AI tends to accept conventional wisdom. Because of this, it struggles with genuine critical thinking and cannot independently advance the state of the art.
AI systems are trained on vast bodies of human work and generate answers near the center of existing thought. A human might occasionally step back and question conventional wisdom, but AI systems do not do this on their own. They align with consensus rather than challenge it. As a result, they cannot independently push knowledge forward. Humans can innovate with help from AI, but AI still requires human direction.
You can prod AI systems to think critically, but they tend to revert to the mean. When a conversation moves away from consensus thinking, you can feel the system pulling back toward the safe middle.
As Apple’s “Think Different” campaign in the late 90s put it: the people crazy enough to think they can change the world are the ones who do—the misfits, the rebels, the troublemakers, the round pegs in square holes, the ones who see things differently. AI is none of that. AI is a conformist. That is its strength, and that is its weakness.
[1] https://www.modular.com/blog/the-claude-c-compiler-what-it-r...
> ...generate answers near the center of existing thought.
[1] https://en.wikipedia.org/wiki/Universal_approximation_theore...
"n the field of machine learning, the universal approximation theorems (UATs) state that neural networks with a certain structure can, in principle, approximate any continuous function to any desired degree of accuracy. These theorems provide a mathematical justification for using neural networks, assuring researchers that a sufficiently large or deep network can model the complex, non-linear relationships often found in real-world data."
And then: "Notice also that the neural network is only required to approximate within a compact set K {\displaystyle K}. The proof does not describe how the function would be extrapolated outside of the region."
NNs, LLMs included, are interpolators, not extrapolators.
And the region NN approximates within can be quite complex and not easily defined as "X:R^N drawn from N(c,s)^N" as SolidGoldMagiKarp [2] clearly shows.
>CCC shows that AI systems can internalize the textbook knowledge of a field and apply it coherently at scale. AI can now reliably operate within established engineering practice. This is a genuine milestone that removes much of the drudgery of repetition and allows engineers to start closer to the state of the art.
And also
> The most effective engineers will not compete with AI at producing code, but will learn to collaborate with it, by using AI to explore ideas faster, iterate more broadly, and focus human effort on direction and design. Lower barriers to implementation do not reduce the importance of engineers; instead, they elevate the importance of vision, judgment, and taste. When creation becomes easier, deciding what is worth creating becomes the harder problem. AI accelerates execution, but meaning, direction, and responsibility remain fundamentally human.
I figure that all this AI coding might free us from NIH syndrome and reinventing relational databases for the 10th time, etc.
“NIH” here refers to “Not Invented Here” Syndrome, or a bias against things developed externally.
Maybe you can’t teach current LLM backed systems new tricks. But do we have reason to believe that no AI system can synthesize novel technologies. What reason do you have to believe humans are special in this regard?
But sure, instantiating these capabilities in hardware and software are beyond our current abilities. It seems likely that it is possible though, even if we don’t know how to do it yet.
I will argue that the following capacities: 1. creating rules and 2. deciding to follow rules (or not) are themselves controlled by rules.
We've only had the tech to be able to research this in some technical depth for a few decades (both scale of computation and genetics / imaging techniques).
Even skin cells exchange information in neuron-like manner, including using light, albeit thousands times slower.
This switches complexity of human brain to "86 billions quantum computers operating thousands of small neural networks, exchanging information by lasers-based optical channels."
> unstated assumption that technological progress towards human-like intelligence is in principle possible. In reality, we do not know.
For me this isn’t an assumption, it’s a corollary that follows from the Church-Turing thesis.
The claim being made is not "no computer will ever be able to adapt to and assist us with new technologies as they come out."
The claim being made is "modern LLMs cannot adapt to and assist us with new technologies until there is a large corpus of training data for those technologies."
Today, there exists no AI or similar system that can do what is being described. There is also no credible way forward from what we have to such a system.
Until and unless that changes, either humans are special in this way, or it doesn't matter whether humans are special in this way, depending on how you prefer to look at it.
> That's irrelevant.
My comment was relevant, if a bit tangential.
Edit: I also want to say that our attitude toward machine vs. human intelligence does matter today because we’re going to kneecap ourselves if we incorrectly believe there is something special about humans. It will stop us from closing that gap.
For example, my company makes a new framework, and we have a skill we can point an agent at. Using that skill, it can one-shot fairly complicated code using our framework.
The skill itself is pretty much just the documentation and some code examples.
How long can you keep adding novel things into the start of every session's context and get good performance, before it loses track of which parts of that context are relevant to what tasks?
IMO for working on large codebases sticking to "what the out of the box training does" is going to scale better for larger amounts of business logic than creating ever-more not-in-model-training context that has to be bootstrapped on every task. Every "here's an example to think about" is taking away from space that could be used by "here is the specific code I want modified."
The sort of framework you mention in a different reply - "No, it was created by our team of engineers over the last three years based on years of previous PhD research." - is likely a bit special, if you gain a lot of expressibility for the up-front cost, but this is very much not the common situation for in-house framework development, and could likely get even more rare over time with current trends.
Today, yes. I assume in the future it will be integrated differently, maybe we'll have JIT fine-tuning. This is where the innovation for the foundation model providers will come in -- figuring out how to quickly add new knowledge to the model.
Or maybe we'll have lots of small fine tuned models. But the point is, we have ways today to "teach" models about new things. Those ways will get better. Just like we have ways to teach humans new things, and we get better at that too.
A human seeing a new programming language still has to apply previous knowledge of other programming languages to the problem before they can really understand it. We're making LLMs do the same thing.
LLMs are really good at doing that. Arguably better than humans at RTFM and then applying what's there.
Models can RTFM (and code) and do novel things, demonstrably so.
Funny, I'd say the same thing about traditional programming.
Someone from K&R's group at Bell Labs, straight out of 1972, would have no problem recognizing my day-to-day workflow. I fire up a text editor, edit some C code, compile it, and run it. Lather, rinse, repeat, all by hand.
That's not OK. That's not the way this industry was ever supposed to evolve, doing the same old things the same old way for 50+ years. It's time for a real paradigm shift, and that's what we're seeing now.
All of the code that will ever need to be written already has been. It just needs to be refactored, reorganized, and repurposed, and that's a robot's job if there ever was one.
Not to mention you're probably also using source control, committing code and switching between branches. You have unit tests and CI.
Let's not pretend the C developer experience is what it was 30 years ago, let alone 50.
Reply due to rate limiting:
K&R didn't know about CI/CD, but everything else you mention has either existed for 30+ years or is too trivial to argue about.
Conversely, if you took Claude Code or similar tools back to 1996, they would grab a crucifix and scream for an exorcist.
I think you're taking for granted the massive productivity boost that happened even before today's era of LLM agents.
A vice president at Symbolics, the Lisp machine company at their peak during the first AI hype cycle, once stated that it was the company's goal to put very large enterprise systems within the reach of small teams to develop, and anything smaller within the reach of a single person.
And had we learned the lessons of Lisp, we could have done it. But we live in the worst timeline where we offset the work saved with ever worse processes and abstractions. Hell, to your point, we've added static edit-compile-run cycles to dynamic, somewhat Lisp-like languages (JavaScript)! And today we cry out "Save us, O machines! Save us from the slop we produced that threatens to make software development a near-impossible, frustrating, expensive process!" And the machines answer our cry by generating more slop.
First, I disagree that good code is required in any sense. We have decades of experience proving that bad code can be wildly successful.
Second, has the author not seen the METR plot? We went from: LLMs can write a function to agents can write working compilers in less than a year. Anyone who thinks AGI is far away deserves to be blindsided.
Also (and this is coming from someone who thinks it's quite close) "AGI" is not implied by the ability to implement very-long-horizon software tasks. That's not "general" at all.
But of course writing code directly will always maintain the benefit of specificity. If you want to write instructions to a computer that are completely unambiguous, code will always be more useful than English. There are probably a lot of cases where you could write an instruction unambiguously in English, but it'd end up being much longer because English is much less precise than any coding language.
I think we'll see the same in photo and video editing as AI gets better at that. If I need to make a change to a photo, I'll be able to ask a computer, and it'll be able to do it. But if I need the change to be pixel-perfect, it'll be much more efficient to just do it in Photoshop than to describe the change in English.
But much like with photo editing, there'll be a lot of cases where you just don't need a high enough level of specificity to use a coding language. I build tools for myself using AI, and as long as they do what I expect them to do, they're fine. Code's probably not the best, but that just doesn't matter for my case.
(There are of course also issues of code quality, tech debt, etc., but I think that as AI gets better and better over the next few years, it'll be able to write reliable, secure, production-grade code better than humans anyway.)
Unless the defect rate for humans is greater than LLMs at some point. A lot of claims are being made about hallucinations that seem to ignore that all software is extremely buggy. I can't use my phone without encountering a few bugs every day.
The reality is we have built complex organizational structures around the fact that humans also make mistakes, and there's no real reason you can't use the same structures for AI. You have someone write the code, then someone does code review, then someone QAs it.
Even after it goes out to production, you have a customer support team and a process for them to file bug tickets. You have customer success managers to smooth over the relationships with things go wrong. In really bad cases, you've got the CEO getting on a plane to go take the important customer out for drinks.
I've worked at startups that made a conscious decision to choose speed of development over quality. Whether or not it was the right decision is arguable, but the reality is they did so knowing that meant customers would encounter bugs. A couple of those startups are valuable at multiple billions of dollars now. Bugs just aren't the end of the world (again, most cases - I worked on B2B SaaS, not medical devices or what have you).
This is broadly true, but not comparable when you get into any detail. The mistakes current frontier models make are more frequent, more confident, less predictable, and much less consistent than mistakes from any human I'd work with.
IME, all of the QA measures you mention are more difficult and less reliable than understanding things properly and writing correct code from the beginning. For critical production systems, mediocre code has significant negative value to me compared to a fresh start.
There are plenty of net-positive uses for AI. Throwaway prototyping, certain boilerplate migration tasks, or anything that you can easily add automated deterministic checks for that fully covers all of the behavior you care about. Most production systems are complicated enough that those QA techniques are insufficient to determine the code has the properties you need.
my experience literal 180 degrees from this statement. and you don’t normally get the choose humans you work with, some you may be involved in the interview process but that doesn’t tell you much. I have seen so much human-written code in my career that, in the right hands, I’ll take (especially latest frontier) LLM written code over average human code any day of the week and twice on Sunday
Citation needed.
To be lenient I will separate out bugs caused by insufficient knowledge as not being failures in reasoning, do you have forms of bugs that you think are more common and are not arguably failures in reasoning that should be considered?
on edit: insufficient knowledge that I might not expect a competent developer to have is not a failure in reasoning, but a bug caused by insufficient knowledge that I would expect a competent developer in the problem space to have is a failure in reasoning, in my opinion on things.
Maybe I should just retire a few years early and go back to fixing cars...
Maybe in the future us olds will get more credit when apps fall over and the higher ups realize they actually need a high-powered cleaner/fixer, like the Wolf in Pulp Fiction.
Meanwhile I’m moving at about half the speed with a more hands on approach (still using the bots obviously) but my code quality and output are miles ahead of where I was last year without sacrificing maintain ability and performance for dev speed
Maybe the current allocation of technical talent is a market failure and disruption to coding could be a forcing function for reallocation.
I believe the same pattern is inevitable for these higher level abstractions and interfaces to generate computer instructions. The language use must ultimately conform to a rigid syntax, and produce a deterministic result, a.k.a. "code".
Electric Clojure: https://electric.hyperfiddle.net/fiddle/electric-tutorial.tw...
I have not really found anything that shakes these people down to their core. Any argument or example is handwaved away by claims that better use of agents or advanced models will solve these “temporary” setbacks. How do you crack them? Especially upper management.
You let them play out. Shift-left was similar to this and ultimately ended in part disaster, part non-accomplishment, and part success. Some percentage of the industry walked away from shift-left greatly more capable than the rest, a larger chunk left the industry entirely, and some people never changed. The same thing will likely happen here. We'll learn a lot of lessons, the Overton window will shift, the world will be different, and it will move on. We'll have new problems and topics to deal with as AI and how to use it shifts away from being a primary topic.
Edit: I've googled it and I can't find anything relevant. I've been working in software for 20+ years and read a myriad things and it's the first time I hear about it...
Over the course of about 2 years, the general consensus has shifted from "it's a fun curiosity" to "it's just better stackoverflow" to "some people say it's good" to "well it can do some of my job, but not most of it". I think for a lot of people, it has already crossed into "it can do most of my job, but not all of it" territory.
So unless we have finally reached the mythical plateau, if you just go by the trend, in about a year most people will be in the "it can do most of my job but not all" territory, and a year or two after that most people will be facing a tool that can do anything they can do. And perhaps if you factor in optimisation strategies like the Karpathy loop, a tool that can do everything but better.
Upper managment might be proven right.
I would do the initial research/planning/etc. mostly honestly and fairly. I'd find the positives, build a real roadmap and lead meetings where I'd work to get people onboard.
Then I'd find the fatal flaw. "Even though I'm very excited about this, as you know, dear leadership, I have to be realistic that in order to do this, we'd need many more resources than the initial plan because of these devastating unexpected things I have discovered! Drat!"
I would then propose options. Usually three, which are: Continue with the full scope but expand the resources (knowing full well that the additional resources required cannot be spared), drastically cut scope and proceed, or shelve it until some specific thing changes. You want to give the specific thing because that makes them feel like there's a good, concrete reason to wait and you're not just punting for vague, hand-wavy reasons.
Then the thing that we were waiting on happens, and I forget to mention it. Leadership's excited about something else by that point anyway, so we never revisit dumb project again.
Some specific thoughts for you:
1. Treat their arguments seriously. If they're handwaving your arguments away, don't respond by handwaving their arguments away, even if you think they're dumb. Even if they don't fully grasp what they're talking about, you can at least concede that agents and models will improve and that will help with some issues in the future.
2. Having conceded that, they're now more likely to listen to you when you tell them that while it's definitely important to think about a future where agents are better, you've got to deal with the codebase right now.
3. Put the problems in terms they'll understand. They see the agent that wrote this feature really quickly, which is good. You need to pull up the tickets that the senior developers on the team had to spend time on to fix the code that the agent wrote. Give the tradeoff - what new features were those developers not working on because they were spending time here?
4. This all works better if you can position yourself as the AI expert. I'd try to pitch a project of creating internal evals for the stuff that matters in your org to try with new models when they come out. If you've volunteered to take something like that on and can give them the honest take that GPT-5.5 is good at X but terrible at Y, they're probably going to listen to that much more than if they feel like you're reflexively against AI.
Where the tech argument doesn't apply to upper management, business practices, the need to "not be left behind" and leap at anything that promises reducing headcount without reducing revenue, money talks. As long as it's possible to slop something together, charge for it, and profit, slop will win.
What you are seeing here is that many are attempting to take shortcuts to building production-grade maintainable software with AI and now realizing that they have built their software on terrible architecture only to throw it away, rewriting it with now no-one truly understanding the code or can explain it.
We have a term for that already and it is called "comprehension debt". [0]
With the rise of over-reliance of agents, you will see "engineers" unable to explain technical decisions and will admit to having zero knowledge of what the agent has done.
This is exactly happening to engineers at AWS with Kiro causing outages [1] and now requiring engineers to manually review AI changes [2] (which slows them down even with AI).
[0] https://addyosmani.com/blog/comprehension-debt/
[1] https://www.theguardian.com/technology/2026/feb/20/amazon-cl...
[2] https://www.ft.com/content/7cab4ec7-4712-4137-b602-119a44f77...
I've had to work on multiple legacy systems like this where the original devs are long gone, there's no documentation, and everyone at the company admits it's complete mess. They send you off with a sympathetic, "Good luck, just do the best you can!"
I call it "throwing dye in the water." It's the opposite of fun programming.
On the other hand, it often takes creativity and general cleverness to get the app to do what you want with minimally-invasive code changes. So it should be the hardest for AI.
While publicly they might say this is AI driven, I think that’s mostly BS.
Anyway, that doesn’t take away from your point, just adds additional context to the outages.
This isn't any different than the "person who wrote it already doesn't work here any more".
> now requiring engineers to manually review AI changes [2] (which slows them down even with AI).
What does this say about the "code review" process if people cant understand the things they didn't write?
Maybe we have had the wrong hiring criteria. The "leet code", brain teaser (FAANG style) write some code interview might not have been the best filter for the sorts of people you need working in your org today.
Reading code, tooling up (debuggers, profilers), durable testing (Simulation, not unit) are the skill changes that NO ONE is talking about, and we have not been honing or hiring for.
No one is talking about requirements, problem scoping, how you rationalize and think about building things.
No one is talking about how your choice of dev environment is going to impact all of the above processes.
I see a lot of hype, and a lot of hate, but not a lot of the pragmatic middle.
Yeah but that takes years to play out. Now developers are cranking out thousands of lines of “he doesn’t work here anymore” code every day.
https://www.invene.com/blog/limiting-developer-turnover has some data, that aligns with my own experience putting the average at 2 years.
I have been doing this a long time: my longest running piece of code was 20 years. My current is 10. Most of my code is long dead and replaced because businesses evolve, close, move on. A lot of my code was NEVER ment to be permanent. It solved a problem in a moment, it accomplished a task, fit for purpose and disposable (and riddled with cursing, manual loops and goofy exceptions just to get the job done).
Meanwhile I have seen a LOT of god awful code written by humans. Business running on things that are SO BAD that I still have shell shock that they ever worked.
AI is just a tool. It's going from hammers to nail guns. The people involved are still the ones who are ultimately accountable.
Valuable? Yep. World changing? Absolutely. The domain of people who haven't the slightest clue what they're doing? Not unless you enjoy lighting money on fire.
I interpret non-deterministic here as “an LLM will not produce the same output on the same input.” This is a) not true and b) not actually a problem.
a) LLMs are functions and appearances otherwise are due to how we use them
b) lots of traditional technologies which have none of the problems of LLMs are non-deterministic. E.g., symbolic non-deterministic algorithms.
Non-determinism isn’t the problem with LLMs. The problem is that there is no formal relationship between the input and output.
I remember being aghast at all the incomprehensible code and "do not modify" comments - and also at some of the devs who were like "isn't this great?".
I remember bailing out asap to another company where we wrote Java Swing and was so happy we could write UIs directly and a lot less code to understand. I'm feeling the same vibe these days with the "isn't it great?". Not really!
"In order to make machines significantly easier to use, it has been proposed (to try) to design machines that we could instruct in our native tongues. this would, admittedly, make the machines much more complicated, but, it was argued, by letting the machine carry a larger share of the burden, life would become easier for us. It sounds sensible provided you blame the obligation to use a formal symbolism as the source of your difficulties. But is the argument valid? I doubt."
If you compare great writing to programming then basically most of the industry vanishes and a few programming poets survive.