detail is not easy! Part of what makes you a good employee is the shit you picked up over the past several months & years that is joe ingrained in your mind when you start working on new projects.
[dead]
> some of the best engineers I know have insane setups with nvim or emacs. They invest in their tool chain, and are now bringing AI into.
I find this likely, but totally irrelevant to their success now and in the future. AI tools are as much of a trivial choice as any other text editor features. The resulting time spent and code quality are the same regardless of personal preferences.
> It's really hard to know what/how to even describe real problems to these tools
I would argue that if you can't describe the problem in plain language then you don't have a very good chance of solving it with code or otherwise.
Personally I find that the act of describing the problem will often reveal a good solution...then it's just a matter of seeing if the LLM agrees with me or if it has a difference idea (for better or worse).
As software grows the problems go such that the effort to completely describe the necessary changes in plain language is often not all that much shorter than the code. Especially if you have a really really good autocomplete for the boilerplate parts of the code. So LLMs being really good at the tedious autocomplete makes LLMs have less marginal utility for the "write the whole thing" for certain types of work.
But the sneakier part of the problem is that as business rules get more complex it's usually harder to completely describe the problem in plain language than in a more formally specified language. For instance, plain-language "apply the user's promo code" doesn't capture the nasty if/else tree you might hit when you're deep in the codebase and see that there's already a bunch of restrictions on other types of promo codes and the product manager didn't think about which of those restrictions should apply to this new promo code. And at this point you're gonna need to use plain language to describe and refine the problem with the product manager - but if you instead were relying on an LLM to turn your short sentence into the right code, it might pick the wrong thing when it comes to modifying that existing code.
Loading
> ...describe the problem in plain language...
Oh, I get it. They're saying you should be able to write it in C.
[Jokes aside, I would be interested in hearing from bridge builders, aerospace engineers, or nuclear scientists. Pretty sure they're using math and not 'plain language'.]
So the author is providing some personal annotations and opinions on a summary of a “new paper” which was actually published five months ago, which itself was a summary of research with the author’s personal annotations and opinions added? These are exactly the kind of jobs that I want AI to automate.
It's more likely that AI will let more people "write" random blogs and articles about things they haven't sufficiently actually researched... you're gonna get more spam, not less.
The solution is more AI, in the browser to tell you this link is to a rubbish AI blog, AI in the mail client to tell you which email content to ignore and AI in the fridge, in the bathroom. Till each person is so tightly locked in a group bubble they have no idea how colourful the world is and how many different personalities exist that for the most bring something to the table
Having your opinions and personal remarks automated by “AI” sounds really smart.
Both humans and coding agents have their strengths and weaknesses, but I've been appreciating help from coding agents, especially with languages or frameworks where I have less expertise, and the agent has more "knowledge", either in its weights or in its ability to more quickly ingest documentation.
One weakness of coding agents is that sometimes all it sees are the codes, and not the outputs. That's why I've been working on agent instructions/tools/MCP servers that empower it with all the same access that I have. For example, this is a custom chat mode for GitHub Copilot in VS Code: https://raw.githubusercontent.com/Azure-Samples/azure-search...
I give it access to run code, run tests and see the output, run the local server and see the output, and use the Playwright MCP tools on that local server. That gives the agent almost every ability that I have - the only tool that it lacks is the breakpoint debugger, as that is not yet exposed to Copilot. I'm hoping it will be in the future, as it would be very interesting to see how an agent would step through and inspect variables.
I've had a lot more success when I actively customize the agent's environment, and then I can collaborate more easily with it.
For me it's simple: even the best models are "lazy" and will confidently declare they're finished when they're obviously not, and the immensely increased amount of training effort to get ChatGPT 5's mild improvements on benchmarks suggests that that quality won't go away anytime soon.
Sounds like it's partially about a nuanced trade-off. It can just as well be too eager and add changes I didn't ask for. Being lazy is better than continuing on a bad path.
There's a long distance between "nuanced behavior" and what it actually does now, which is "complete 6 items of an explicit 10-item task list and then ask the user again if they want to continue".
gpt-5 is extremely cheap, what makes you think they couldn't produce a larger, smarter, more expensive model?
gpt-5 was created to be able to service 200m daily active users.
> what makes you think they couldn't produce a larger, smarter, more expensive model?
Because they already did try making a much larger, more expensive model, it was called GPT-4.5. It failed, it wasn't actually that much smarter despite being insanely expensive, and they retired it after a few months.
Loading
I'm fatigued by these articles that just broadly claim AI can't code because its painting a broad stroke against a widely diverse use of AI for different stacks.
It's horribly outdated way of thinking that an singular AI entity would be able to handle all stacks all problems directed at it because no developer is using it that way.
AI is a great tool for both coders and artists and these outlandish titles that grab attention really seem to be echo chambers aimed at people who are convinced that AI isn't going to replace them which is true but the opposite is also true.
A lot of comments here seem to be similar. I see people claiming that AI has all but taken over doing their work for them, and others claiming that it's almost useless. But usually, nobody even briefly mentions what the work is (other than, presumably, something related to programming).
I imagine there's a big difference in using AI for building, say, an online forum vs. building a flight control system, both in terms of what the AI theoretically can do, and in terms of what we maybe should or should not be letting the AI do.
Yeah. I use it for analytics/dataviz stuff (which involves a lot of python to run spark jobs, glue different APis to get some extra column of data, making png or svg pictures,, making D3 based web sites in html and JavaScript). That all works pretty well.
I also write high performance Go server code, where it works a lot less well. It doesn't follow my rules for ptr APIs or using sync mutexes or atomic operations across a code base. It (probably slightly older version than SOTA) didn't read deep call chains accurately for refactoring. It's still worth trying but if that was my sole work it would probably not be worth it.
On the other hand for personal productivity, emacs functions and config, getting a good .mcp.json, it is also very good and generates code that partakes in the exponential growth of good code. (Unlike data viz where there is a tendency to build something and then the utility declines over time).
I can confidently state that for CRUD web apps, its truly over as in those jobs are never going to command the same wages it once used to.
With the recent models its now encroaching similarly on all fronts, I think the next few iterations we'll see LLM solidify itself as a meta-compiler that will be deployed locally for more FCS type systems.
At the end of the day the hazards are still same with or without AI, you need checks and bounds, you need proper vetting of code and quality but overall it probably doesn't make sense to charge an hourly rate because an AI would drastically cut down such billing schemes.
For me "replacement" is largely a 70~80% reduction in either hourly wages, job positions or both and from the job market data I see it can get there.
Loading
Well, AI really can't code any more than a compiler can. They all require a human to write the original code, even the machine does translate it into other code.
And until the day that humans are no longer driving the bus that will remain the case.
You can say generate a c program that uses gcc 128 bit floats and systematically generates all quadratic roots in order of the max size of their minimal polynomial coefficients, and then sort them and calculate the distribution of the intervals between adjacent numbers, and it just does it. That's qualitatively different from the compilers I have used. Now I was careful to use properly technical words to pull in the world of numeric computation and c programming. But still saved me a lot of time. It was even able to bolt on multithreaded parallelism to speed it up using c stuff I never heard of.
Loading
Using a calculator won't make you a mathematician, but a mathematicians with a calculator can show you amazing things.
Calculators wont give you completely wrong results, not even once, where "AI" does that way too often. If calculators did too, mathemeticians simply would not use them.
Most specifically, random wrong results. Some calculators have issues with rounding, but if you understand those issues, it's consistent.
Imagine driving your car, you turn right, but today turning right slams on the brakes, and 10 people rear end you! That's current AI.
Yes in terms of raw LLM, but with some tools or a MCP the “AI” will never be wrong.
Loading
Loading
Nope, that's not the reason. It's because it's just a query that probabilistically creates a garden path of tokens out of a compressed form of the training data, requiring a real coder to evaluate which parts of are useful.
Amazing how someone writing for an IEEE website can't keep their eyes on the fundamentals.
I don't like the phrase "real coder". It's not clear what it means.
I really like the word "assistant" for what we have today. The AI code assistant tools available today, like Claude Code and GitHub Copilot, can't replace humans in doing software development. Not even close. But they are often useful to human developers, and today, that's the more important measure.
I've been spending time with various AI tools, especially Claude Code and GitHub Copilot. They're amazing one minute, and they make bone-headedly bad recommendations the next. It takes effort to learn how to create good prompts, and if you want the results to be good, you have to review and critique the results. I'm particularly concerned about security. They're definitely happy to write insecure code. If you know what you're doing, prompt them well, and review their results, you can get good results.
I don't know if they'll ever reach "full autonomy". They don't need to get there to be useful.
> I don't know if they'll ever reach "full autonomy". They don't need to get there to be useful
They do, unless you just want to babysit them all the time
Maybe babysitting a bad machine sounds great to you, it sounds like the torment nexus to me
LLMs are not ... well ...
I've just tried to get ChatGPT to help me generate a plumbing part that I will 3D print. Bit bizarre but the part in question (which is currently cracked) holds the other part in place and that provides the water seal with two O rings.
I got so far with OpenSCAD. I started off with describing a part that Chat came back with as a "cup". Fair enough and so far so magical. Then I asked it to put a helical screw thread on the inside of the part. It suggested a popular library for this (BOSL2) and then got things a bit wrong.
This is how it goes: Chat int al will get you close but as things get more specialized it will start to go off piste and tumble.
I own a couple of slide rules. If I read them incorrectly then I will get a calculation wrong. Chat n that are just very fancy slide rules and I still find them fabulous.
I'm tired of the anthropomorphization marketing behind AI driving this kind of discussion. In a few years, all this talk will sound as dumb as stating "MS Word spell checker will replace writers" or "Photoshop will replace designers".
We'll reap the productivity benefits from this new tool, create more work for ourselves, output will stabilize at a new level and salaries will stagnate again, as it always happens.
Generative AI is replacing writers, designers, actors, ... it is nothing like just a spell checker or Phtoshop.
Everyday, I see ads on YouTube with smooth-talking, real-looking AI-generated actors. Each one represents one less person that would have been paid.
There is no exact measure of correctness in design; one bad bit does not stop the show. The clients don't even want real art. Artists sometimes refer to commercial work as "selling out", referring to hanging their artistic integrity on the hook to make a living. Now "selling out" competes with AI which has no artistic integrity to hang on the hook, works 24 hours a day for peanuts and is astonishingly prolific.
> Everyday, I see ads on YouTube with smooth-talking, real-looking AI-generated actors. Each one represents one less person that would have been paid.
Were AI-generated actors chosen over real actors, or was the alternative using some other low-cost method for an advertisement like just colorful words moving around on a screen? Or the ad not being made at all?
The existence of ads using generative AI "actors" doesn't prove that an actor wasn't paid. This is the same logical fallacy as claiming that one pirated copy of software represents a lost sale.
Loading
Loading
Loading
Loading
Loading
I agree with your sentiment. But where I struggle is: to what degree do each of those ads “represent one less person who would have been paid” versus those that represent one additional person who would not be able to afford to advertise in that medium.
Of course that line of reasoning reduces similar to other automation / minimum wage / etc discussions
Loading
Loading
YouTube has the lowest quality ads of any online platform I use by several orders of magnitude. AI being used for belly fat and erectile dysfunction ads is not exactly good for its creative reputation
Loading
There's a difference between taking one thing and putting something else in it's spot, and truly REPLACING something. Yes, some ads have AI generated actors. You know because you can tell because they're "not quite right", rather than focusing on the message of the ad. Noticing AI in ads turns more people off than on, so AI ads are treated by a lot of people as an easy "avoid this company" signal. So those AI ads are in lieu of real actors, but not actually replacing them because people don't want to watch AI actors in an ad. The ad ceases to be effective. The "replacement" failed.
Loading
> Generative AI is replacing writers, designers, actors, ... it is nothing like just a spell checker or Phtoshop.
For cheap stuff it’s true. However, nobody wants to watch or listen generated content and this will wear thin aside from the niche it’ll take hold of and permanently replace humans.
“ Everyday, I see ads on YouTube with smooth-talking, real-looking AI-generated actors. Each one represents one less person that would have been paid.”
The thing is that they would not have paid for the actor anyway. It’s that having an “actor” and special effects for your ads cost nothing, so why not?
The quality of their ads went up, the money changing hands did not change.
Anecdata: I know writers, editors, and white collar non-tech workers of all kinds who use AI daily and like it.
When GPT3.5 first landed a lifelong writer/editor saw a steep decrease in jobs. A year later the jobs changed to "can you edit this AI generated text to sound human", and now they continue to work doing normal editing for human or human-ish writing while declining the slop-correction deluge because it is terrible work.
I can't help but see the software analogy for this.
And as people get more used to the patterns of AI it’s getting called out more and more.
> Each one represents one less person that would have been paid
or equally, one more advert which (let's say rightly) wouldn't have been made.
seriously though, automation allows us to do things that would not have been possible or affordable before. some of these are good things.
I'm not a "real coder" either, but it sounds like the "No True Scotsman" trap when people say, “AI can’t be a real coder,” and then redefine “real coder” to mean something AI can’t currently do (like full autonomy or deep architectural reasoning). This makes the claim unfalsifiable and ignores the fact that AI already performs several coding tasks effectively. Yeah, I get it, context handling, long-horizon planning, and intent inference all stink, but the tools are all 'real' to me.
That's based on the assumption models would not soon cross that treshold of autonomy and self-reflection that suddenly makes an escalating number of jobs (with cheap humanoids, even physical) automatable for ridiculous pricing. Even if this isn't certain, likelihood could be considered quite high and thus we urgently need a public debate / design process for the peaceful, post-commercial, post-competitive, open-access post-scarcity economy some (RBE / commoning community) have been sketching for years and years. Seems this development defies most people's sense of imagination - and that's precisely why we need to raise public awareness for the freedom and fun OPEN SOURCE EVERYTHING & Universal Basic Services could bring to our tormented world. 2 billion without access to clean water? we can do much better if we break free from our collective fixation on money as the only means and way to deal with things ever.
You say it as a joke, but spell check has replaced certain tiers of editors. And Photoshop has replaced certain tiers of designers.
Not a joke.
Proofreaders still exist, despite spell checker. Art assistants still exist, despite Photoshop. There's always more work to do, you just incorporate the new tools and bump the productivity, until it gets so commoditized it stops being a competitive advantage.
Saying AI "replaces" anyone is just a matter of rhetoric to justify lower salaries, as always.
Bad ones
People are stupid, always have been - took thousands of years to accept brain as the seat of thought because “heart beat faster when excited, means heart is source of excitement”.
Heck, people literally used to think eyes are the source of light since everything is dark when you close them.
People are immensely, incredibly, unimaginably stupid. It has taken a lot of miracles put together to get us where we are now…but the fundamentals of what we are haven’t changed.
> took thousands of years to accept brain as the seat of thought because “heart beat faster when excited, means heart is source of excitement”
So what you are saying is that beings without a central nervous system cannot experience "excitement"?
or perhaps the meaning of too many words has changed, and their context. When Hippocrates claimed that the brain was an organ to cool the blood, perhaps he meant that we use our thought to temper our emotions, i.e. what he said agrees with our modern understanding.
However, many people read Hippocrates and laugh at him, because they think he meant the brain was some kind of radiator.
Maybe because we stopped talking about "excitable" people as being "hot-blooded"
Loading
You're confusing ignorance with stupidity. People at the time were coming to the best conclusions they could based on the evidence they had. That isn't stupid. If humans were truly "incredibly, unimaginably stupid" we wouldn't have even gotten to the point of creating agriculture, much less splitting the atom. We didn't get here through "miracles," we got here through hard work and intelligence.
Stupid is people in 2025 believing the world is flat and germ theory is a hoax. Ignorance becomes stupidity when our species stands on the shoulders of giants but some people simply refuse to open their eyes.
Loading
In a few years AI will have progressed a fair bit in a way that MS spell checker didn't.
> tired of anthropomorphization
The thing is trained on heaps and heaps of human output. You better anthropomorphize if you want to stay ahead of the curve.
> all this talk will sound as dumb as stating "MS Word spell checker will replace writers" or "Photoshop will replace designers".
You cannot use just a spell checker to write a book (no matter how bad) or photoshop (non-AI) plugins to automatically create meaningful artwork, replacing human intervention.
Business people "up the ladder" are already threatening with reducing the workforce and firing people because they can (allegedly) be replaced by AI. No writer was ever threatened by a spellchecker.
Hollywood studio execs are putting pressure on writers, and now they can leverage AI as yet another tool against them.
I'm tired of all the "yet another tool" reductionism. It reeks of cope.
It took under a decade to get AI to this stage - where it can build small scripts and tiny services entirely on its own. I see no fundamental limitations that would prevent further improvements. I see no reason why it would stop at human level of performance either.
There’s this saying that humans are terrible at predicting exponential growth. I believe we need another saying, those who expect exponential growth have a tough time not expecting it.
It’s not under a decade for ai to get to this stage but multiple decades of work, with algorithms finally able to take advantage of gpu hardware to massively excel.
There’s already feeling that growth has slowed, I’m not seeing the rise in performance at coding tasks that I saw over the past few years. I see no fundamental improvements that would suggest exponential growth or human level of performance.
Loading
> ... entirely on its own
ok, ok! just like you can find for much less computation power involved using a search engine, forums/websites having if not your question, something similar or a snippet [0] helping you solve your doubt... all of that free of tokens and companies profiting over what the internet have build! even FOSS generative AI can give billions USD to GPU manufacturers
[0] just a silly script that can lead a bunch of logic: https://stackoverflow.com/questions/70058132/how-do-i-make-a...
You can’t see any bottlenecks? Energy? Compute power? Model limitations? Data? Money?
Loading
So maybe the truth is somewhere in between - there is no way AI is not going to have a major societal impact - like social media.
If we don't see some serious fencing, I will not be surprised by some spectacular AI-caused failures in the next 3 years that wipe out companies.
Business typically follows a risk-based approach to things, and in this case entire industries are yolo'ing.
> I see no fundamental limitations
How about the fact that AI is only trained to complete text and literally has no "mind" within which to conceive or reason about concepts? Fundamentally, it is only trained to sound like a human.
Loading
your comment reeks of hype. no evidence whatsoever for your prediction, just an assertion that you personally don't see it not coming true
It took closer to 100 years for AI to get to this stage. Check out: https://en.wikipedia.org/wiki/History_of_artificial_intellig...
I suspect once you have studied how we actually got to where we are today, you might see why your lack of seeing any limitations may not be the flex you think it is.
When are you starting time from? AI has been a topic of research for over 70 years
We see massive initial growth followed by a slowdown constantly.
There is zero reason to think AI is some exception that will continue to exponentially improve without limit. We already seem to be at the point of diminishing returns. Sinking absurd amounts of money and resources to train models that show incremental improvements.
To get this far they have had to spend hundreds of billions and have used up the majority of the data they have access to. We are at the point of trying to train AI on generated data and hoping that it doesn’t just cause the entire thing degrade.
> I see no fundamental limitations that would prevent further improvements
How can you say this when progress has so clearly stagnated already? The past year has been nothing but marginal improvements at best, culminating in GPT-5 which can barely be considered an upgrade over 4o in terms of pure intelligence despite the significant connotation attached to the number.
Loading
>> It reeks of cope.
haha, well said, I've got to remember that one. HN is a smelly place when it comes to AI coping.
Loading
There is so much more than coding, it’s the business rules, the constraints held in other parts of the company, the undocumented code that holds a company with no written context … you start a project from scratch ? AI can help you but you better be highly specific because hallucinations and errors will be there along the way. Now for the hundreds or billion of loc of private source code softwares that holds thousands of companies with not tightly formatted context and most of it just in some people’s memory … how would AI replace that ? Try let an AI rewrite a full ERP responsible of business logic with costly consequences on business and money … It will be just a tool, where sometimes we loose time and most time save a bit of time if we use it well
I think everyone is looking for back and white switches. Either coding agents are writing your code or they aren't. Humans will always be in the mix in some form, but the amount and skills they use is going to be radically different as time goes on.
I personally haven't written any significant code by hand since claude code landed. I also have a high tolerance for prompting and re-prompting. Some of my colleagues would get upset if it wasn't mostly one shotting issues and had a really low tolerance for it going off the rails.
Since gpt-5-high came out, I rarely have to re-prompt. Strong CI pipeline and well defined AGENTS.md goes an incredibly long way.
I can't help thinking that a big factor is how bad our tools were before. There are many leaking abstractions, boilerplate code and cruft. Dealing with that fills a sizeable percent of a programmer's job.
These tools seem great because they are less sensitive than humans to the mess and lift us over the tedious work. But at the same time, they're giving us an excuse to not fix what needed to be fixed and, in doing so, they're adding more crap to the heap.
Maybe what we need is forcing the tools to build on a simpler base, so we can keep an understanding of the results.
There's been a strong theme recently here on HN of confusing programming (the act of writing code to meet specifications) and Engineering(the writing of specifications, and the oversight of said process, along with overview of testing).
AI is definitely not ready for an Engineering role. My recent experience with ChatGPT5(Preview) via Visual Studio Code tells me that it might perform acceptably as a junior programmer. However, because I'm an old retired self taught programmer who only ever managed 1 other programmer, I lack the experience to know what's acceptable as a Junior Programmer at FAANG and elsewhere.
> There's been a strong theme recently here on HN of confusing programming (the act of writing code to meet specifications) and Engineering(the writing of specifications, and the oversight of said process, along with overview of testing).
You're making a distinction that might be interesting in some esoteric sense, but that doesn't exist in the industry. When it comes to software, the architects are also the construction crew. There's nothing to confuse if the terms are effectively synonymous.
You're making a distinction that might be interesting in some esoteric sense, but that doesn't exist in the industry.
Sure it exists. Not everywhere, not for all software, but it does exist.
I think this is generally true, and there are SO MANY blog posts and articles. Using something like Claude Code to build an entire SaaS from nothing can seem like magic, but there is still a point where the code is too big and any LLM will lose context.
But there is a "sweet spot" where it's amazing, specifically highly targeted tasks with a specific context. I wanted a simple media converter app that tied into ffmpeg, and I didn't want to install any of the spammy or bloated options I found... so I got Claude to build one. It took about 30 minutes and works great.
I also asked it to update some legacy project, and it fell down a testing a loop where it failed to understand the testing database was missing. Had I not had years of knowledge, I would've looked at the output and suggestions Claude was giving and spent hours on it... but it was a simple command that fixed it. As with all new tech, your milage will vary.
Anecdotes are unreliable, for one, your described use case and the tools you are using suggest you are at a very basic level and unable to extract the full capabilities of the tooling and models which many of use to solve if not complete complex software.
Just be aware that AI is a tool not a replacement but a human apt at AI as a tool will replace the former.
No offense but your experience with AI is fairly primitive if that's where you're at.
Please elaborate.
Share what you built and how you prompted and what you are making from it, and how many tokens you paid to rummage through.
Loading
Loading
I agree, using "chatgpt 5 in visual studio code" screams unfamiliarity with what is current.
just stop yeah?
The article is so banal
My best understanding/explanation of LLM tools is "plausible extension of a context"
That is, you have some context, ie the prompt and any other text, and the LLM produces a plausible continuation or alteration of that prompt and text.
My intuition leads me to a thought like: To progress, the context must compress into a fractal representation.
I feel very confident that someone smarter and MUCH better paid than me is already working on this.
True, but it's catching up fast. A year ago, I used AI for small OS scripts. It worked fine and saved me the time of looking up switches for commands. Now, I can ask it to create a simple game of about 200 lines, and it does a pretty good job of writing bug-free code within a few seconds. It's only going to get better. Even if the tech doesn't improve further, I can see a future where all apps are endlessly configurable.
A big part of my career has been the modification of enterprise software to fit a company's needs. Rarely was any one addition more than a few hundred lines of code. I can see a future where there will be simple options for a non-coder to add to an app.
True, it's not a coder, but that doesn't mean it won't fundamentally change how apps are made and it will reduce the number of master programmers needed. It won't replace all programmers, but it will greatly reduce the number that are needed, which country they work in and the language they use to program apps.
Programming has mainly been a career that requires the individual to understand English. That is changing. I can see a future where code can be created in multiple human languages. Programming was well-paid because relatively few people had the expertise to do it. That won't be the case, and the pay will adjust downward as needed. AI might not be a coder, but it will let many more people become coders. In the future, coding will be in the same pay range as clerical work. Companies will be hiring Programming Clerks rather than Programming Engineers.
> Programming has mainly been a career that requires the individual to understand English.
Disagree, programming is a career where in order to be good you can
1. Break down a big problem into smaller ones, creating abstractions
2. Implement those abstractions one by one to end up with a full program
3. Refactor those abstractions if requirements change or (better) reimplement an abstraction a different way
You do all of this to make complex software digestible by a human, so that they don't have to have the entire system 'in context'.
This prophesied view of software development will mean you end up with code that's likely only maintainable by the model itself.
I can't imagine the vendor lock in of that.... You have the source, but it is in such a state that no human can maintain it?
> I can't imagine the vendor lock in of that.... You have the source, but it is in such a state that no human can maintain it?
It’s much worse than that.
What happens when the erroneous output caused by model blind spots gets fed back into the model?
Those blind spots get reinforced.
Doesn’t matter how small that error rate is (and it’s not small). The errors will compound.
Vendor lock-in won’t matter because it will simply stop working/become totally unrecoverable.
I think you're right that LLMs are democratizing access to coding, but unless and until AI models reach a point where they can say 'no' to their users, the scenario you're imagining ('endlessly configurable apps') will probably lead to software that collapses under its own complexity.
Years ago, I supported a team of finance professionals who were largely quite competent at coding but knew nothing about software engineering. They had thousands of scripts and spreadsheets: they used version control, but kept separate long-lived branches for client-specific variations of different models. There were no tests for anything; half the tools would break when the clocks changed.
They weren't dumb, but their incentives weren't about building anything we might recognize as an engineered application. I suspect something similar will happen turning end users loose with AI.
“If it takes longer to explain to the system all the things you want to do and all the details of what you want to do, then all you have is just programming by another name”
If it's taking you that long to direct the AI, then either you're throwing too small a problem at it, or too big a problem at it, or you're not directing its attention properly.
In a way, your prompts should feel like writing user documentation:
Refactor all of the decode functions in decoder.rs to return the number of bytes decoded
in addition to the decoded values they already return. While refactoring, follow these principles:
* Use established best practices and choose the most idiomatic approaches.
* Avoid using clone or unwrap.
* Update the unit tests to account for the extra return values, and make sure the tests check the additional return values.
When you're finished, run clippy and fix any issues it finds. Then run rustfmt on the crate.I feel you've done a decent job of disguising "you're not approaching it with the right mindset".
It's more "you need to learn how a tool really works in order to use it most effectively"
Talking to a chat bot like you'd talk to a human is a quick way to be disappointed. They don't work that way, despite sometimes sounding like they do.
Loading
My read was along the lines of "you're holding it wrong"
And then for some bizarre reason sometimes it doesn't really work, AI added a bunch of random shit, then you can feel your rage bubbling up in real-time as you have to re-prompt it.
I've yet to have that happen. But then again so far I've only used it for Rust, and it's hard for the AI to maintain hallucinations with such a strict compiler.
Am I the only person who stops reading an article now when I see the use of em dashes?
AI is apparently not ready to take my order at Popeyes either