Video wird geladen...
Video konnte nicht geladen werden
"Software Engineering Will Be Automatable in 12 Months," Anthropic CEO Dario Amodei predicts that AI models will be able to do 'most, maybe all' of what software engineers do end-to-end within 6 to 12 months, shifting engineers to editors.
4,485,871 Aufrufe • vor 8 Monaten •via X (Twitter)
36 Kommentare

Didn’t he say the same thing 12 months ago?

Absolute bullshit. Spend an hour with enterprise software tied to three different legacy systems and get back to me.

Plan B

its not going to be any different in 12 mo . The core ai wont be . the harnesses will improve and the plugins

He said same thing 12 months ago 🤣

the bubble burst is going to be incredible

pop

Snake oil salesman says what

Everyone's focused on "12 months." Nobody's asking: what happens to the 10 million solo devs building weekend projects RIGHT NOW? AI won't replace engineers. It'll create 100x more builder-operators who ship without writing a single line. The question isn't "will my job disappear?" It's "am I building my own leverage before everyone else does?"

Shifting developers to (product) engineers

Keyboards (2026 edition)

AI has helped with coding, no doubt. But it seems like we heard this stuff a year ago, yet us developers are still here developing and Claude Code, in all of its glory still have problems with complex or enterprise tasks.

He says this every 3-6 months. We are far from it.

a new role will take its place : software engineering engineering

here's the breakdown of that clip plus the rest of the Davos video:

My Prediction: I don't buy it. He is wrong. Software engineers will always be needed. AI is just a model, and that's all it will ever be. No model makes perfect predictions, thus AI will not be able to produce all code.

Learn to coal.

the job description becomes "senior prompt reviewer with git access"

Engineers to Editors? Hopefully more Emacs style with clever LISP macros than Regedit ones.

He has not written a single line of code and he has no idea what he is talking about.

These guys are on drugs - Claude couldn’t even fix a fire base CORS issue, kept spewing out garbage

This video is old, from a year ago, and it didn't happen; there's just a lot of hype.

Interesting. Debugging prompts will be the new bottleneck then .

writing code -> editing code, yes but still need human there, so headline is misleading

They said that 12month ago. So far the only thing ai has successfully automated is CEO’s

i prefer this edition

I think they found the planate where 10 sec is equal to 10 years

Even faster if you're using Anthropic models in Kilo :)

Its the new compiler. You'll still need engineers to use them because you need people who understand what they do and what they produce to use them

Next year they will be saying the same thing 😂

What he said back in March, because everybody is misquoting him:

this is very helpful! and I agree with you, I think he's more spot on than people realize

I’ve been coding for 16 years. Eight months ago I doubted him; now ~98–99% of my code is written by Claude Code or Codex, and I have higher output and ambition than ever. I barely type code anymore. At a global level it wouldn’t surprise me if most code output already comes from LLMs. I won’t make the mistake of doubting him again.

Of course, that's nonsense! Writing code and even the overall design of software is merely the final stage of a fieldwork process that engineers must undertake by observing processes and engaging with users. They use context as information and experience to anticipate scenarios. AI confined to a data center lacks the capacity to do that fieldwork. It will simply be a better employee for coding and avoiding tedious tasks.

14th of March 2025... It's wonderful how these people get to declare whatever garbage they want, capture the media cycle and then shamelessly do the same next year without any consequence.

software engineering is much more than writing code. system design, tradeoffs, ownership, etc are all involved. an AI will amplify competence but it does not replace it. for example if you have a manager with no software engineering experience use AI to try and build a production grade distributed system, you’re cooked. the system will fail because as the manager passes garbage into the AI, he’ll get garbage out (and he probably won’t even know it). the tool is only as good as the person using it

