You are not anthropic.

I have a theory that this current era of “AI will write all the code and engineers will vibe-ship straight to production” will feel like a fever dream a few years from now.

Everyone used to try and learn from Google, but as was commonly stated “You are not google”. And similarly, I think we can have a parallel “You are not Anthropic.” This idea of engineers guiding the future of their code by writing a million lines of code in a year. Like Anthropic is practically speaking still in their ZIRP era. Money to Anthropic is still free.

I see lots of articles that talk about close to 100% of code shipped with AI, and it makes me wonder - Why? Why ship a one line change with AI? I suppose this is similar to say - writing code in C might be more efficient than writing code in Python, but no point in trying to know both C and Python. But, someone pointed out in this article where a few engineers at Anthropic shipped a million lines of code in a year, this was ~600 lines of code per MR compared to the industry standard of like 5, and the bigger comparison - Firefox, with 10 years of MR’s is only 2.5 million lines of code. Did Anthropic build an internal tool in a year wixth 40% of the function/value/maturity of Firefox?

Why?

  • Don’t want to lose my coding skills
  • I like coding, this is my hobby
  • Continue to guide the architecture, keep things non verbose.
  • I think it’s good to have a single workflow - so by maintaing this workflow, you can regulate the flow of tokens and not get interrupted by running out of tokens.