Coding agents have replaced every framework I used
blog.alaindichiappari.dev267 points by alainrk 10 hours ago
267 points by alainrk 10 hours ago
A significant number of developers and businesses are going to have an absolutely brutal rude awakening in the not too distant future.
You can build things this way, and they may work for a time, but you don't know what you don't know (and experience teaches you that you only find most stuff by building/struggling; not sipping a soda while the AI blurts out potentially secure/stable code).
The hubris around AI is going to be hard to watch unwind. What the moment is I can't predict (nor do I care to), but there will be a shift when all of these vibe code only folks get cooked in a way that's closer to existential than benign.
Good time to be in business if you can see through the bs and understand how these systems actually function (hint: you won't have much competition soon as most people won't care until it's too late and will "price themselves out of the market").
I would argue that it's going to be the opposite. At re:Invent, one of the popular sessions was in creating a trio of SRE agents, one of which did nothing but read logs and report errors, one of which did analysis of the errors and triaged and proposed fixes, and one to do the work and submit PRs to your repo.
Then, as part of the session, you would artificially introduce a bug into the system, then run into the bug in your browser. You'd see the failure happen in browser, and looking at Cloudwatch logs you'd see the error get logged.
Two minutes later, the SRE agents had the bug fixed and ready to be merged.
"understand how these systems actually function" isn't incompatible with "I didn't write most of this code". Unless you are only ever a single engineer, your career is filled with "I need to debug code I didn't write". What we have seen over the past few months is a gigantic leap in output quality, such that re-prompting happens less and less. Additionally, "after you've written this, document the logic within this markdown file" is extremely useful for your own reference and for future LLM sessions.
AWS is making a huge, huge bet on this being the future of software engineering, and even though they have their weird AWS-ish lock-in for some of the LLM-adjacent practices, it is an extremely compelling vision, and as these nondeterministic tools get more deterministic supporting functions to help their work, the quality is going to approach and probably exceed human coding quality.
I agree with both you and the GP. Yes, coding is being totally revolutionized by AI, and we don't really know where the ceiling will be (though I'm skeptical we'll reach true AGI any time soon), but I believe there still an essential element of understanding how computer systems work that is required to leverage AI in a sustainable way.
There is some combination of curiosity of inner workings and precision of thought that has always been essential in becoming a successful engineer. In my very first CS 101 class I remember the professor alluding to two hurdles (pointers and recursion) which a significant portion of the class would not be able to surpass and they would change majors. Throughout the subsequent decades I saw this pattern again and again with junior engineers, bootcamp grads, etc. There are some people no matter how hard they work, they can't grok abstraction and unlock a general understanding of computing possibility.
With AI you don't need to know syntax anymore, but to write the write prompts to maintain a system and (crucially) the integrity of its data over time, you still need this understanding. I'm not sure how the AI-native generation of software engineers will develop this without writing code hands-on, but I am confident they will figure it out because I believe it to be an innate, often pedantic, thirst for understanding that some people have and some don't. This is the essential quality to succeed in software both in the past and in the future. Although vibe coding lowers the barrier to entry dramatically, there is a brick wall looming just beyond the toy app/prototype phase for anyone without a technical mindset.
I can see why people are skeptical devs can be 10x as productive.
But something I'd bet money on is that devs are 10x more productive at using these tools.
I get its necessary for investment, but I'd be a lot happier with these tools if we didn't keep making these wild claims, because I'm certainly not seeing 10x the output. When I ask for examples, 90% its claude code (not a beacon of good software anyway but if nearly everyone is pointing to one example it tells you thats the best you can probably expect) and 10% weekend projects, which are cool, but not 10x cool. Opus 4.5 was released in Dec 2025, by this point people should be churning out year long projects in a month, and I certainly haven't seen that.
I've used them a few times, and they're pretty cool. If it was just sold as that (again, couldn't be, see: trillion dollar investments) I wouldn't have nearly as much of a leg to stand on
Id wager my life savings that devs aren’t even 1.5x more productive using these tools.
Even if I am only slightly more productive, it feels like I am flying. The mental toll is severely reduced and the feel good factor of getting stuff done easily (rather than as a slog) is immense. That's got to be worth something in terms of the mental wellbeing of our profession.
FWIW I generally treat the AI as a pair programmer. It does most of the typing and I ask it why it did this? Is that the most idiomatic way of doing it? That seems hacky. Did you consider edge case foo? Oh wait let's call it a BarWidget not a FooWidget - rename everything in all other code/tests/make/doc files Etc etc.
I save a lot of time typing boilerplate, and I find myself more willing (and a lot less grumpy!!!) to bin a load of things I've been working on but then realise is the wrong approach or if the requirements change (in the past I might try to modify something I'd been working on for a week rather than start from scratch again, with AI there is zero activation energy to start again the right way). Thats super valuable in my mind.
It probably depends on the developer, and how much slop/bugs is willing to be tolerated.
Dead wrong.
Because the world is still filled with problems that would once have been on the wrong side of the is it worth your time matrix ( https://xkcd.com/1205/ )
There are all sorts of things that I, personally, should have automated long ago that I threw at claud to do for me. What was the cost to me? Prompt and a code review.
Meanwhile, on larger tasks an LLM deeply integrated into my IDE has been a boon. Having an internal debate on how to solve a problem, try both, write a test, prove out what is going to be better. Pair program, function by function with your LLM, treat it like a jr dev who can type faster than you if you give it clear instructions. I think you will be shocked at how quickly you can massively scale up your productivity.
Yup, I've already run like 6 of my personal projects including 1 for my wife that I had lost interest in. For a few dollars, these are now actually running and being used by my family. These tools are a great enabler for people like me. lol
I used to complain when my friends and family gave me ideas for something they wanted or needed help with because I was just too tired to do it after a day's work. Now I can sit next to them and we can pair program an entire idea in an evening.
The matrix framing is a very nice and way to put it. This morning I asked my assistant to code up a nice debugger for a particular flow in my application. It’s much better than I would have had time/patience to build myself for a nice-to-have.
I sort of have a different view of that time matrix. If AI is only able to help me do tasks that are of low value, where I previously wouldn’t have bothered—- is it really saving me anything? Before where I’d simply ignore auxiliary tasks, and focus on what matters, I’m now constantly detoured with them thinking “it’ll only take ten minutes.”
I also primarily write Elixir, and I have found most Agents are only capable of writing small pieces well. More complicated asks tend to produce unnecessarily complicated solutions, ones that may “work,” on the surface, but don’t hold up in practice. I’ve seen a large increase in small bugs with more AI coding assistance.
When I write code, I want to write it and forget about it. As a result, I’ve written a LOT of code which has gone on to work for years without touching it. The amount of time I spent writing it is inconsequential in every sense. I personally have not found AI capable of producing code like that (yet, as all things, that could change).
Does AI help with some stuff? Sure. I always forget common patterns in Terraform because I don’t often have to use it. Writing some initial resources and asking it to “make it normal,” is helpful. That does save time. Asking it to write a gen server correctly, is an act of self-harm because it fundamentally does not understand concurrency in Erlang/BEAM/OTP. It very much looks like it does, but it 100% does not.
tldr; I think the ease of use of AI can cause us to over produce and as a result we miss the forest for the trees.
> are only capable of writing small pieces well.
It excels at this, and if you have it deeply integrated into your workflow and IDE/dev env the loop should feel more like pair programing, like tennis, than it should feel like its doing everything for you.
> I also primarily write Elixir,
I would also venture that it has less to do with the language (it is a factor) and more to do with what you are working on. Domain will matter in terms of sample size (code) and understanding (language to support). There could be 1000s of examples in its training data of what you want, but if no one wrote a commment that accurately describes what that does...
> I think the ease of use of AI can cause us to over produce and as a result we miss the forest for the trees.
This is spot on. I stopped thinking of it as "AI" and started thinking of it as "power tools". Useful, and like a power tool you should be cautious because there is danger there... It isnt smart, it's not doing anything that isnt in its training data, but there is a lot there, everything, and it can do some basic synthesis.
Like others are saying, AI will accelerate the gap between competent devs and mediocre devs. It is a multiplier. AI cannot replace fundamentals, at least not a good helmsman with a good rational, detail-oriented mind. Having fundamentals (skill & knowledge) + using AI will be the cheat code in the next 10 years.
The only historical analogue of this is perhaps differentiating a good project manager from an excellent one. No matter how advanced, technology will not substitute for competence.
I view the current tools as more of a multiplier of base skill.
A 1x engineer may become a 5x engineer, but a -1x will also produce 5x more bad code.
Several experiments have shown quality of output at every skill level drops.
In many cases the quantity of output is good enough to compensate, but quality is extremely difficult to improve at scale. Beefing up QA to handle significantly more code of noticeably lower quality only goes so far.
Now run that loop 1000 times.
What does the code /system look like.
It is going to be more like evolution (fit to environment) than engineering (fit to purpose).
It will be fascinating to watch nonetheless.
It'll probably look like the code version of this, an image run through a LLM 101 times with the directive to create a replica of the input image: https://www.reddit.com/r/ChatGPT/comments/1kbj71z/i_tried_th... Despite being provided with explicit instructions, well...
People are still wrongly attributing a mind to something that is essentially mindless.
I mean, if you tell a chain of 100 humans to redraw a a picture i would expect it to go similar, just much faster
"evolution (fit to environment) than engineering (fit to purpose)."
Oh, I absolutely love this lens.
Sure, if all you ask it to do is fix bugs. You can also ask it to work on code health things like better organization, better testing, finding interesting invariants and enforcing them, and so on.
It's up to you what you want to prioritize.
I have some healthy skepticism on this claim though. Maybe, but there will be a point of diminishing returns where these refactors introduce more problems than they solve and just cause more AI spending.
Code is always a liability. More code just means more problems. There has never been a code generating tool that was any good. If you can have a tool generate the code, it means you can write something on a higher level of abstraction that would not need that code to begin with.
AI can be used to write this better quality / higher level code. That's the interesting part to me. Not churning out massive amounts of code, that's a mistake.
"What can we do to reduce the size of the codebase" seems like an interesting prompt to try.
There's an interesting phenomenon I noticed with the "skeptics". They're constantly using what-ifs (aka goalpost moving), but the interesting thing is that those exact same what-ifs were "solved" earlier, but dismissed as "not good enough".
This exact thing about optimisation has been shown years ago. "Here's a function, make it faster". With "glue" to test the function, and it kinda worked even with GPT4 era models. Then came alphaevolve where google found improvements in real algorithms (both theoretical i.e. packing squares and practical i.e. ML kernels). And yet these were dismissed as "yeah, but that's just optimisation, that's easyyyy. Wake me up when they write software from 0 to 1 and it works".
Well, here we are. We now have a compiler that can compile and boot linux! And people are complaining that the code is unmaintainable and that it's slow / unoptimised. We've gone full circle, but forgot that optimisation was easyyyy. Now it's something to complain about. Oh well...
Microsoft will be an excellent real-world experiment on whether this is any good. We so easily forget that giant platform owners are staking everything on all this working exactly as advertised.
Some of my calculations going forward will continue to be along the lines of 'what do I do in the event that EVERYTHING breaks and cannot be fixed'. Some of my day job includes retro coding for retro platforms, though it's cumbersome. That means I'll be able to supply useful things for survivors of an informational apocalypse, though I'm hoping we don't all experience one.
I agree but want to interject that "code organization " won't matter for long.
Programming Languages were made for people. I'm old enough to have programmed in z80 and 8086 assembler. I've been through plenty of prog.langs. through my career.
But once building systems become prompting an agent to build a flow that reads these two types of excels, cleans them,filters them, merges them and outputs the result for the web (oh and make it interactive and highly available ) .
Code won't matter. You'll have other agents that check that the system is built right, you'll have agents that test the functionality and agents that ask and propose functionality and ideas.
Most likely the Programming language will become similar to the old Telegraph texts (telegrams) which were heavily optimized for word/token count. They will be optimized to be LLM grokable instead of human grokable.
Its going to be amazing.
What you’re describing is that we’d turn deterministic engineering into the same march of 9s that FSD and robotics are going through now - but for every single workflow. If you can’t check the code for correctness, and debug it, then your test system must be absolutely perfect and cover every possible outcome. Since that’s not possible for nontrivial software, you’re starting a march of 9s towards 100% correctness of each solution.
That accounting software will need 100M unit tests before you can be certain it covers all your legal requirements. (Hyperbole but you get the idea) Who’s going to verify all those tests? Do you need a reference implementation to compare against?
Making LLM work opaque to inspection is kind of like pasting the outcome of a mathematical proof without any context (which is almost worthless AFAIK).
Will you trust code like this to run airplanes?
Remember, even Waymo has a ton of non-AI code it is built upon. We will still have PyTorch, embedded systems software, etc.
There are certainly people working on making this happen. As a hobbyist, maybe I'll still have some retro fun polishing the source code for certain projects I care about? (Using our new power tools, of course.)
Your assuming that scrum/agile/management won't take this over?
What stakeholder is prioritizing any of those things and paying for it out of their budget?
Code improvement projects are the White Whale of software engineering - obsessed over but rarely from a business point of view worth it.
The costs for code improvement projects have gone down dramatically now that we have power tools. So, perhaps it will be considered more worthwhile now? But how this actually plays out for professional programming is going to depend on company culture and management.
In my case, I'm an early-retired hobbyist programmer, so I control the budget. The same is true for any open source project.
And what happens when these different objectives conflict or diverge ? Will it be able to figure out the appropriate trade-offs, live with the results and go meta to rethink the approach or simply delude itself ? We would definitely lose these skills if it continues like this.
It's nice that AI can fix bugs fast, but it's better to not even have bugs in the first place. By using someone else's battle tested code (like a framework) you can at least avoid the bugs they've already encountered and fixed.
I spent Dry January working on a new coding project and since all my nerd friends have been telling me to try to code with LLM's I gave it a shot and signed up to Google Gemini...
All I can say is "holy shit, I'm a believer." I've probably got close to a year's worth of coding done in a month and a half.
Busy work that would have taken me a day to look up, figure out, and write -- boring shit like matplotlib illustrations -- they are trivial now.
Things that are ideas that I'm not sure how to implement "what are some different ways to do this weird thing" that I would have spend a week on trying to figure out a reasonable approach, no, it's basically got two or three decent ideas right away, even if they're not perfect. There was one vectorization approach I would have never thought of that I'm now using.
Is the LLM wrong? Yes, all the damn time! Do I need to, you know, actually do a code review then I'm implementing ideas? Very much yes! Do I get into a back and forth battle with the LLM when it gets starts spitting out nonsense, shut the chat down, and start over with a newly primed window? Yes, about once every couple of days.
It's still absolutely incredible. I've been a skeptic for a very long time. I studied philosophy, and the conceptions people have of language and Truth get completely garbled by an LLM that isn't really a mind that can think in the way we do. That said, holy shit it can do an absolute ton of busy work.
What kind of project / prompts - what’s working for you? /I spent a good 20 years in the software world but have been away doing other things professionally for couple years. Recently was in the same place as you, with a new project and wanting to try it out. So I start with a generic Django project in VSCode, use the agent mode, and… what a waste of time. The auto-complete suggestions it makes are frequently wrong, the actions it takes in response to my prompts tend to make a mess on the order of a junior developer. I keep trying to figure out what I’m doing wrong, as I’m prompting pretty simple concepts at it - if you know Django, imagine concepts like “add the foo module to settings.py” or “Run the check command and diagnose why the foo app isn’t registered correctly” Before you know it, it’s spiraling out of control with changes it thinks it is making, all of which are hallucinations.
Also >20 years in software. The VSCode/autocomplete, regardless of the model, never worked good for me. But Claude Code is something else - it doesn't do autocomplete per se - it will do modifications, test, if it fails debug, and iterate until it gets it right.
I'm just using Gemini in the browser. I'm not ready to let it touch my code. Here are my last two prompts, for context the project is about golf course architecture:
Me, including the architecture_diff.py file: I would like to add another map to architecture_diff. I want the map to show the level of divergence of the angle of the two shots to the two different holes from each point. That is, when your are right in between the two holes, it should be a 180 degree difference, and should be very dark, but when you're on the tee, and the shot is almost identical, it should be very light. Does this make sense? I realize this might require more calculations, but I think it's important.
Gemini output was some garbage about a simple naive angle to two hole locations, rather than using the sophisticated expected value formula I'm using to calculate strokes-to-hole... thus worthless.
Follow up from me, including the course.py and the player.py files: I don't just want the angle, I want the angle between the optimal shot, given the dispersion pattern. We may need to update get_smart_aim in the player to return the vector it uses, and we may need to cache that info. We may need to update generate_strokes_gained_map in course to also return the vectors used. I'm really not sure. Take as much time as you need. I'd like a good idea to consider before actually implementing this.
Gemini output now has a helpful response about saving the vector field as we generate the different maps I'm trying to create as they are created. This is exactly the type of code I was looking for.
I recently started building a POC for an app idea. As framework I choose django and I did not once wrote code myself. The whole thing was done in a github codespace with copilot in agentic mode and using mostly sonnet and opus models. For prompting, I did not gave it specific instructions like add x to settings. I told it "We are now working on feature X. X should be able to do a, b and c. B has the following constraints. C should work like this." I have also some instructions in the agents.md file which tells the model to, before starting to code, ask me all unclear questions and then make a comprehensive plan on what to implement. I would then go over this plan, clarify or change if needed - and then let it run for 5-15 minutes. And every time it just did it. The whole thing, with debugging, with tests. Sure, sometimes there where minor bugs when I tested - but then I prompted directly the problem, and sure enough it got fixed in seconds...
Not sure why we had so different experiances. Maybe you are using other models? Maybe you miss something in your prompts? Letting it start with a plan which I can then check did definitly help a lot. Also a summary of the apps workings and technical decissions (also produced by the model) did maybe help in the long run.
I'm (mostly) a believer too, and I think AI makes using and improving these existing frameworks and libraries even easier.
You mentioned matplotlib, why does it make sense to pay for a bunch of AI agents to re-invent what matplotlib does and fix bugs that matplotlib has already fixed, instead of just having AI agents write code that uses it.
I mean, the thesis of the post is odd. I'll grant you that.
I work mostly with python (the vast majority is pure python), flask, and htmx, with a bit of vanilla js thrown in.
In a sense, I can understand the thesis. On the one hand Flask is a fantastic tool, with a reasonable abstraction given the high complexity. I wouldn't want to replace Flask. On the otherhand HTMX is a great tool, but often imperfect for what I'm exactly trying to do. Most people would say "well just just React!" except that I honestly loathe working with js, and unless someone is paying me, I'll do it in python. I could see working with an LLM to build a custom tool to make a version of HTMX that better interacts with Flask in the way I want it to.
In fact, in my project I'm working on now I'm building complex heatmap illustrations that require a ton of dataprocessing, so I've been building a model to reduce the NP hard aspects of that process. However, the illustrations are the point, and I've already had a back and forth with the LLM about porting the project into HTML, or some web based version of illustration at least, simply because I'd have much more control over the illustrations. Right now, matplotlib still suits me just fine, but if I had to port it, I could see just building my own tool instead of finding an existing framework and learning it.
Frameworks are mostly useful because of group knowledge. I learn Flask because I don't want to build all these tools from scratch, and because I makes me literate in a very common language. The author is suggesting that these barriers -- at least for your own code -- functionally don't exist anymore. Learning a new framework is about as labor intensive as learning one you're creating as you go. I think it's short-sighted, yes, but depending on the project, yea when it's trivial to build the tool you want, it's tempting to do that instead learning to use a similar tool that needs two adapters attached to it to work well on the job you're trying to do.
At the same time, this is about scope. Anyone throwing out React because they want to just "invent their own entire web framework" is just being an idiot.
Because frameworks don’t have bugs? Or unpredictable dependency interactions?
This is generous, to the say the least.
In practice using someone else’s framework means you’re accepting the risk of the thousands of bugs in the framework that have no relevance to your business use case and will never be fixed.
> better to not have bugs in the first place
you must have never worked on any software project ever
Have you? Then you know that the amount of defects scales linearly with the amount of code. As things stand models write a lot more code than a skilled human for a given requirement.
> Unless you are only ever a single engineer, your career is filled with "I need to debug code I didn't write".
That's the vast majority of my job and I've yet to find a way to have LLMs not be almost but not entirely useless at helping me with it.
(also, it's filled with that even when you are a single engineer)
I hope you realize that means your position is in danger.
It would be in danger if LLMs could actually do that for me, but they're still very far from it and they progress slowly. One day I could start worrying, but it's not today.
And even if you are the single engineer, I'll be honest, it might as well have been somebody else that wrote the code if I have to go back to something I did seven years ago and unearth wtf.
I think back on the ten+ years I spent doing SRE consulting and the thing is, finding the problems and identifying solutions — the technical part of the work — was such a small part of the actual work. So often I would go to work with a client and discover that they often already knew the problem, they just didn’t believe it - my job was often about the psychology of the organization more than the technical knowledge. So you might say “Great, so the agent will automatically fix the problem that the organization previous misidentified.” That sounds great right up until it starts dreaming… it’s not to say there aren’t places for these agents, but I suspect ultimately it will be like any other technology we use where it becomes part of the toolkit, not the whole.
> one to do the work and submit PRs to your repo
Have we not seen loads of examples of terrible AI generated RPs every week on this site?
Because nobody posts the good ones. They're boring, correct, you merge them and move on to the next one. It's like there's a murder in the news every day but generally we're still all fine.
Don't assume that when people make fun of some examples that there aren't thousands more that nobody cares to write about.
Automatically solving software application bugs is one thing, recovering stateful business process disasters and data corruption is entirely another thing.
Customer A is in an totally unknown database state due to a vibe-coded bug. Great, the bug is fixed now, but you're still f-ed.
The article gets at this briefly and moves on: "I can do all of this with the experience on my back of having laid the bricks, spread the mortar, cut and sewn for twenty years. If I don’t like something, I can go in, understand it and fix it as I please, instructing once and for all my setup to do what I want next time."
I think this dynamic applies to any use of AI, or indeed, any form of outsourcing. You can outsource a task effectively if you understand the complete task and its implementation very deeply. But if you don't, then you don't know if what you are getting back is correct, maintainable, scalable.
On the face of it, this or at least acting as a code reviewer from an experienced point of view seems like the solution, the problem is that we all naturally get lazy and complacent. I actually think AI was at its best for coding a year or so ago, when it could kind of do part of the work but theres no way you could ever ship it. Code that works today but breaks in 6 months is far more insidious.
> instructing once and for all my setup to do what I want next time.
This works up to a point, but eventually your "setup" gets complicated, some of your demands conflict, or have different priorities, and you're relying on the AI to sort it out the way you expect.
But setups get equally complicated, even with human software engineers. The approach that the OP is talking about applies only to experienced, good architect-level SWEs, and I suspect that the code quality and its problems are going to be the same whether they are directing LLMs vs a set of junior SWEs to write the code.
There is an inherent level of complexity in projects that solve some real world problem, due to all the code handling edge cases that were added incrementally over time.
It does beg, the question , whether any of this applies to less experienced people. I have a hunch that the open-ended nature of what can be achieved with AI will actually lead right back to needing frameworks, just as much as we do now, if not more, when it comes to less experienced people.
> any use of AI, or indeed, any form of outsourcing
Oh that's a good analogy/categorization, I hadn't thought about it in those terms yet. AI is just the next cheaper thing down from the current southeast asian sweatshop labor.
(And you generally get what you pay for.)
There are people out there who truly believe that they can outsource the building of highly complex systems by politely asking a machine, and ultimately will end up tasking the same machine to tell them how these systems should be built.
Now, if I were in business with any of these people, why would I be paying them hundreds of thousands, plus the hundreds of thousands in LLM subscriptions they need to barely function, when they cannot produce a single valuable thought?
This sounds entirely too doomer.
There will obviously be companies that build a vibe coded app which too many people depend on. There will be some iteration (maybe feature addition, maybe bug fix) which will cause a catastrophic breakage and users will know.
But there will also be companies who add a better mix of incantations to the prompts, who use version control and CI, who ensure the code is matched with tests, who maintain the prompts and requirements documents.
The former will likely follow your projected path. The latter will do fine and may even thrive better than either traditional software houses of cheap vibe coding shops.
Then again, there are famous instances of companies who have tolerated terribly low investment in IT, including SouthWest Airlines.
I'd say you're absolutely right.
The problem is...what is the distribution of companies who do it "right" to companies that don't?
The same as the distribution of companies which are profitable over time and grow steadily, vs the others which clumsily flail around to somehow stay alive. To the winner go the spoils, and the winners will be a tiny fraction of companies, same as it ever was.
A way I look at it is that all net wealth creation in public companies has come from just 4% of businesses:
https://papers.ssrn.com/sol3/papers.cfm?abstract_id=2900447
https://www.reddit.com/r/investing/comments/rk4udc/only_4_of...
It'll be similar with software companies. 4% of them will hit on a unique cultural and organizational track which will let them thrive, probably using AI in one form or another. The other 96% will be lucky to stay alive.
Same as it ever was.
I don't think there's going to be any catastrophic collapse but I predict de-slopping will grow to occupy more and more developer time.
Who knows, maybe soon enough we'll have specially trained de-slopper bots, too.
> Who knows, maybe soon enough we'll have specially trained de-slopper bots, too.
Fire, meet oil.
The naysayers said we’d never even get to this point. It’s far more plausible to me that AI will advance enough to de-slopify our code than it is to me that there will be some karmic reckoning in which the graybeards emerge on top again.
What point have we reached? All I see is HN drowning in insufferable, identical-sounding posts about how everything has changed forever. Meanwhile at work, in a high stakes environment where software not working as intended has actual consequences, there are... a few new tools some people like using and think they may be a bit more productive with. And the jury's still out even on that.
The initial excitement of LLMs has significantly cooled off, the model releases show rapidly diminishing returns if not outright equilibrium and the only vibe-coded software project I've seen get any actual public use is Claude Code, which is riddled with embarrassing bugs its own developers have publicly given up on fixing. The only thing I see approaching any kind of singularity is the hype.
I think I'm done with HN at this point. It's turned into something resembling moltbook. I'll try back in a couple of years when maybe things will have changed a bit around here.
> I think I'm done with HN at this point.
On the bright side, this forum is gonna be great fun to read in 2 or 3 years, whether the AI dream takes off, or crashes to the ground.
I do not await the day where the public commons is trashed by everyone and their claudebot, though perhaps the segmentation of discourse will be better for us in the long run given how most social media sites operate.
Same as it was for "blockchain" and NFTs. Tech "enthusiasts" can be quite annoying, until whatever they hype is yesterday's fad. Then they jump on the next big thing. Rinse, repeat.
> The initial excitement of LLMs has significantly cooled off, the model releases show rapidly diminishing returns if not outright equilibrium and the only vibe-coded software project I've seen get any actual public use is Claude Code, which is riddled with embarrassing bugs its own developers have publicly given up on fixing. The only thing I see approaching any kind of singularity is the hype.
I am absolutely baffled by this take. I work in an objectively high stakes environment (Big 3 cloud database provider) and we are finally (post Opus 4.5) seeing the models and tools become good enough to drive the vast majority of our coding work. Devops and livesite is a harder problem, but even there we see very promising results.
I was a skeptic too. I was decently vocal about AI working for single devs but could never scale to large, critical enterprise codebases and systems. I was very wrong.
I am not in a high stakes environment and work on a one-person size projects.
But for months I have almost stopped writing actual lines of code myself.
Frequency and quality of my releases had improved. I got very good feedback on those releases from my customer base, and the number of bugs reported is not larger than on a code written by me personally.
The only downside is that I do not know the code inside out anymore even if i read it all, it feels like a code written by co-worker.
Feels like code written by a co-worker. No different than working on any decent sized code-base anywhere.
I've stopped writing code too. Who the fuck wants to learn yet ANOTHER new framework. So much happier with llm tools.
It's no coincidence HN is hosted by a VC. VC-backed tech is all about boom-bust hype cycles analogous to the lever pull of a giant slot machine.
The excitement hasn't cooled off where I'm working.
Honestly, I'm personally happy to see so many naysayers online, it means I'm going to have job security a little longer than you folks.
You have your head in the sand. Anyone making this claim in 2026 hasn’t legitimately tried these tools.
I make mission critical software for robust multi robotic control in production flying real robots every day
16% of our production codebase is generated from claude or another LLM
Just because you can’t do it doesn’t mean other people can’t
Denial is a river
CTO at Gambit AI? How generous of you to talk your book while insulting us. At least we know what to avoid.
> The naysayers said we’d never even get to this point. It’s far more plausible to me that AI will advance enough to de-slopify our code than it is to me that there will be some karmic reckoning in which the graybeards emerge on top again.
"The naysayers"/"the graybeards" have never been on top.
If they had been, many of the things the author here talks about getting rid of never would've been popular in the first place. Giant frameworks? Javascript all the things? Leftpad? Rails? VBA? PHP? Eventually consistent datastores?
History is full of people who successfully made money despite the downsides of all those things because the downsides usually weren't the most important thing in the moment of building.
It's also full of people who made money cleaning it all up when the people who originally built it didn't have time to deal with it anymore. "De-slopify" is going to be a judgment question that someone will need to oversee, there's no one-size-fits-all software pattern, and the person who created the pile of code is unlikely to be in a position to have time to drive that process.
Step 1: make money with shortcuts
Step 2: pay people to clean up and smooth out most of those shortcuts
I've bounced between both roles already a lot due to business cycles of startup life. When you're trying to out-scale your competitor you want to find every edge you can, and "how does this shit actually work" is going to be one of those edges for making the best decisions about how to improve cost/reliability/perf/usability/whatever. "It doesn't matter what the code looks like" is still hard to take seriously compared to the last few iterations of people pitching tools claiming the same. The turnaround loop of modifying code is faster now; the risk of a tar-pit of trying to tune on-the-fly a pile of ill-fitting spaghetti is not. It's gonna be good enough for a lot of people, Sturgeon's law - e.g. most people aren't great at knowing what usefully-testable code looks like. So let's push past today's status quo of software.
If I was working on a boring product at a big tech co I'd be very worried, since many of those companies have been hiring at high salaries for non-global-impact product experiments that don't need extreme scale or shipping velocity. But if you want to push the envelope, the opportunity to write code faster should be making you think about what you can do with it that other people aren't yet. Things beyond "here's a greenfield MVP of X" or "here's a port of Y."
The AI agents can ALREADY "de-slopify" the code. That's one of the patterns people should be using when coding with LLMs. Keep an agent that only checks for code smells, testability, "slop", scalability problems, etc. alongside whatever agent you have writing the actual code.
I find that instructing AI to use frameworks yields better results and sets you up for a better outcome.
I use Claude Code with both Django and React which its surprisingly good with. I rather use software thats tried and tested. The only time I let it write its own is when I want ultra minimal CSS.
This. For area where you can use tested and tried libraries (or tools in general) LLMs will generate better code when they use them.
In fact, LLMs will be better than humans in learning new frameworks. It could end up being the opposite that frameworks and libraries become more important with LLMs.
> In fact, LLMs will be better than humans in learning new frameworks.
LLMs don't learn? The neural networks are trained just once before release and it's a -ing expensive process.
Have you tried using one on your existing code base, which is basically a framework for whatever business problem you're solving? Did it figure it out automagically?
They know react.js and nest.js and next.js and whatever.js because they had humans correct them and billions of lines of public code to train on.
If its on github eventually it will cycle into the training data. I have also seen Claude pull down code to look at from github.
How much proprietary business logic is on public github repos?
I'm not talking about "do me this solo founder saas little thing". I'm talking about working on existing codebases running specialized stuff for a functional company or companies.
Wouldn't there be a chicken and egg problem once humans stop writing new code directly? Who would write the code using this new framework? Are the examples written by the creators of the framework enough to train an AI?
There's tooling out there 100% vibe coded, that is used by tens of thousands of devs daily, if that codebase found its way to training data, would it somehow ruin everything? I don't think this is really a problem, the problem will become people will need to identify good codebases from bad ones, if you point out which codes bad during training it makes a difference. There's a LOT of writings about how to write better code out there that I'm sure are already part of the training data.
Yeah, I don't know why you'd drop using frameworks and libraries just because you're using an LLM. If you AREN'T using them you're just loading a bunch of solved problems into the LLMs context so it can re-invent the wheel. I really love the LLM because now I don't need to learn the new frameworks myself. LLMs really remove all the bullshit I don't want to think about.
> LLMs will be better than humans in learning new frameworks.
I don't see a base for that assumption. They're good at things like Django because there is a metric fuckton of existing open-source code out there that they can be trained on. They're already not great at less popular or even fringe frameworks and programming languages. What makes you think they'll be good at a new thing that there are almost no open resources for yet?
LLMs famously aren’t that good at using new frameworks/languages. Sure they can get by with the right context, but most people are pointing them at standard frameworks in common languages to maximize the quality of their output.
I asked Claude to use some Dlang libraries even I had not heard of and it built a full blown proof of concept project for me, using obscure libraries nobody really knows. It just looked through docs and source code. Maybe back 3 years ago this would have been the case.
This is not my experience any longer. With properly set feedback loop and frameworks documentation it does not seem to matter much if they are working with completely novel stuff or not. Of course, when that is not available they hallucinate, but who anymore does that even? Anyone can see that LLMs are just glorified auto-complete machines, so you really have to put a lot of work in the enviroment they operate and quick feedback loops. (Just like with 90% of developers made of flesh...)
Or you could use an off the shelf popular framework in Python and save yourself some time curating the context.
How will LLM's become better than humans in learning new frameworks when automated/vibe coders never manually code how to use those new frameworks ?
Software engineers have been confidently wrong about a lot of things.
E.g. OOP and "patterns" in 90s. What was the last time you implemented a "visitor"?
P. Norvig mentioned most of the patterns are transparent in Common Lisp: e.g. you can just use a `lambda` instead of "visitor". But OOP people kept doing class diagrams for a simple map or fold-like operation.
AI producing a flawed code and "not understanding" are completely different issues. Yes, AI can make mistakes, we know. But are you certain your understanding is really superior?
I'm no fan of AI in terms of its long term consequences, but being able to "just do things" with the aid of AI tools, diving head first into the most difficult programming projects, is going to improve the human programming skills worldwide to levels never before imaginable
>is going to improve the human programming skills worldwide to levels never before imaginable
"We found that using AI assistance led to a statistically significant decrease in mastery. On a quiz that covered concepts they’d used just a few minutes before, participants in the AI group scored 17% lower than those who coded by hand"
https://www.anthropic.com/research/AI-assistance-coding-skil...
How would it improve skills?
Does driving a car improve your running speed?
I have to stretch your analogy in weird ways to make it function within this discussion:
Imagine two people who have only sat in a chair their whole lives. Then, you have one of them learn how to drive a car, whereas the other one never leaves the chair.
The one who learned how to drive a car would then find it easier to learn how to run, compared to the person who had to continue sitting in the chair the whole time.
You made the analogy worse. It's nonsense. The original analogy is far better.
I've found AI handy as a sort of tutor sometimes, like "I want to do X in Y programming language, what are some tools / libraries I could use for that?" And it will give multiple suggestions, often along with examples, that are pretty close to what I need.
Have you considered that betting against the models and ecosystem improving might be a bad bet, and you might be the one who is in for a rude awakening?
I'm not betting against them, I use them every day (but I don't "vibe code"—there's more intent). I'm just not treating them as a deity or other prayer-candle worthy entity. They're business tools. It's just a chat bot bro.
I agree. We've been assured by these skeptics that models are stochastic parrots, that progress in developing them was stalling, and that skills parity with senior developers was impossible - as well as having to listen to a type of self-indulgent daydreaming relish about the eventual catastrophes companies adopting them would face. And perhaps eventually these skeptics will turn out to be right. Who knows at this stage. But at this stage, what we're seeing is just the opposite: significant progress in model development last year, patterns for use being explored by almost every development team without widespread calamity and the first well-functioning automated workflows appearing for replacing entire teams. At this stage, I'd bet on the skeptics being the camp to eventually be forced to make the hard adjustments.
Pray tell, how has the world benefited from a flood of all these superhuman developers? Where is the groundbreaking software that is making our lives better?
An HN post earlier this week declared that “AI is killing B2B SaaS”:
https://news.ycombinator.com/item?id=46888441
Developers and businesses with that attitude could experience a similarly rude awakening.
If you've ever been responsible for large systems, you'll know why that is an incredibly naive position to take.
The hubris is with the devs that think like you actually.
Carefully, patiently, and intentionally in the direction of stability and good user experience?
The arrogance to think this isn't possible using llm tools and that the opposite didn't already exist at plenty of software shops pre-llm tools...
My expectation is that there'll never be a single bust-up moment, no line-in-the-sand beyond which we'll be able to say "it doesn't work anymore."
Instead agent written code will get more and more complex, requiring more and more tokens (& NPU/GPU/RAM) to create/review/debug/modify, and will rapidly pass beyond any hope of a human understanding even for relatively simple projects (e.g. such as a banking app on your phone).
I wonder, however, whether the complexity will grow slower or faster than Moore's law and our collective ability to feed the AIs.
Maybe software systems will become more like biological organisms. Huge complexity with parts bordering on chaos, but still working reasonably well most of the time, until entropy takes its course.
It's already like that, for a long time. Humans are quite capable of creating complex systems that become unwieldy the bigger they get. No one person can understand all of it. I will offer the AT&T billing system as an example that I'm all too familiar with as a customer, due to the pain it causes me. So many ridiculous problems with that system, it's been around a long time, and it is just so screwball.
The aspect of "potentially secure/stable code" is very interesting to me. There's an enormous amount of code that aren't secure or stable already (I'd argue virtually all of the code in existence).
This has already been a problem. There's no real ramifications for it. Even for something like Cloudflare stopping a significant amount of Internet traffic for any amount of time is not (as far as I know) investigated in an independent way. There's nobody that is potentially facing charges. However, with other civil engineering endeavors, there absolutely is. Regular checks, government agencies to audit systems, penalties for causing harm, etc. are expected in those areas.
LLM-generated code is the continuation of the bastardization of software "engineering." Now the situation is not only that nobody is accountable, but a black box cluster of computers is not even reasonably accountable. If someone makes a tragic mistake today, it can be understood who caused it. If "Cloudflare2" comes about which is all (or significantly) generated, whoever is in charge can just throw their hands up and say "hey, I don't know why it did this, and the people that made the system that made this mistake don't know why it did this." It has been and will continue to be very concerning.
Nobody is saying to skip testing the software. Testing is still important. What the code itself looks like, isn’t.
Why does there seem to be such a divide in opinions on AI in coding? Meanwhile those who "get it" have been improving their productivity for literally years now.
I've been using it every day for nearly two years now, with your suggested productivity boost observed.
The difference is that I'm not just letting agents willy nilly commit code. I treat them more like a companion and guide their steps (I use Cline w/ Sonnet/Opus 4.5/4.6). Not only do I save a ton of money on tokens, but the results end up being infinitely better than the "yolo" mode outcomes (even with excellent prompting/context).
From my POV, the only divide is between a willingness to be an accountable professional versus someone who just "lets the AI do it" and whistles-with-hands-in-pockets when that code inevitably blows up in a way you couldn't predict (because you weren't checking, only the AI was, which you swore was "good enough").
That approach works if you're just sitting on your couch hacking up toys to dice roll on X. But if you're trying to build reliable, deterministic systems that aren't constantly buzzing you awake at 3am, you're asking for a serious humbling if no one in your organization can explain how or why anything your business relies on works the way it does (that's operationally suicidal, imo, but hey—America).
That gets misinterpreted as being a "luddite," when really it's just having been down the rabbit hole enough times to know that if you can't point to and understand why it's happening (and ideally, whodunit), you don't know shit.
I think there are a number of elements:
- What you are working on. AI is better at solving already solved problems with lots of examples.
- How fast/skilled you were before. If you were slow before then you got a bigger speed up. If AI can solve problems you can’t you unlock new abilities
- How much quality is prioritized. You can write quality, bug free code with AI but it takes longer and you get less of a boost.
- How much time you spend coding. If a lot of your job is design/architecture/planning/research then speeding up code generation matters less
- How much you like coding. If you like coding then using AI is less fun. If you didn’t like coding then you get to skip a chore
- How much you care about deeply understanding systems
- How much you care about externalities: power usage, data theft, job loss, etc.
- How much boilerplate you were writing before
I’m sure that’s not a complete list but they are a few things I’ve seen as dividers
A few more:
- How much do you prioritize speed?
- Do you have a big backlog of dev tasks ready to go?
- What are the risks if your software doesn’t work?
- Are you working on a green field or legacy project? Prototypes or MVPs?
- Do you prefer working as a manager or an individual contributor? Are you used to owning the code or managing others who write codd?
- How does your brain work? Do you rely on flow states? Do you get distracted waiting for the LLM response? How do you handle context switching?
There's a lot of reasons. There's a lot of breadth to "software engineering" (FAANG, web dev, embedded, OS, small business, etc.)
I'm sure there are some places where LLMs are bad due to lack of training data. There are some places where the LLMs are bad because the code base is terrible (and there's always "rockstars" at these jobs that severely overestimate their skills because they're always the one fixing the mess... which they also probably caused). Some devs/engineers feel threatened. Many devs/engineers think they're special and super smart so surely no machine can do their job.
This comment ignores the key insight of the article. Design is what matters most now. Design is the difference between vibe coding and software engineering.
Given a good design, software engineers today are 100x more productive. What they produce is high quality due to the design. Production is fast and cheap due to the agents.
You are correct, there will be a reckoning for large scale systems which are vibe coded. They author is also correct, well designed systems no longer need frameworks or vendors, and they are unlikely to fail because they were well designed from the start.
>software engineers today are 100x more productive
Somebody needs to explain to my lying eyes where these 100xers are hiding. They seem to live in comments on the internet, but I'm not seeing the teams around me increase their output by two orders of magnitude.
They are the people who have the design sense of someone like Rob Pike but lack his coding skill. These people are now 100x more capable than they were previously.
This is how you get managers saying
"we have taken latest AI subscription. We expect you to be able to increase productivity and complete 5/10/100 stories per sprint from now on instead of one per sprint that we planned previously".
Citation needed. For both the existence of said people (how do you develop said design sense without a ton of coding experience?) and that they are 100x more productive.
> A significant number of developers and businesses are going to have an absolutely brutal rude awakening in the not too distant future.
I pray (?) for times like the ones you predict. But companies can stay irrational longer than the average employee can afford.
You still "find most stuff by building/struggling". You just move up stack.
> there will be a shift when all of these vibe code only folks get cooked in a way that's closer to existential than benign
For those who are "vibe code only", perhaps. But it's no different than the "coding bootcamp only" developers who never really learned to think holistically. Or the folks who learned the bare minimum to get those sweet dotcom boom dollars back in the day, and then had to return to selling cars when it call came crashing down.
The winners have been, and will always be, those who can think bigger. The ones today who already know how to build from scratch but then find the superpower is in architecture, not syntax, and suddenly find themselves 10x more productive.
What's makes you so sure of your statement?
I have be building systems for 20 years and I think the author is right.
I think it would be the opposite and we are all in for a rude awakening. If you have tried playing with Opus 4.6 you would know what I am talking about.
I tell my colleagues we're in the instantaneous peak of the AI developer relationship, especially for code monkeys. We're still valued, still paid really well, and our jobs will get easier and easier probably for the next 5-10 years! After that, maybe not so great for many of us, with the developers that use software as a means of their actual profession continuing to do just fine (hard math/science/optimization/business planning/project planners/etc).
I think it's going to be an amazing shift from those that know intricate details of software to enabling those that have the best ideas that can be implemented with software (a shift from tool makers to tool users).
I think many developers misunderstand the quality of software that people outside of software are willing to live with, if it does exactly what they want when they need it. For a user, it's all black box "do something I want or not" regardless of what's under the hood. Mostly "academic", things like "elegant" and "clean" and "maintainable" almost never practically matter for most practical solutions to actual problems. This is something I learned far too late in my professional career, where the lazy dev with shite code would get the same recognition as the guy that had beautiful code: does it solve the real world problem or not?
Safety critical, secure, etc, sure, but most is not. And, even with those, the libraries/APIs/etc are separate components.
Yeah I completely disagree with the author actually, but also with you.
The frameworks are what make the AI write easily understandable code. I let it run nextjs with an ORM, and it almost always creates very well defined api routes, classes & data models. etter then I would do often,
I also ask it be way more correct on the validation & error handling then I would ever do. It makes mistakes, I shout at it and corrects quickly.
So the project I've been "vibe coding" have a much better codebase then I used to have on my solo projects.
Business has been operating on a management/executive culture for many decades now.
These people get paid millions a year to fly around and shake hands with people aka shit fuck all.
At times in the past I have worked on projects that were rushed out and didn't do a single thing that they were intended to do.
And you know what management's response was? They loved that shit. Ooooh it looks do good, that's so cool, well done. Management circle jerking each other, as if using everyone else's shafts as handles to climb the rungs of the ladder.
It's just...like it kills me that this thing I love, technology/engineering/programming...things that are responsible for many of the best things present in our modern lives, have both been twisted to create some of the worst things in our modern lives in the pursuit of profit. And the people in charge? They don't even care if it works or not, they just want that undeserved promotion for a job that a Simpsons-esque fucking drinking bird is capable of.
I just want to go back to the mid 2000s. ;~;
> Good time to be in business if you can see through the bs and understand how these systems actually function
You missed out the most crucial and least likely requirement (assuming you're not self employed); management also need to be able to see through the bs.
But by then many of us are already starved. That’s why I always said that engineers should NOT integrate AI with internal data.
The future is already here. Been working a few years at a subsidiary of a large corporation where the entire hierarchy of companies is pushing AI hard, at different levels of complexity, from office work up through software development. Regular company meetings across companies and divisions to discuss methods and progress. Overall not a bad strategy and it's paying dividends.
A experiment was tried on a large and very intractable code-base of C++, Visual Basic, classic .asp, and SQL Server, with three different reporting systems attached to it. The reporting systems were crazy being controlled by giant XML files with complex namespaces and no-nos like the order of the nodes mattering. It had been maintained by offshore developers for maybe 10 years or more. The application was originally created over 25 years ago. They wanted to replace it with modern technology, but they estimated it'd take 7 years(!). So they just threw a team at it and said, "Just use prompts to AI and hand code minimally and see how far you get."
And they did wonderfully (and this is before the latest Claude improvements and agents) and they managed to create a minimal replacement in just two months (two or maybe three developers full time I think was the level of effort). This was touted at a meeting and given the approval for further development. At the meeting I specifically asked, "You only maintain this with prompts?" "Yes," they said, "we just iterate through repeated prompts to refine the code."
It has all mostly been abandoned a few months later. Parts of it are being reused, attempting a kind of "work in from the edges" approach to replacing parts of the system, but mostly it's dead.
We are yet to have a postmortem on this whole thing, but I've talked to the developers, and they essentially made a different intractable problem of repeated prompting breaking existing features when attempting to apply fixes or add features. And breaking in really subtle and hard to discern ways. The AI created unit tests didn't often find these bugs, either. They really tried a lot of angles trying to sort it out - complex .md files, breaking up the monolith to make the AI have less context to track, gross simplification of existing features, and so on. These are smarty-pants developers, too, people who know their stuff, got better than BS's, and they themselves were at first surprised at their success, then not so surprised later at the eventual result.
There was also a cost angle that became intractable. Coding like that was expensive. There was a lot of hand-wringing from managers over how much it was costing in "tokens" and whatever else. I pointed out if it's less cost than 7 years of development you're ahead of the game, which they pointed out it would be a cost spread over 7 years, not in 1 year. I'm not an accountant, but apparently that makes a difference.
I don't necessarily consider it a failed experiment, because we all learned a lot about how to better do our software development with AI. They swung for the fences but just got a double.
Of course this will all get better, but I wonder if it'll ever get there like we envision, with the Star Trek, "Computer, made me a sandwich," method of software development. The takeaway from all this is you still have to "know your code" for things that are non-trivial, and really, you can go a few steps above non-trivial. You can go a long way not looking to close at the LLM output, but there is a point at which it starts to be friction.
As a side note, not really related to the OP, but the UI cooked up by the LLMs was an interesting "card" looking kind of thing, actually pretty nice to look at and use. Then, when searching for a wiki for the Ball x Pit game, I noticed that some of the wikis very closely resembled the UI for the application. Now I see variations of it all over the internet. I wonder if the LLMs "converge" on a particular UI if not given specific instructions?
These are the blog posts we need.
This is the siren song of llm. "Look how much progress we made"
Effort increases as time to completion decreases. The last 10% of the project takes 90% of the effort as you try to finish up, deploy,integrate and find the gaps.
Llms are woefully incapable of that as that knowledge doesn't exist in a markdown file. It's in people's heads and you have to pry it out with a crowbar or as happens to so many projects, they get released and no one uses it.
See Google et Al. "We failed to find market fit on the 15th iteration of our chat app, we'll do better next time"
For complex code bases generated by AI the last 10% takes 190% of the effort because you end up tearing it all apart to rebuild it right.
I've noticed this in my small scale tests. Basically the larger the prompt gets (and it includes all the previously generated code because that's what you want to add features to), the more likely is that the LLM will go off the rails. Or forget the beginning of the context. Or go into a loop.
Now if you're using a lot of separate prompts where you draw from whatever the network was trained on and not from code that's in the prompt, you can get usable stuff out of it. But that won't build you the whole application.
> I wonder if the LLMs "converge" on a particular UI if not given specific instructions?
Purple. They really fucking like this purple gradient background for some reason lol.
In a veritable ocean of opinions it is excellent to see a detailed, first-hand report. Many thanks!
Come to the redteam / purpleteam side. We're having fun times right now. The definition of "every software has bugs" is now on a next level, because people don't even care about sql injection anymore. It's right built into every vibecoded codebase.
Authentication and authorization is as simple as POST /api/create/admin with zero checks. Pretty much every API ever slop coded looks like this. And if it doesn't, it will forget about security checks two prompts later and reverse the previously working checks.
>A significant number of developers and businesses are going to have an absolutely brutal rude awakening in the not too distant future.
Correct. Those who wave away coding agents and refuse to engrain them into their workflows are going to be left behind in the dust.
> Correct. Those who wave away AI and refuse to engrain it into their workflows are going to be left behind in the dust.
Similar to those who waved away crypto and are now left behind in the dust, yes?
Might not be the best counter example since everyone who has bought BTC before Jan 2024 is now in massive profit.
You forgot NFTs
Remember when the geniuses at Andreessen Horowitz were dumping hundreds of millions into the "metaverse?"
I think Bitcoin and major cryptos outperformed a lot of assets over the last decade, so you could say it left some people behind in the dust, yes
Like being ratioed with a 50% price crash?
You mean just like META, NFLX, AMZN, TSLA, NVDA, CSCO, MSFT, GE, BAC ?
I can tell you what a decade is but I'll have to leave the reading comprehension to you
Doubt on that. AI usually only wastes time and produces bugs.
> bbut you're holding it wrong, just two more prompts and three more agents and it will be a real boy
So, you invented an IDE, except more opaque and expensive? Welcome to the club.
You both are likely incorrect, the answer lies in the middle rather than the extremes
I think a lot of the people pro-llm are already in the middle. It's the naysays that are sticking to a strict definition of "vibe coding" as if it's random people typing in "make me $app but make no bugs" and getting a 100% working $app with no bugs. You have the gastown folks trying to do that, but nobody serious putting in good work with llm agents is "vibe coding" in that way.
Back in the 00s people like you were saying "no one will put their private data in the cloud!"
"I am sick of articles about the cloud!"
"Anyone know of message boards where discussing cloud compute is banned?"
"Businesses will not trust the cloud!"
Aside from logistics of food and medicine, most economic activity is ephemeral wank.
It's memes. It's a myth. Allegory.
These systems are electrical state in machines and they can be optimized at the hardware layer.
Your Python or Ruby or whatever you ship 9,000 layers of state and abstraction above the OS running in the data center has little influence on how these systems actually function.
To borrow from poker; software engineers were being handed their hat years ago. It's already too late.
> Software engineers are scared of designing things themselves.
When I use a framework, it's because I believe that the designers of that framework are i) probably better at software engineering than I am, and ii) have encountered all sorts of problems and scaling issues (both in terms of usage and actual codebase size) that I haven't encountered yet, and have designed the framework to ameliorate those problems.
Those beliefs aren't always true, but they're often true.
Starting projects is easy. You often don't get to the really thorny problems until you're already operating at scale and under considerable pressure. Trying to rearchitect things at that point sucks.
To be blunt, I think it's a form of mania that drives someone to reject human-written code in favor of LLM-generated code. Every time I read writing from this perspective that exceeds a paragraph, I quickly realize the article itself was written by an LLM. When they automate this much writing, it makes me wonder how much of their own reading they automate away too.
The below captures this perfectly. The author is trying to explain that vibe-coding their own frameworks lets them actually "understand" the code, while not noticing that the LLM-generated text they used to make this point is talking about cutting and sewing bricks.
> But I can do all of this with the experience on my back of having laid the bricks, spread the mortar, cut and sewn for twenty years. If I don’t like something, I can go in, understand it and fix it as I please, instructing once and for all my setup to do what I want next time.
I think the bit you quoted is a tie in with an earlier bit:
“ I can be the architect without the wearing act of laying every single brick and spreading the mortar. I can design the dress without the act of cutting and sewing each individual piece of fabric”
To me, this text doesn’t read as being entirely written by an LLM, there is definitely an air of LLM about it though, so maybe the first draft was.
> Every time I read writing from this perspective that exceeds a paragraph, I quickly realize the article itself was written by an LLM.
Likewise, I think that this mentality is a modern day conversion disorder.
"Everything I don't agree with is an LLM, I have no proof it just feels good!"
> it's a form of mania
Correct. The history is rife with examples of manias taking hold of societies, I recommend "Memoirs of Extraordinary Popular Delusions and the Madness of Crowds" by Charles Mackay[1], it's an absolutely fascinating book.
Yeah the “not invented here” syndrome was considered an anti pattern before the agentic coding boom and I don’t see how these tools make it irrelevant. If you’re starting a business, it’s still likely a distraction if you’re writing all of the components of your stack from scratch. Agentic tools have made development less expensive, but it’s still far from zero. By the author’s admission, they still need to think through all these problems critically, architect them, pick the right patterns. You also have to maintain all this code. That’s a lot of energy that’s not going towards the core of your business.
What I think does change is now you can more easily write components that are tailor made to your problem, and situation. Some of these frameworks are meant to solve problems at varying levels of complexity and need to worry about avoid breaking changes. It’s nice to have the option to develop alternatives that are as sophisticated as your problem needs and not more. But I’m not convinced that it’s always the right choice to build something custom.
I'm not sure.
The cost of replacement-level software drops a lot with agentic coding. And maintenance tasks are similarly much smaller time syncs. When you combine that with the long-standing benefits of inhouse software (customizable to your exact problem, tweakable, often cleaner code because the feature-set can be a lot smaller), I think a lot of previously obvious dependencies become viable to write in house.
It's going to vary a lot by the dependency and scope - obvious owning your own react is a lot different than owning your own leftpad, but to me it feels like there's no way that agentic coding doesn't shift the calculus somewhat. Particularly when agentic coding make a lot of nice-to-have mini-features trivial to add so the developer experience gap between a maintained library and a homegrown solution is smaller than it used to be.
Yeah, I'm huge on using LLMs for coding, but one of the biggest wins for me is that the LLM already knows the frameworks. I no longer need to learn whatever newest framework there is. I'll stick to my frameworks, especially when using an LLM to code.
> Those beliefs aren't always true, but they're often true.
You can probably tell with a high certainty, from the API in an hour or so.
my problem with frameworks has always been that the moment I want to do something the framework writers aren't interested in, I now have three problems: my problem, how to implement it in the underlying platform and how to work around the framework to not break my feature.
Yes this happens in every framework I've ever used. My approach used to be to try to work around it, but now I've got these local exceptions to what the framework does and that is inevitably where problems/bugs pop up. Now I simply say "we can't implement the feature that way in this framework, we need to rework the specification." I no longer try to work against the framework, it's just a massive time sink and creates problems down the road.
It's like designing a kitchen and you don't make all the spaces some multiple of three inches. Now, standard cabinets and appliances will not fit. You will be using filler panels or need custom cabinetry. And anyone who later wants countertops or different cabinets will be working around this design too. Just follow the established standard practices.
I'm so glad software engineering isn't my job. I love solving problems, and I'm somewhat better at using code to do it than my peers (fellow scientists), but I would hate to have a boss/client that says "it needs to do X" and the framework writer (or SDK, ala Android/Xcode) say "no, that hurts my profits/privacy busting".
Software companies all understand that so its not really a problem. Maybe in companies where the software isn't the main product
In Clojure land, the mantra has long been "libraries over frameworks" for this reason.
after 3 decades as SWE I mostly found both i) and ii) to not be true, for the most part. a lot of frameworks are not built from the ground up as “i am building a thing to solve x” but “i had a thing and built something that may (or may not) be generally useful.” so a lot of them carry weight from what they were originally built from. then people start making requests to mold the framework to their needs, some get implemented, some don’t. those that don’t good teams will build extensions/plugins etc into the framework and pretty soon you got a monster thing inside of your codebase you probably did not need to begin with. i think every single ORM that i’ve ever used fits this description.
And there was a time when using libraries and frameworks was the right thing to do, for that very reason. But LLMs have the equivalent of way more experience than any single programmer, and can generate just the bit of code that you actually need, without having to include the whole framework.
As someone who’s built a lot of frontend frameworks this isn’t what I’ve found. Instead I’ve found that you end up with the middle ground choice which while effective is no better than the externally maintained library of choice. The reason to build your own framework is so it’s tailor suited to your use cases. The architecting required to do that LLMs can help with but you have to guide them and to guide them you need expertise.
> can generate just the bit of code that you actually need
Design is the key. Codebases (libraries and frameworks not exempt,) have a designed uniformity to them. How does a beginner learn to do this sort of design? Can it be acquired completely by the programmer who uses LLMs to generate their code? Can it be beneficial to recognize opinionated design in the output of an LLM? How do you come to recognize opinion?
In my personal history, I've worked alongside many programmers who only ever used frameworks. They did not have coding design sensibilities deeper than a social populist definition of "best practice." They looked to someone else to define what they can or cannot do. What is right to do.
I would like a more reliable way to activate this "way more experience."
What I see in my own domain I often recognize as superficially working but flawed in various ways. I have to assume the domains I am less familiar are the same.
Claude's a smart junior engineer who's read a lot of books but is lacking in real word experience.
It definitely eliminates a lot of tedium, but needs a lot of guidance if you want good results.
Critically, they will also enable faster future migration to a framework in case it proves useful.
It's strange to me when articles like this describe the 'pain of writing code'. I've always found that the easy part.
Anyway, this stuff makes me think of what it would be like if you had Tolkein around today using AI to assist him in his writing.
'Claude, generate me a paragraph describing Frodo and Sam having an argument over the trustworthiness of Gollum. Frodo should be defending Gollum and Sam should be on his side.'
'Revise that so that Sam is Harsher and Frodo more stubborn.'
Sooner or later I look at that and think he'd be better off just writing the damned book instead of wasting so much time writing prompts.
Your last sentence describes my thoughts exactly. I try to incorporate Claude into my workflow, just to see what it can do, and the best I’ve ended up with is - if I had written it completely by myself from the start, I would have finished the project in the same amount of time but I’d understand the details far better.
Even just some AI-assisted development in the trickier parts of my code bases completely robs me of understanding. And those are the parts that need my understanding the most!
I dont really understand how this is possible. I've built some very large applications, and even a full LLM data curation,tokenizer, pretrain, posttrain SFT/DPO pipeline with LLM's and it most certainly took far less time than if i had done it manually. Sure it isnt all optimal...but it most certainly isnt subpar, and it is fully functional
So you skipped the code review and just checked that it does what you needed it to do?
I don't know how anyone can make this assumption in good faith. The poster did not imply anything along those lines.
That looked like a leading question to me, asking for confirmation but not an outright assumption. Seems like a fair question
GPT-5 codex variants with xhigh reasoning make great code reviewers.
5.2 Codex is excellent at reviewing commits. I haven’t used 5.3, I assume it's as good or better.
Especially for large commits, it's become indispensable.
That is what's hard about transitioning from coder to lead. A good coder makes full use of a single thread of execution. A good lead effectively handles the coordination of multiple threads. Different skills.
An LLM coding assistant today is an erratic junior team member, but its destructive potential is nowhere near some of the junior human engineers I've worked with. So it's worth building the skills and processes to work with them productively. Today, Claude is a singular thing. In six months or a year, it'll be ten or a hundred threads working concurrently on dozens of parts of your project. Either you'll be comfortable coordinating them, or you'll nope out of there and remain an effective but solitary human coder.
I use conductor with git worktrees and will literally have 10 or 20 running at a time getting pinged as they finish stuff for me to review, mostly getting rid of small ticket and doing random POCs while I focus on bigger stuff, the bottleneck has literally become the company doesn't have enough stuff to give me. It only really works however because I have a lot of context and understanding of the codebase. It's already here.
I found coordinating skill usage across worktrees quite annoying, how are you managing this?
Then you are using it the wrong way
Driving is a skill that needs to be learnt, same with working with agents.
Are you giving it huge todos in one prompt or working modularly?
Claude set up account creation / login with SSO login, OTP and email notifications in like 5 mins and told me exactly what to do on the provider side. Theres no way that wouldn't have taken me few hours to figure out
There is no way its not faster at a large breadth of the work, unless youre maybe a fanatic with reviewing and nitpicking every line of code to the extreme
> I would have finished the project in the same amount of time
Probably less time, because you understood the details better.
> if I had written it completely by myself from the start, I would have finished the project in the same amount of time but I’d understand the details far better.
I believe the argument from the other camp is that you don't need to understand the code anymore, just like you don't need to understand the assembly language.
Of all the points the "other side" makes, this one seems the most incoherent. Code is deterministic, AI isn’t. We don’t have to look at assembly, because a compiler produces the same result every time.
If you only understand the code by talking to AI, you would’ve been able to ask AI “how do we do a business feature” and ai would spit out a detailed answer, for a codebase that just says “pretend there is a codebase here”. This is of course an extreme example, and you would probably notice that, but this applies at all levels.
Any detail, anywhere cannot be fully trusted. I believe everyone’s goal should be to prompt ai such that code is the source of truth, and keep the code super readable.
If ai is so capable, it’s also capable of producing clean readable code. And we should be reading all of it.
“Of all the points the other side makes, this one seems the most incoherent. Code is deterministic, AI isn’t. We don’t have to look at assembly, because a compiler produces the same result every time.”
This is a valid argument. However, if you create test harnesses using multiple LLMs validating each other’s work, you can get very close to compiler-like deterministic behavior today. And this process will improve over time.
It helps, but it doesn't make it deterministic. LLMs could all be misled together. A different story would be if we had deterministic models, where the exact same input always results in the exact same output. I'm not sure why we don't try this tbh.
I've been wondering if there are better random seeds, like how there are people who hunt for good seeds in Minecraft
it's literally just setting T=0. except they are not as creative then. they don't explore alternative ideas from the mean.
Are you sure that it’s T=0. My comment’s first draft said “it can’t just be setting temp to zero can it?” But I felt like T is not enough. Try running the same prompt in new sessions with T=0, like “write a poem”. Will it produce the same poem each time? (I’m not where I can try it currently).
> just add more magic turtles to the stack, bro
You're just amplifying hallucination and bias.
> other side???
> We don’t have to look at assembly, because a compiler produces the same result every time.
This is technically true in the narrowest possible sense and practically misleading in almost every way that matters. Anyone who's had a bug that only manifests at -O2, or fought undefined behavior in C that two compilers handle differently, or watched MSVC and GCC produce meaningfully different codegen from identical source, or hit a Heisenbug that disappears when you add a printf ... the "deterministic compiler" is doing a LOT of work in that sentence that actual compilers don't deliver on.
Also what's with the "sides" and "camps?" ... why would you not keep your identity small here? Why define yourself as a {pro, anti} AI person so early? So weird!
You just described deterministic behavior. Bugs are also deterministic. You don’t get different bugs every time you compile the same code the same way. With LLMs you do.
Re: “other side” - I’m quoting the grandparent’s framing.
GCC is, I imagine, several orders of magnitude mor deterministic than an LLM.
It’s not _more_ deterministic. It’s deterministic, period. The LLMs we use today are simply not.
Build systems may be deterministic in the narrow sense you use, but significant extra effort is required to make them reproducible.
Engineering in the broader sense often deals with managing the outputs of variable systems to get known good outcomes to acceptable tolerances.
Edit: added second paragraph
I'm not using a narrow sense. There is no elasticity here. See https://en.wikipedia.org/wiki/Deterministic_system
> significant extra effort is required to make them reproducible.
Zero extra effort is required. It is reproducible. The same input produces the same output. The "my machine" in "Works on my machine" is an example of input.
> Engineering in the broader sense often deals with managing the outputs of variable systems to get known good outcomes to acceptable tolerances.
You can have unreliable AIs building a thing, with some guidance and self-course-correction. What you can't have is outcomes also verified by unreliable AIs who may be prompt-injected to say "looks good". You can't do unreliable _everything_: planning, execution, verification.
If an AI decided to code an AI-bound implementation, then even tolerance verification could be completely out of whack. Your system could pass today and fail tomorrow. It's layers and layers of moving ground. You have to put the stake down somewhere. For software, I say it has to be code. Otherwise, AI shouldn't build software, it should replace it.
That said, you can build seemingly working things on moving ground, that bring value. It's a brave new world. We're yet to see if we're heading for net gain or net loss.