sirnicolaz

vibe-coders should be pretty happy about the fact that places like this exist, given that's it's ultimately thanks to non-vibe-coded OSS that LLM can improve.

I'll join the rebels

This does still not include any QA. The fact that A.I. tells you the implications of your changes does not take away the effort and time you need to verify the correctness end to end

One week to both refactor a big code base and also make sure everything works as expected? Not the estimate I would expect from a senior analyst, too optimistic. Fine that you fix the regressions for free, but the last thing I want is a one-week refractory going to production and fire fighting later.

Ironic that this very article has been partially written with AI... kind of lost the drive to read it

Now, let's imagine your work all day with an LLM...

I spent 10 minutes figuring out how I could just follow a few instances without having to run a docker container on my machine or keeping N tabs open all the time.

Not a chance to figure that out (I am no mastodon user, so hard for me to guess).

These platforms are promising, but they need to remember that the average user will turn away immediately if there is no clear guidance on how to do the basic things.

never

Oddly comforting

https://archive.ph/z24oE

I remember when working for one of my old clients last year that the business as usual slack conversation was:

- lead AI engineer asking tech advice - me proposing something (out of my experience and knowledge) - he invoking inline slack bot to ask if what I said made any sense - me telling that the AI arguments were kind of off - he invoking again the same both asking why I said that - .... so on... - third developer kicking in responding with obviously ChatGPT generated message

I left the company shortly after.

I hope for his mental sanity that he didn't

"people should aim to reengineer humanity and eliminate illness, old age, suffering and vulnerability"

sure, that's exactly where AI is headed.

also: just imagine a world where people don't age. just imagine it for a second.

AI usage

I used ai for a single rust scope issue that google wasn’t giving me clear answers for.

Nowadays, this needs to be highlighted. Congrats

I had a professor at my CS university (one of the greatest I had) who used to say (in 2008): "a developer should write no more than 5 lines of code a day"

Because the SWE benchmarks for LLM coding are done on python code bases, hence you are likely gonna have superior results

Thanks for the breakdown! I myself tried to do some poor man calculation, and using this website https://tokenomy.ai/tools/energy-usage-estimator I have came to the conclusion that 1000 developers using 100k$ worth of tokens in a month (20B) would equal the total estimated energy used to build the Great Pyramid

Ps: yes apparently at Meta there are devs consuming a yearly salary of tokens in a month, according to a recent NYT article https://www.nytimes.com/2026/03/20/technology/tokenmaxxing-a...

Hi! Also in my 40s, 20+ years of experience. I took the coding career because I like solving problems with code. I use agents for one-shot tasks, quickly understand big code bases, finding what I need in messy documentation, have a test suite foundation (things which I always didn't like much). I will continue this way for now because, honestly, if my job has to become reviewing tons of code vomited by an AI the whole day, I'd rather become a bar tender. I won't jump the panic train now just because everyone is obsessed with coding agents. Wait a couple of years, see what will actually remain out of this bubble and then see if and how to pivot, now it's not the time. Indeed if you have interests and budget, do go back to school and reinvent yourself, but that should be a decision independent from the current AI hype.

Consider that SWE benchmarking is mainly done with python code. It tells something

It's the very reason why I don't get this brutal attempt to replace developers with AI. I would have considered a more reasonable approach to use it for things were "chatting" is the main skill.

https://www.anthropic.com/news/where-stand-department-war

This made my day, thank you

One thing is playing Pokemon, one thing is decide who to kill. Also: if they are planning to use it on the field, there is going to be a velocity issue. Claude and any other LLM require a non negligible amount of time to ingest the input and spit the output.

"fully autonomous weapons and mass domestic surveillance"

I still don't buy this discussion. How exactly do they want to use an llm for autonomous weapons, given it's not even possible to reliably have a piece of code written without having to review it?

And how is a 1M token window model suppposed to be useful for mass surveillance?

Honest questions, I am sure I am missing some details. Because so far it looks like a very sophisticated marketing strategy.

This is like comparing the danger of a machine gun to that of a block of lead.

This should get +1000 upvotes, a gold mine of useful information . Thanks!

This is TopTal. I had the same issue. I am gonna just close the account, they basically nuked all my portfolio (all react apps that have less than 1kb initial payload, indeed too small to be proper).

Exactly the same thing

I still miss the times when I would decide to watch a movie based on the cover at the VHS store or based on a recommendation. Much faster, way more serendipity.

A bit of incoherence between this

'We are destroying software telling new programmers: “Don’t reinvent the wheel!”. But, reinventing the wheel is how you learn how things work, and is the first step to make new, different wheels.'

and this

'We are destroying software pushing for rewrites of things that work.'

But generally speaking I grok it.