Top engineers at Anthropic, OpenAI say AI now writes 100% of their code

[tayo42]

I feel like collectively we should just stop talking about this and speculating and just see what happens in a year.

Is programming obsolete, or is it needed more then ever becasue we need to fix all the crap generated. Let's just all follow up at the end of the year.

Itll be quiet though if we're still hand writing code though I think

[atomicnumber3]

Hasn't AI been annoying the piss out of us for well over a year now? I've definitely been hearing "10x productivity" for that long.

So - where is it all? Where's all the 10 years of software dev that happened over the last year? Where's the companies blowing their competitors out of the water by compressing a decade of production into a year?

The proof is in the pudding. Or lack thereof. If the claims were anything like they say, we'd see something by now.

[readme]

I can't really trust this kind of statement since there is an obvious conflict of interest.

You can write lots of code with an LLM.

"I did ~20 PRs yesterday all coded by AI"

I guess anthropic makes you use a pull request for minor code changes.

[mifreewil]

Real estate agents say now is a good time to buy!

[hehrhrbrh]

Yeah, we’ve noticed.

Anthropic over billed me by $300 last month.

[EddieRingle]

I didn't get as far as experiencing their billing system; I decided to finally look into Claude recently and discovered that their pricing page's content area was completely blank. It loads on my phone if I'm off WiFi, so I assume my PiHole is blocking something there. That being said, if anyone at Anthropic is reading this, I'm primarily an Android dev but if you'd like to hire me to write a simple static HTML page just like I taught myself back when I was in elementary school, I think it could significantly improve your conversion rates or whatever, and I've been unable to even get an interview anywhere for the past 7 months so I really think we'd be helping each other out.

(Suffice to say I passed on Claude/Claude Code for the time being.)

[illist-ell1s]

Every time I ask Opus 4.5 to generate code it is a mess, by mess I mean a human would not have done it that way, it is sub-optimal for maintainability and improvements. The more complex the worse. But generally at least the code works. If I need to modify it, what a nightmare, better tell it to modify so AI slop leads to more AI slop.

It is the same as writing a story. It ain't gonna win The Hugo Award. We know it is AI, full of cliches and corporate speak. I think the difference is people are used to crap code, where as they won't read a crap book.

Most code is crap because 1) engineers don't care about the software 2) high turn over 3) getting rushed

So maybe AI just generates code at the level most people are used to, I have no idea because compared to my code Claude code is bad, but bad is not useless. I put the bad code in its little corner and it if it does its job it is fine.

[KellyCriterion]

you have to give it clear statements what you are looking for, how the solution should work and what you are expecting.

also insert into prompts (or in the base prompt) someting like (e.g. example from my Claude setup) >>

- Configuration as data, logic as loops

- Dynamic rendering (adapts to data size/content)

- Single source of truth (no scattered magic numbers)

- Defensive coding (null checks, division by zero)

- O(1) lookups over O(n) searches where possible

[whattheheckheck]

What's your github profile? I really want to admire the best of the best

[Grimblewald]

LLMs, in my experience, can solve solved problems for you well. It is extremely rare they produce quality code for novel things.

therefore what I hear from an announcment like this is "we have stopped innovating".

this is a bad sign, not a good sign.

[warrofua]

Curious opinion.

Is Claude Cowork not innovation? https://www.reddit.com/r/accelerate/comments/1qbg9bu/anthrop... https://jpcaparas.medium.com/the-claude-code-creator-says-ai...

is this not innovation? >> https://openai.com/prism/

A better argument might be that greenfield applications are currently way easier for AI to build out, I do concede that point.

[browningstreet]

I believe this to the same degree that I believe Soderbergh films his movies on iPhones.

[infomaniac]

It shows.

They use all those spare brain tokens testing the product behaves correctly, right?

Right?!

[eclipticplane]

You're absolutely right!

[almosthere]

I mean to be honest, we're a company of 12 devs, I think 100% of our code is agentic now. We almost treat it as a liability to hand write something.

Apparently cursor keeps track and Opus4.5 is the model our top 3 devs use.

[warrofua]

Yep, can confirm, been doing it this way since co-founding a healthtech startup in 2024 (and for personal projects even before that).

It is like a 180 degree different mindset than what typical SWE requires.

[thebenedict]

I'd love to hear more about the workflow. Do you open PRs and is there human code review? What scale is your software used at?

[scrubs]

Ok, I'll bite: 12 devs in what domain of work + 1 paragraph description of system's roles and responsibilities, please.

[nomel]

What domain, if I may ask? I assume web?

[abacadaba]

they just like me

[villgax]

I mean CRUD apps aren’t really that difficult of a coding gig to begin with.

[tester756]

they're lying