> fully understand every change and every single line of the code.

im probably just not being charitable enough to what you mean, but thats an absurd bar that almost nobody conforms to even if its fully handwritten. nothing would get done if they did. But again, my emphasis is on that im probably just not being charitable to what you mean.

You're most likely being pedantic, like when someone says they understand every single line of this code:

    x = 0
    for i in range(1, 10):
      x += i
    print(x)

They don't mean they understand silicon substrate of the microprocessor executing microcode or the CMOS sense amplifiers reading the SRAM cells caching the loop variable.

They just mean they can more or less follow along with what the code is doing. You don't need to be very charitable in order to understand what he genuinely meant, and understanding code that one writes is how many (but not all) professional software developers who didn't just copy and paste stuff from Stackoverflow used to carry out their work.

you drew it to its most uncharitable conclusion for sure, but ya thats pretty much the point i was making.

How deep do i need to understand range() or print() to utilize either, on the slightly less extreme end of the spectrum.

But ya, im pretty sure its a point that maybe i coulda kept to myself and been charitable instead.

Understand your code in this day and age likely means hit the point of deterministic evaluation.

print(X) is a great example. That's going to print X. Every time.

Agent.print(x) is pretty likely to print X every time. But hey, who knows, maybe it's having an off day.

How is that an absurd bar? If you're handwriting code, you'd need to know what you actually want to write in the first place, hence you understand all the code you write. Therefore the code the AI produces should also be understood by you. Anything else than that is indeed vibe coding.

A lot of developers don't actually understand the code they write. Sure nowadays a lot of code is generated by LLMs, but in the past people just copied and pasted stuff off of blogs, Stack Overflow, or whatever other resources they could find without really understanding what it did or how it worked.

Jeff Atwood, along with numerous others (who Atwood cites on his blog [1]) were not exaggerating when the observed that the majority of candidates who had existing professional experience, and even MSc. degrees, were unable to code very simple solutions to trivial problems.

[1] https://blog.codinghorror.com/why-cant-programmers-program/

its an absurd bar if you are being a uncharitable jerk like i was, the layers go deep, and technically i can claim I have never fully grasped any of my code. It is likely just a dumb point to bring up tbh.

I saw your reply to another comment [0], I see what you mean now. By "understand each line of code" I meant that one would know how that for loop works not the underlying levels of the implementation of the language. I replied initially because lots of vibe coding devs in fact do not read all the code before submitting, much less actually review it line by line and understand each line.

[0] https://news.ycombinator.com/item?id=47894279

Well that is how it mostly worked until recently... unless if the developer copied and pasted from stackoverflow without understanding much. Which did happen.

Could they have meant "every line of code being committed by the LLM" within the current scope of work?

That's how I read it, and I would agree with that.

It's a good point. To me this really comes down to the economics of the software being written.

If it's low-stakes, then the required depth to accept the code is also low.

I do. If you don't, maybe you shouldn't be writing software professionally. And yes, I've written both DBs and compilers so I do understand what is happening down to the CMOS. I think what you are doing is just cope.

nah, you're kinda encapsulating what i viewed in my mind:

at what level of abstraction can you claim to actually "understand" the code?

You're claiming to understand down to the CMOS, but you are failing to even engage with what level understanding should be accepted. is "down to the CMOS" the bar? because then you're gonna be on an uphill battle as potentially the only human who traces a simple hello world python script down to it, because thats not how people develop software with high level languages.

is understanding the print()'s underlying code the bar? seems fairly gatekeepy, its kinda intuitive what a print does, everyone trusts its gonna do what its designed to do in the same way we trust the water that comes out of our faucets.

I mean "understanding it just like when you hand wrote the code in 2019."

Obviously I don't mean "understanding it so you can draw the exact memory layout on the white board from memory."

You don't understand every change you make in the PRs you offer for review?

Im locked in for a year of claude pro, I encountered the same issues as you a couple weeks ago, Id get like one solid plan done and really really hope it was a 1 shot because that was legit all i was gonna get out of it for those 5 hours, and it would be ~10% of weekly usage to really make me feel scared to hit send.

I got the 20$ gpt tier, and now i just use claude to craft MD plan docs instead, and then i hand them off to gpt 5.4 and it has been working great. can do about 4x as much work or so based on my feelings(not accurate). if i have just small simple stuff to do i might still fire those off with sonnet and that seems plenty viable, but as soon as its an opus tier task i swap to this workflow.

Little annoying as now im kinda trying to manage a .claude/ and an .opencode/ folder but i kinda just have the .opencode/ stuff reference the .claude/ stuff so its a little less bleh.

I've been keeping within my usage because ive been in a funk a bit, but when i was slightly more worried id sorta just juggle whether claude or gpt would handle writing some initial tests as it did seem to kinda be imbalanced otherwise. seems like gpt just spam resets weekly usage throughout the week anyway so its prolly nbd.

you could try customer support, that chat bot will happily loop you with some more non answers, but try to make you feel good about those non answers :)

This test makes perfect sense with their actions the last few weeks, they think they've done enough to transition into the general public and away from devs and our goodwill no longer is something they should be concerned with.

Its funny that openai, who in my eyes went for the general public rather than devs initially, seems to be semi pivoting and catching all the fallout from anthropic's recent behavior.

It is a massive bummer, up until those few weeks ago, i was hard pulling for anthropic for quite some time, now i just dont care and hope something dope emerges quickly that signals i wont ever have to consider either of them.

I think they will try to get into the enterprise as soon as possible - looking out for big purchases - figma, atlassian, asana, monday, etc...

gpt allows you to wire their models into other CLI tools, I'm advising everyone I know to lean that direction. Not trying to become hostage to something like claude's ecosystem for the rest of my development career.

feel like its beyond optimistic on their part, just starting to hear their name be blended with companies desires on job listings, and they are destroying the goodwill of the devs who surely are the main reason their name has landed there. They aren't dug in like a microsoft, maybe they get some staying power for nocode people who feel trapped, but im done with their nonsense already and won't recommend them anywhere. Other stuff is good enough already to match.

they need the devs on board for that to matter, i can get whatever i want done with lesser models already. It is quite literally about just who is not gonna give me the shittiest experience, and at anthropic it sure seems they are determined to annoy everyone since they started gaining in popularity.

kinda surprised to see this type of take out of someone who participates on this website. I feel like this is the place where I have seen that middle ground surface the most. Just the overall shift in the past year from semi handwaving to feeling like it must be embraced, and identifying the problems it creates and how to address them. I feel this is all exactly what you are mentioning.

I think AI as a proper utilized tool, is amazing, I think our lack of restraint when just throwing it into everyone's hands without understanding of the tools they are using, is horrifying. I'd imagine a lot of the community here echos that same sentiment, but maybe not, and i am just making assumptions.

The overall sentiment on here might be in the middle, but I feel like that is more because half the posts and comments are railing against AI slop and half are about exciting new AI models or tools.

that usage page is getting really cluttered, really fast. Two months ago, mine was simply a 5 hour usage window, and the extra usage portion.

Now its 5 Hour usage, Weekly Usage, Claude Design, Daily Routine Runs, and the Extra Usage portion...

I hesitated 100% when i saw caveman gaining steam, changing something like this absolutely changes the behaviour of the models responses, simply including like a "lmao" or something casual in any reply will change the tone entirely into a more relaxed style like ya whatever type mode.

I think a lot of people echo my same criticism, I would assume that the major LLM providers are the actual winners of that repo getting popular as well, for the same reason you stated.

> you will barely save even 1% with such a tool

For the end user, this doesnt make a huge impact, in fact it potentially hurts if it means that you are getting less serious replies from the model itself. However as with any minor change across a ton of users, this is significant savings for the providers.

I still think just keeping the model capable of easily finding what it needs without having to comb through a lot of files for no reason, is the best current method to save tokens. it takes some upfront tokens potentially if you are delegating that work to the agent to keep those navigation files up to date, but it pays dividends when future sessions your context window is smaller and only the proper portions of the project need to be loaded into that window.