All of your bullet points look deceptive or dishonest to me. They are either untrue or something that won't stop the march of progress. When the car was invented was it a useful counter-argument to suggest 99.9999% of miles traveled so far were horses or bicycles? No.

> When the car was invented was it a useful counter-argument to suggest 99.9999% of miles traveled so far were horses or bicycles? No.

I feel it's possibly pertinent to point out that cars didn't use existing horses or bicycles as fuel/building materials/<some other analogy here>, whereas LLMs ingest the software people have written previously during their creation.

No, you're the dishonest one. The only bullet point a reasonable person might disagree with it "99.9999%". The rest is true.

- Many real devs criticize "AI". Documented: https://bsky.app/profile/robpike.io/post/3matwg6w3ic2s?ref=i...

- LLMs are trained without respecting the licenses of the original work, not even giving attribution which almost all OSS licenses require. AGPL required derivative works to be also AGPL - this should include the model and all its output for any reasonable meaning of derived work.

- SOTA models even today produce absolute garbage. A week ago, Claude Sonnet 4.6 tried to call one constructor from the body of another in C# using syntax which doesn't exist. Less glaring issues are completely normalized. This is why "agentic" generation is so popular today - it puts guard rails around the slop.

- I and other devs I've talked to are not interested in the mechanical writing of code but in the additional understanding which comes from engaging deeply with the problem and solution.