I think veteran engineers have always known that the real problems with velocity have always been more organizational than technical. The inability for the business to define a focused, productive roadmap has always been the problem in software engineering. Constantly jumping to the next shiny thing that yields almost no ROI but never allowing systemic tech debt to be addressed has crippled many company's I have worked at in the long-term.
> The inability for the business to define a focused, productive roadmap has always been the problem in software engineering.
Agreed, and I also agree that most developers come to this realization with time and experience. When you have a clear understanding of business rationale, scope, inputs, and desired outputs, the data models, system design and the code fall out almost naturally. Or at least are much more obvious.
- systemic tech debt is now addressable at scale with LLMs. Future models will be good enough to sustain this, if people don’t believe this I would challenge them to explain why. First consider if you understand what scaling laws are like chinchilla and how RL with verification works fundamentally
- I completely agree with you about fundamentally the limitation being the business able to coherently articulate itself and its strategy
- BUT the benefit now is you can basically prototype for free. Before we had to be extremely careful with engineer headcount investment. Now we can try many more things under the same time constraints.
> systemic tech debt is now addressable at scale with LLMs.
Is there any reason to believe this? I've only seen the evidence of the contrary so far.
My experience with AI coding aides is that they, generally:
1. Don't have an opinion.
2. Are trained on code written using practices that increase technical debt.
3. Lack in the greater perspective department, more focused on concrete, superficial and immediate.
I think, I need to elaborate on the first and explain how it's relevant to the question. I'll start with an example. We have an AI reviewer and recently had migrated a bunch of company's repositories from Bitbucket to GitLab. This also prompted a bunch of CI changes. Some projects I'm involved with, but don't have much of an authority, that are written in Python switched to complicated builds that involve pyproject.toml (often including dynamic generation of this cursed file) as well as integration with a bunch of novelty (but poor quality) Python infrastructure tools that are used for building Python distributalbe artifacts.
In the projects where I have an authority, I removed most of the third-party integration. None of them use pyproject.toml or setup.cfg or any similar configuration for the third-party build tool. The project code contains bespoke code to build the artifacts.
These two approaches are clearly at odds. A living and breathing person would either believe one to be the right approach or the other. The AI reviewer had no problems with this situation. It made some pedantic comments about the style and some fantasy-impossible-error-cases, but completely ignored the fact that moving forward these two approaches are bound to collide. While it appears to have an opinion about the style of quotation marks, it completely doesn't care about strategic decisions.
My guess as to why this is the case is that such situations are genuinely rarely addressed in code review. Most productive PRs, from which an AI could learn, are designed around small well-defined features in the pre-agreed upon context. The context is never discussed in PRs because it's impractical (it would usually require too much of a change, so the developers don't even bring up the issue).
And this is where real large glacier-style deposits of tech debt live. It's the issues developers are afraid of mentioning because of the understanding that they will never be given authority and resources to deal with.
You are not wrong about anything you’re saying but like I said this misses the forest for the trees. I’m talking about like the next ~2 years. There is a common idea that we don’t understand this technology or what will happen performance wise. We know a lot more about what’s going to happen than people think. It’s because none of this is new. We’ve known about neural nets since the 40s, we know how RL works on a fundamental level and it has been an active and beautiful field of research for at least 30-40 years, we know what happens when you combine RL with verifiable rewards and throw a lot of compute at it.
One big misconception is that these models are trained to mimic humans and are limited by the quality of the human training data, and this is not true and also basically almost entirely the reason why you have so much bullishness and premature adoption of agentic coding tools.
Coding agents use human traces as a starting point. You technically don’t have to do this at all but that’s an academic point, you can’t do it practically (today). The early training stages with human traces (and also verified synthetic traces from your last model) get you to a point where RL is stable and efficient and push you the rest of the way. It’s synthetic data that really powers this and it’s rejection sampling; you generate a bunch of traces, figure out which ones pass the verification, and keep those as training examples.
So because
- we know how this works on a fundamental level and have for some time
- human training data is a bootstrap it’s not a limitation fundamentally
- you are absolutely right about your observations yet look at where you are today and look at say Claude sonnet 3.x. It’s an entire world away in like a year
- we have imperfect benchmarks all with various weaknesses yet all of them telling the same compelling story. Plus you have adoption numbers and walled garden data that is the proof in the pudding
The onus is on people who say “this is plateauing” or “this has some fundamental limitation that we will not get past fairly quickly”.
> [O]rganizations which design systems (in the broad sense used here) are constrained to produce designs which are copies of the communication structures of these organizations.
Any competent engineer should understand that engineering is just the assembly line side of product development. Deciding when to release which feature, bug fixes, etc. and the development/management of the product in general has always been the real challenge, and a lot of the strategy involved in doing this relies on feedback loops that AI cannot speed up. Though at the same time I do feel like leaders on the business side often scapegoat engineer's speed as an excuse instead of taking responsibility for poor decisions on their end.
I get what youre trying to say but this is actually a bad picture to defend. product and engineering should go hand in hand, with one side informing the other. Engineers sctually giving a shit about a product will tell product possibilities they havent even considered, product people caring about engineering will not propose utterly stupid things. and I for one can spot when a product is well designed but poorly made, as well as when a product is perfectly crafted yet useless. the sweetspot is both. and even with the speed multiplier of AI, having a proud in the craft and being actually good in it as an engineer makes a night and day difference for the final result.
> I think veteran engineers have always known that the real problems with velocity have always been more organizational than technical.
I don't think this comment is fair or grounded. There are plenty of process bottlenecks that are created by developers. Unfortunately I have a hefty share of war stories where a tech lead's inability to draft a coherent and clear design resulted in project delays and systems riddled with accidental complexity required to patch the solution enough to work.
Developers are a part of the process and they are participants of both the good parts and the bad parts. If business requirements are not clear, it's the developer's job to work with product owners to arrive at said clarity.
> Unfortunately I have a hefty share of war stories where a tech lead's inability to draft a coherent and clear design resulted in project delays and systems riddled with accidental complexity required to patch the solution enough to work
This is also an organizational problem (bad hiring/personal management). If you put an incompetent individual at the helm of a project, then resources (especially time) will be spent horrendously and you will have more problems down the line. That’s true for all type of organizations and projects.
yes, most places I have worked were hobbled by the organizations being completely idiotic.
which is why engineers want to be left alone to code, historically. Better to be left alone than dealing with insane bureaucracy. But even better than that is working with good bureaucracy. Just, once you know it's insane, there's not really anything that you can personally do about it, so you check out and try to hold onto a semblance of sanity in the realm you have control over, which is the code.
It’s part of the problem but AI also can crush this on pure lines of code and functionality alone. It can put out 100,000 lines of somewhat decent code in a day. That usually takes months or years of manual coding for a team.
There is a reason that kLOC / FP were rightly shunned out of being measurable metrics years ago.
The same clown show seems to be resurging with "tokens".
There is, in my opinion, no real formula or metric that you can define for "good" code or "bad" code.
Tickets and ceremonial activities, however abstract that into a N-nary status value that seems easier to judge upon.
And now they're almost forcing us to produce machine-made tech-debt at an industrial scale. The AI craze isn't going to produce the boon some people think it will. And the solution? More AI, unfortunately.
No. It's not more AI. The solution is designing and sticking to development process that is more resilient to errors than the one that's currently happening. This isn't a novel idea. Code reviews weren't always part of the process, neither was VCS, nor bug tracker etc.
The way AI is set up today, it's trying to replicate the (hopefully) good existing practices. Possibly faster. The real change comes from inventing better practices (something AI isn't capable of, at least not the kind of AI that's being sold to the programmers today).
What better practices do you mean? Are you saying we just need different more agentic-friendly practices that ensure scaled reliability beyond what we can manually check? If so I totally agree.
AI is 100% capable fundamentally of making new processes. Look I mean it’s not like I think opus 4.7 is all you need, but how can you argue with the fact that adoption since 4.5 has been an inflection point? That’s kind of proof that reliability has reached a level that serious usage is possible. That’s over a period of months. When you zoom out further you see this is extremely predictable even a few years ago, despite the absolute hissy fits thrown on HN when CEOs began saying this.
Agentic coding is verifiable and this implies there are very few practical limits to what it can do. Combine that with insanely active research on tackling the remaining issues (hallucinations — which are not a fundamentally unsolvable problem at a practical level, context rot, continual learning etc)
I want this question to have an interesting answer, but everyone knows that if this question ever goes to the courts, ownership will go to the people in charge with the money. The idea that Anthropic may not own Claude Code just because Claude wrote it is wishful thinking.
The work-for-hire doctrine actually supports your intuition more than the AI authorship question does. The reason Anthropic likely owns Claude Code has little to do with whether Claude wrote it and everything to do with the employment contracts of the engineers who directed it. The DMCA takedown question is genuinely interesting though because DMCA requires the claimant to assert copyright ownership in good faith. If a court later found the codebase was predominantly AI-authored and therefore not copyrightable, the 8,000 takedowns could be challenged as bad faith DMCA claims. That is a different and more tractable legal question than the ownership one.
Work-for-hire doctrine doesnt automagically absolve you from IP law. Microsoft and Intel already learned this in the nineties when they paid San Francisco Canyon Company to steal Apple code.
The San Francisco Canyon case is a good example of exactly the right distinction. Work-for-hire determines who owns the output, but if the process of creating that output involved copying protected material, the infringement claim runs separately. The piece makes this point on the open source contamination section: owning the output and having a clean chain of title to the output are different questions. You can own AI-generated code and still have a copyleft problem in it.
I have trouble believing that the DMCA claims would be found to be in bad faith when they were made at a time when the question of what degree of human input is required to acquire copyright on AI generate code hasn't been resolved at all.
It doesn't seem like bad faith to think that copyright is stronger than the courts end up thinking, just being mistaken.
fair correction, updated the piece to reflect this. Bad faith under DMCA requires knowing the claim is false, not merely being wrong. A good faith belief in copyright ownership, even one that turns out to be mistaken, is a defense. The more accurate framing is that if the codebase is found to be predominantly AI-authored, the takedowns would fail on the threshold question of whether there is a valid copyright to assert, which is a different issue from intent.
As a developer, the fact that my source code passed through a compiler - an automated tool - doesn't give the author of the compiler any claim on my executable code.
As an artist, the fact that I used, e.g., Rebelle to paint a digital painting, or that I used Lightroom (including generative AI to fill, or other ML/AI tools to de-noise and sharpen my image) in editing a photograph, doesn't give EscapeMotion, Adobe, or Topaz, any claims to my product.
Why, then, would there be any chance that use of a tool like Claude - a tool that's super-advanced to be sure, but at the end of the day operates by way of a mathematical algorithms - would confer any claims to Anthropic?
If a court later found the codebase was predominantly AI-authored and therefore not copyrightable
Is figuring out the appropriate prompts to use in directing Clause qualitatively different than using a (much) higher-level abstraction in coding? That is, there was never any talk as we climbed the abstraction layer from machine code to assembly to Fortran or C to 4GLs to Rust etc., that the assembler/compiler/IDE builder would have any ownership claim on the produced executable. In what sense can Anthropic et al assert that their tool, which just transforms our directives to some lower-level representation, creates ownership of that lower-level representation?
Best part is, it's likely to have a different answer in every country, who knows what'll happen, not every country implicitly sides with the ones with the most money.
It's not wishful thinking, and ownership isn't a foregone conclusion.
Sure the courts could mint a communist society with a few weird decisions about property rights, but this being the US do you really suppose that's likely?
There's really no legal question of any kind that models aren't people and therefore cannot own property (and also cannot enter into legal contract as would be required to reassign the intellectual property they don't and can't own)
The catch-22 is that the fact that models aren't people is only relevant if you treat them similar to a person. Like the US Copyright Office's opinion which treats it similar to a freelancer. If you treat the LLM as a machine similar to a camera, with the author expressing their existing intent through the tools of this machine, ownership is back on the table and more or less how it was before LLMs.
Well if the camera in addition to choosing autoexposure also decided how to frame the shots, which lens to use, where to stand, and everything else salient to the artistry of photography -- all without direct human intervention, then I would think the situation would again be analogous. If the camera could do all that because an intern was holding it, the intern would still own the shots even if their employer gave them the assignment.
That's why the intern signs an employment contract that reassigns their rights to their employer!!
Too late to edit, but OpenAI certainly doesn't want ownership or liability, for the CSAM they've produced. They certainly don't want ownership/liability of code which does $ONLYAWFULTHING.
They won't want to own code that is malicious\illegal\used in crime, although it's really weird to me that no one (in LEO) seems to care that, for example, grok generates CSAM, revenge porn, probably other illegal things, so they'll probably get to have their cake and eat it too.
Those things have precise legal definitions which it may not be entirely clear that an LLM can even generate them - especially in the USA where the 1st covers things that many would think illegal (and are illegal in other countries).
I've been worried for some time now that genAI will effectively kill the market for dev tools and so we will be stuck with our current dev tools for a long time. If everyone is using LLMs to write code, the only dev tools anyone will use will be the ones that the LLMs use. We will be stuck with NPM forever.
I think the opposite may be true.
If dev tools are broken and it annoys someone, they can more easily build a better architecture, find optimizations and release something that is in all ways better. People have been annoyed with pip forever, but it was the team behind uv that took on pip's flaws as a primary concern and made a better product.
I think having a pain point and a good concept (plus some eng chops) will result in many more dev tools - that may be cause different problems, but in general, I think more action is better than less.
this is exactly what I mean though. Instead of the community building a better tool that we collectively contribute to and work with, genAI is going to silo all the good stuff with individual developers and teams instead. Because its so cheap to create these tools, no one is going to bother publishing new ones for everyone, so we will essentially be stuck with what we have forever now.
What kind of tools do you have on your mind specifically? My experience is that LLM can create me a decent dev tool that I wouldn't ever bother making so nice myself.
It's extremely weird that 40 years after TurboPascal, 30 years after Delphi and VBA, we've only regressed in terms of truly integrated development environments.
Heck, even programming languages have regressed. Python and Javascript are less type safe than Java circa 2005. Even though we have technology needed to make type safe languages much more ergonomic, since then.
I think there is a misunderstanding about the therapeutic effects of psychedelics. The drugs themselves may alter physical structure in your brain a little bit - but what they really do is temporarily give you a different perspective - they change your point of view. That skewing of perspective is (I believe) where the therapeutic effect from these drugs arises.
If you are deeply curious about these types of drugs, you need to remember that they all wear off eventually. Lots of very smart and happy people have taken these drugs and experienced no harm.
This is somewhere between "False" and "So misleading about an astronomically small risk that we should just treat it as False".
Driving or riding in a car is a more likely cause of PTSD - you might be involved in a horrific crash.
Nothing in this world is risk free, but if we dropped the cultural stigma and history, and these were just discovered by Pfizer today and went through regular FDA processes, this class of drugs would have a risk profile lower than SSRIs, benzodiazepines, and most other drugs used for psychiatric purposes.
Have you known at least a few people who have taken psychedelics, then had a chance to see how they are doing in the years afterwards?
The harm is much more apparent to observers than it is to the psychedelic user him or herself.
If I'm wrong about psychedelics, I'm wrong in my claim that they routinely cause PTSD specifically, not about the claim that they routinely cause some kind of long-term harm. I admit that they also often improve people, including people whose psychedelic use was unsupervised. I.e., I'm making a statistical claim, not a categorical one.
I get my PTSD claim from Dr K of the "Healthy Gamer" YT channel, who is a Harvard-train psychiatrist. I can provide a citation if there is interest.
Tales of a person's life and level of functioning steeply declining after taking a psychedelic, then staying that way for years, are common, e.g., on this web site over the years. Here is one example, and yes, I realize that in the same comment section are people who claim to have been helped by psychedelic use.
If we ignore what people say about their own experience with psychedelics and focus only on what people say about people they have known who have taken the drugs, the reports are overwhelming negative unless the reports are by researchers and clinicians reporting on psychedelic use in which the entire experience is supervised by a skilled therapist (which I do not criticize).
P.S., benzos and SSRIs are both bad drugs that do more harm than good, IMHO, so your assertion that psychedelics are better than them is not saying much.
long time lurker but created an account to reply here. i've taken plenty of psychedelics from around 18 y/o on a regular basis (once or twice every other month with frequent breaks of several months and then more intense periods of heavier usage) until i was about 29 and lost interest. i've tried DMT, LSD (my favorite. have done large doses of 800 microgram), different kind of shrooms...
drugs have repeatedly given me profound and connected experiences. it makes you feel connected with people and the world because your ego is reduced and you let everything in your surroundings fill you up instead. your mental barriers and preconcieved notions fall apart and you just accept what is happening around you.
I know several people with hereditary mental health disorders who's ailments have been trigger by drug use but i don't think you can blame the drugs here. a traumatic experience could trigger it too.
while i would not call my self and addict, i was a thrill seeker in my younger years for sure. today i'm a successful SWE, homeowner in a major western city and have a loving partner. plenty of my friends who were with me doing these drugs have similar lives today.
> I know several people with hereditary mental health disorders who's ailments have been trigger by drug use but i don't think you can blame the drugs here. a traumatic experience could trigger it too.
If they had a 50% chance of developing the disorder without taking the drug and an 80% chance of developing it with the drug (for example) of course you can blame it. There has to be some nuance here: these drugs are not nearly as dangerous as many make them out to be, but they are not without risk either. People can be seriously harmed by them, or, more likely, just have a bad time.
> these drugs are not nearly as dangerous as many make them out to be, but they are not without risk either
yes i agree. my girlfriend has never done any drugs except alcohol, weed (handful of times) and prescribed drugs from the dr. i have never recommend her to try psychedelics but I am always honest about what a massive positive impact it had on my life. i would consider myself depressed when i was in my teens. psychedelics (and meditation, philosophy books, and thought provoking conversations) helped me break out of my mental prison. if you treat drugs like a tool, like you would a sharp knife, you can unlock beautiful things -- but the knife might cut you.
just like with most things in life - leaving the safety of your home carries a certain risk. when you're swimming there is a risk you'll drown. bouldering, climbing can cause you to fall and break your neck. driving on the motorway has a relatively high chance of causing you a premature death. i can go on...
the people i know who's mental health issues have been excaberated by drugs are minimal compared to the ones i know who have used drugs and are perfectly normal people. some folks were heavy psychedelic / mdma users but you would never know that if you met them on the street or had a conversation with them...
You are overstating or overgeneralizing the strength of psychedelics as a class of drug. Most people who take them are not taking enough to produce a PTSD-level response.
I developed PTSD after my finding my 3yo son floating in a pool face down (I luckily saved and revived him - he's fine now) and it would take a very intense psychedelic experience to come anywhere close to that kind of emotional content.
Claiming the entire class of drugs are a potent cause of PTSD rings of reefer madness propaganda to me.
>the entire class of drugs are a potent cause of PTSD
That is indeed my claim. More precisely, it is my secondary claim that (like I say in a cousin comment) I am less confident of than my primary claim that psychedelics are a potent cause of some sort of long-term severe harm.
A person's having had PTSD does not automatically make the person an expert on what sorts of experiences can be traumatizing. There is more to it than the just the intensity of the emotions. PTSD is very complicated and difficult to understand (which is why many with PTSD have no clue that they even have it).
Dr K says BTW that it is the loss of the sense of self that can be traumatizing in psychedelic use.
Ehh. I've done mushrooms, lsd, etc. about once to three times a year pretty much my whole adult life (decades). I find it fun. I have a relaxed good time with like minded friends and that's it. I think the whole "mind awakening" nonsense is just as much nonsense as the PTSD or worse folks. Perhaps someone with underlying severe mental health issues might experience things differently. But for folks in a pretty healthy headspace, it's just a recreational drug with extremely low addiction potential and zero hang over. What's not to like?
PTSD is not usually what happens when taken without supervision either. I think there's a large chasm of experiences between lifelong healing and lifelong damage with regard to psychedelics.
I have pretty limited experience with it and came to the conclusion that it's not for me. But of the people I know who do them or have at one time, I don't think I know anyone whose life has been changed by them.
All of the nay-sayers in the comments here are thinking about this from the POV of a person who reached intellectual maturity without LLMs and now use it as a force multiplier, and rightly so.
However, I think that take is too short-sighted and doesn't take into account the effect that these products have on minds that have not yet reached maturity. What happens when you've been using ChatGPT since grade school and have effectively offloaded all the hard stuff to AI through college? Those people won't be using it as a force multiplier - they will be using it to perform basic tasks. Ray-Ban sells glasses now with LLMs built in with a camera and microphone so you can constantly interact with it all day. What happens when everyone has one of these devices and use it for everything?
I think you are looking at this from a too-narrow lens. What happens when people have ChatGPT built into their eyeglasses and they use it for literally everything. Ray-Ban is already selling this as a product.
Boot-camp grads are not self-taught, they went to a boot-camp. Boot-camp people are career opportunists. Nothing wrong with career opportunists, just saying that they are different than a self-taught dev.
Those kinds of thought processes are the kinds that produce value.
Deciding what to build and how to build it is often harder than building.
What LLMs of today do is basically super-autocomplete. It's a continuation of the history of programming automation: compilers, more advanced compilers, IDEs, code generators, LINTers, autocomplete, codeinsight, etc.
reply