The philosophy still works, you just have to change your view. Instead of trying to work side by side with the agent on every turn (inside of your IDE), instead the agent performs a unit of work and then you review it. You can use your IDE to view the diff, or another diffing tool.
If you've dug in sufficiently on plan mode, then what the agent is executing is not a surprise and shouldn't need input. If it does, the plan was insufficient and/or the context around the request (agents.md, lessons.md, or whatever tools and documents you use ) weren't sufficient.
EDIT: Maybe it doesn't work in cursor, but I continue to use vscode to review diffs and dig in on changes.
Honest question: Why does it matter? They got the product shipped and got millions of paying customers and totally revolutionized their business and our industry.
Engineers using LOC as a measure of quality is the inverse of managers using LOC as a measure of productivity.
More code means more entropy, more room for bugs, harder to find issues, more time to fix, more attack surface, more memory used, more duplication, more inconsistencies... I bet you at some point we'll get someone reporting how AI performance deteriorates as the code base grows, and some blog post about how their team improved the success of their AI by trimming the code base down to less than 100k LOC or something like that.
The principles of good software don't suddenly vanish just because now it's a machine writing the code instead of a human, they still have to deal with the issues humans have for more than half a century. The history of programming is new developers coming up with a new paradigm, then rediscovering all the issues that the previous generation had figured out before them.
The history of programming is also each generation writing far less performant code than the one before it. The history of programming is each generation bemoaning the abstractions, waste and lack of performance of the code of the next generation.
It turns out that there is a tradeoff in code between velocity and quality that smart businesses consider relative to hardware cost/quality. The businesses that are outcompeting others are rarely those who have the highest quality code, but rather those that are shipping quickly at a quality level that is satisfactory for current hardware.
> far less performant code than the one before it.
That worked because of rapid advancements in CPU performance. We’ve left that era.
It’s about more than performance. Code is and always has been a liability. Even with agents, you start seeing massive slowdowns with code base size.
It’s why I can nearly one shot a simple game for my kid in 20 minutes with Claude, but using it at work on our massive legacy codebase is only marginally faster than doing it by hand.
You asked why the size of the code matters, I gave you the answer. If you want to ramble about the non technical aspects of software development talk to someone else, I'm not interested.
I asked a rhetorical question to get the reader to think about a topic. I was not looking for a rote recitation of a well-known textbook answer. Maybe you should not be on the comment section of an engineering website if you find discussion so offensive.
The reason it’s not useful as a measure of productivity is because it’s measure of complexity (not directly, but it’s correlated). But it tells you nothing about whether that complexity was necessary for the functionality it provides.
But given that we know the functionality of Claude Code, we can guess how much complexity should be required. We could also be wrong.
>Why does it matter?
If there’s massively more code than there needs to be that does matter to the end user because it’s harder to maintain and has more surface area for bugs and security problems. Even with agents.
Among the hundreds of thousands of lines of code that Anthropic produced was one that leaked the source code. It is likely to be a config file, not part of the Claude Code software itself, but it still something to track.
The more lines of code you have the more likely there is for one of them to be wrong and go unnoticed. It results in bugs, vulnerabilities,... and leaks.
It will be exactly that. But that is a 'them' problem. I can look at it a go 'that looks like a bad idea' but they are the ones who have to live with it.
At some point someone will probably take their LLM code and repoint it at the LLM and say 'hey lets refactor this so it uses less code is easier to read but does the same thing' and let it chrun.
One project I worked on I saw one engineer delete 20k lines of code one day. He replaced it with a few lines of stored procedure. That 20k lines of code was in production for years. No one wanted to do anything with it but it was a crucial part of the way the thing worked. It just takes someone going 'hey this isnt right' and sit down and fix it.
When a TUI requires 68 GB of RAM to run, or when they spend a week not being able to find a bug that causes multiple people to immediately run out of tokens, it's not a "them" problem.
It's not that a leader is capable of long-term planning, it's that a system is. I am a big proponent for democracy, but the fact is simple that when you do a massive regime change every ~4 years, nothing big will get done. You have about 2 good years to do something, and most big projects simply require more time than that.
China, unlike the US, can look 10 years into the future and consistently execute towards a goal. That's not because of leaders, it's because the systems are fundamentally designed this way.
It's like the two party system in the US. It's because of first past the post in the Constitution. The system is designed to do this, so it does it. The US is designed to be unable to plan or execute long term vision.
I 100% agree for the fact that biggest systematic root problem USA has is first past the post voting system.
Though how did US managed to be long term thinking since world wars up to ~1980s or 90s? Was it just generational trauma of world wars that allowed to align opinions between parties? And by trauma I mean some combo of real trauma to not have WW2 again to the capitalistict and globalistic drive to be world’s hegemony.
It is not even first past the post that is the problem. Even if you had some sort of ranked voting or parliamentary system you would still end up with the same problem. The person in charge gets changed too frequently to be able to have long term plans. 8 years is too short to execute a plan that will take 10 or 20 years.
I think this is why FDR was a successful president and was able to get so much done. He had 3 complete terms and a partial 4th term.
If you are going to have shorter terms you need to have your successor continue with your plans, but in a liberal democracy you don't know who is going to follow you. Even if your party wins, your successor might not continue with the plan.
In preferential Systems you must chase the centre, in the USA it might be trains versus cars, in Australia because both sides are chasing the centre they will both agree on a train line it's just the specifics they will argue about which believe leads to better outcomes.
At least from my experience I would say change of government won't lead to cancellation of a project or reform just an expansion or contraction in scope.
You can't do something like implement a 1 child policy and stick to it for decades causing a demographic collapse because it wouldn't have broad appeal from the population.
Your last paragraph is my point. Some policies may be good, but not popular. (I'm not suggesting the one child policy is good). How would you be able to continue a policy like the one child policy in a democracy for decades? You wouldn't be able to. With China since their leaders are there longer and because the leaders have a more consistent world view they were able to continue with such a policy.
Modern music has done this to itself. When the human product is already pure corporate slop, it's not hard for AI to compete.
Hopefully AI outcompeting humans at slop sparks a renaissance of humans creating truly beautiful human artwork. And if it doesn't, then was anything of value truly lost?
How much of your super-awesome bandcamp music is topping charts, selling millions, packing mega stadiums, and is penetrating the zeitgeist so deeply that people around the world are addicted to it?
Maybe, just maybe, I'm not talking about "my" music tastes, but offering commentary on the state of music at a global scale. Weird that this point was so hard to follow!
So true. AI music gens like Suno can't do Paul Shapera works even remotely, but can recreate a lot of pop or EDM music very faithfully. There's just no distance to close, it's already mainstreamly bad.
> Modern music has done this to itself. When the human product is already pure corporate slop, it's not hard for AI to compete.
What are you talking about? There’s lots of modern music that’s not corporate slop and that’s absolutely great. Never in history was access to great music as easy as it is now.
I'm talking about modern music. Just because a couple of dweebs on hackernews have "totally amazing underground music" doesn't mean the overall zeitgeist agrees. Regardless of your esoteric music tastes, music by sales and music by charting tells a very different story. And that story is one of replaceable slop.
So find music you like that isn't modern corporate slop. My music right now consists mainly of indie stuff I've found on youtube and daft punk. No plagiarism machine needed, just human-made music
"No plagiarism machine needed, just human-made music"
From wikipedia: Many Daft Punk songs feature vocals processed with effects and vocoders including Auto-Tune, a Roland SVC-350 and the Digitech Vocalist. Bangalter said: "A lot of people complain about musicians using Auto-Tune. It reminds me of the late '70s when musicians in France tried to ban the synthesiser. They said it was taking jobs away from musicians. What they didn't see was that you could use those tools in a new way instead of just for replacing the instruments that came before. People are often afraid of things that sound new."
Did Daft Punk put in a lot of effort to remix existing sounds to make their own music? Yes. Did they type "pls make french house electronic music number 1 chart" into a text box? No. Did they also credit original authors? Yes. I've not gone through their whole library, but for example, Edwin Birdsong has songwriting credit for harder, better, faster, stronger
There's this fallacy with AI generation that people think that all you have to do is type "i lik musik pls remake favrite song but better" and you get amazing results.
This is patently untrue.
It's like how if a junior engineer and a principal engineer use claude opus 4.6 they get radically different results. The junior doesn't have the taste or knowledge to know good from bad so the AI oversteers and slop is made. The principal has finely tuned sense of taste and deep knowledge, so they aggressively steer the AI at every step. This is also true in other AI domains.
To be absolutely clear: you can't make good AI music. Try all you want. Try the prompt you just wrote. Show and tell. It's not something you're going to be able to do.
>AI has just exposed that many "engineers" are "temporarily embarrassed project managers", which is fine in the sense that it makes it clearer who actually enjoys making things and who just wants the end result regardless of how it's made.
AI has also exposed that many "engineers" are just "people who like fiddling with code" and that's fine in the sense that it makes it clear who are the actual engineers who are engineering solutions to real human problems and who just want to tinker with code.
Like imagine slandering a civil engineer "you just want a bridge that is safe and lasts for a century, you don't care about enjoying the journey of construction".
Haha! Your analogy doesn't work on multiple levels. Firstly, if you're outsourcing your work to AI you're not the engineer anymore. A civil engineer is different from a manager of a civil engineering project. Just like I wouldn't call myself an artist if I got AI to generate me some art, I wouldn't call myself a software engineer if I got AI to write all the code for me.
Secondly, it's not just about "enjoying the journey of construction", it's also about caring about the quality of the end results. Getting vibe coded software that is as stable as a "bridge that is safe and lasts for a century" is not a matter of careful engineering decisions, it's mostly a matter of luck, because you don't have the necessary oversight in the quality of the output unless you're doing extensive reviews of the generated code, at which point you greatly diminish the time you're supposedly saving.
False. If you "outsource your code" to a compiler and just write higher level language, you're not an engineer. You literally don't own any of your own code, just an abstraction of it written in human language. See how that works? An engineer can delegate -- period.
- "I wouldn't call myself a software engineer if I got AI to write all the code for me"
If all you do is write code you're not an engineer. I think you fundamentally don't know what engineering is. In a very real sense engineering is what you do when you're not coding. The civil engineer doesn't construct the bridge personally.
- "Secondly, it's not just about "enjoying the journey of construction", it's also about caring about the quality of the end results".
Codemonkeys DON'T CARE about the quality of the end result. They only care about their little corner of the zen garden. Writing real software for real users is by far the worst part of a codemonkeys job.
- "Getting vibe coded software that is as stable as a "bridge that is safe and lasts for a century" is not a matter of careful engineering decisions, it's mostly a matter of luck"
Nonsense. The engineer who spends 90% of his time architecting systems and testing them at a high level is making safer and more stable software than the codemonkey who spends 90% of his time tinkering with the details. Forest for the trees.
- "unless you're doing extensive reviews of the generated code, at which point you greatly diminish the time you're supposedly saving."
Who said anything about "saving time"? We're engineering high quality systems. Some of us spend our time at a higher level, thinking holistically about the system, testing multiple concepts and rapidly iterating. Others demand bespoke handwritten code and in the time allowed can barely finish a single concept with a questionable amount of polish. Whatever their first idea is will ship, and they'll have no real ability to justify the architecture other than vibes.
Yes and no. Engineering does involve delegation but what defines an engineer is is what work they do, not what work they pass onto others.
If it helps you understand this, consider the role of an engineer as someone that makes engineering decisions. If you give a specification to a colleague and ask them to write code for you, then you're delegating those engineering decisions. When you write high level code, yes you allow a compiler or interpreter to determine how to turn your instructions into machine code, but you have made engineering decisions in order to design the end result. If you give instructions via product specifications, then you have acted as a project manager or business analyst, not as an engineer.
To use another analogy, imagine you are a chef and you go to eat at a restaurant you don't work in. When you order from the menu, you are not a chef at that moment, even if your background suggests you are capable of being one. Similarly, ordering code from an AI agent does not give you the right to call yourself an engineer when doing so, as you did very little of the real engineering work to produce the end result.
> If all you do is write code you're not an engineer.
Engineering requires thought and application of thought, and if you're outsourcing both then you don't qualify as an engineer.
> The engineer who spends 90% of his time architecting systems and testing them at a high level is making safer and more stable software than the codemonkey who spends 90% of his time tinkering with the details.
The devil is in the details. A technical architect that doesn't understand the tradeoffs in the designs they're specifying isn't worth the money they earn.
> Who said anything about "saving time"?
Almost everyone that is selling the benefits of AI. Clearly you haven't been paying attention to industry trends.
> Like imagine slandering a civil engineer "you just want a bridge that is safe and lasts for a century, you don't care about enjoying the journey of construction".
Would you currently trust a bridge designed by a civil engineer using AI for all of their calculations ?
> Would you currently trust a bridge designed by a civil engineer using AI for all of their calculations ?
Not a great comparison. I'd agree with you if it was straight up vibe coding.
But co-creating (which is what I do) I create plan, then step through it with Claude. Claude creates a small part of what I want, I review, tweak or ask Claude why it took that approach if its different.
I know the subject matter of what it is creating, so in this sense it is safe, as long as I am reviewing everything.
It gets dangerous if you just let it create something without any interaction or understanding of what is being created.
>Would you currently trust a bridge designed by a civil engineer using AI for all of their calculations ?
Of course. I've seen how sloppy and lazy humans are, and I already use the bridge, and if the safety truly came down to the output of single person, then the risk is already significant.
I must say, I got a chuckle at "using AI to do their calculations". Oh no, my agent is going to write a python script to do basic maths, and check their work against a series of automated tests, the sky is falling!
I'm convinced that most programmers largely hate making software and solving user problems. They just like fiddling with code. And honestly it explains so much.
The prompt isn't very useful. You'd see the exact same prompt on every ticket for me.
Prompt 1: "Research <X> domain, think deeply, and record a full analysis in /docs/TICKET-123-NOTES.md"
Prompt 2: Based on our research, read TICKET-123 and began formulating solutions. Let's think this problem through and come up with multiple potential solutions. Document our solutions in TICKET-123-SOLUTIONS.md
Prompt 3: Based on Solution X, let's formulate a complete plan to implement. Break the work into medium sized tasks that a human could complete in 5-10 hours. Write our plan in TICKET-123-PLAN.md
I've often thought that some of this metadata, such as the research, solutioning and plan could be shared. I think they're valuable for code review. I've also translated these artifacts into other developer documentation paradigms.
But the prompts? You're not getting a lot of value there.
> Prompt 1: "Research <X> domain, think deeply, and record a full analysis in /docs/TICKET-123-NOTES.md"
> Prompt 2: Based on our research, read TICKET-123 and began formulating solutions. Let's think this problem through and come up with multiple potential solutions. Document our solutions in TICKET-123-SOLUTIONS.md
> Prompt 3: Based on Solution X, let's formulate a complete plan to implement. Break the work into medium sized tasks that a human could complete in 5-10 hours. Write our plan in TICKET-123-PLAN.md
Sounds to me that all these 10x - 100x "engineers" can be removed from the loop.
Almost! We are certainly on the precipice of the vast majority of white collar work being removed from the loop.
However, what each domain will tell you (engineering included) is that AI doesn't understand the full context of what you're doing and the point of the business and where to spend effort and where to cut corners. There is definitely still room for competent engineers to iterate here on the solutioning and plans to refine the AI work into something more sturdy.
Although this is only in domains where code quality truly matters. A lot of consumer software without SLA's are just vibe coding full speed now. No code review, AI writing 100% of the code.
What a utopia, where code quality matters in all domains!
In my opinion nearly the opposite is true: modern business solves for the "minimum viable quality". What is the absolute lowest quality the software can be and not tank the business.
If you could prove what "minimum viable quality" actually was this would be true. We have standards and procedures exactly because it is unknowable. One engineers idea of "good enough" might bankrupt the business.
> What a utopia, where code quality matters in all domains!
It does. The degree may not, though.
"We have a threshold of at least 5 hours total uptime every 24 hours" is still a quality bar, even if it is different to "We have a threshold of 99.99% uptime per year".
Maybe you're different, but I prefer to write code that at least attempts to be performant, tidy and readable, as well as working at least 90% of the time. Maybe I don't achieve perfection, but I try to care about the quality of what I write
You seem pretty smart. If suddenly, after over a decade, schizophrenic artifacts appear in one single isolated subject, - a subject well known and documented with equal and greater concerns among highly credible sources - does that perhaps imply that the subject itself may be inducing schizophrenia? Maybe a pathological system is inducing pathological effects? Strangely, I feel fine.
Regardless, gaslight as you will; The public will see the implications, which is that questioning LLMs, to some (you?), is symptomatic of psychological pathology. In my opinion, that level of trust, or Faith, is naive for such a novel but powerful technology.
And the basic premise seems to be: user questions sensitive system attributes. Pathologize user. Imply system is infallible and any doubt suggests mental incapacitation. Point out all possible flaws in user while deflecting any attention to system.
That's tried and true. I wish you luck. Meanwhile, the message becomes clearer and clearer.
Autopilot isn’t full self driving (FSD), most cars these ship with smart cruise control (what autopilot basically is). Do you have fatality statistics for FSD?
If we are just talking about smart cruise control, most cars are using cameras and radar, not lidar yet. But Tesla is special since it doesn’t even use radar for its smart cruise control implementation, so that could make it less safe than other new cars with smart cruise control, but Autopilot was never competing with Waymo.
If you've dug in sufficiently on plan mode, then what the agent is executing is not a surprise and shouldn't need input. If it does, the plan was insufficient and/or the context around the request (agents.md, lessons.md, or whatever tools and documents you use ) weren't sufficient.
EDIT: Maybe it doesn't work in cursor, but I continue to use vscode to review diffs and dig in on changes.
reply