@nthcdr @bbatsov well put.

I believe it's a slippery slope: Once you start using AI assistance for coding, you soon become lazy. After a while (half a year? Two years?) you're in practice dependent on AI to do any coding. You cannot code without it. In the beginning, you review each line, are careful you understand, really understand the code AI writes. As time goes on, you let it slip, let it fly. The vibing colleagues and AI agents throws so many PRs at you that you can't possibly review each line. Onwards.

At one point, why bother to review the code at all? Why not have the AI write a bunch of tests and trust that that's good enough? How about vibing some AI agents to scrutiny the code from different angles (security, performance++)? That surely fixes the problem of the annoying feeling that humans should review and understand this, surely?

Then, later, you might ask: why does the language, libraries and frameworks matter if humans are never to understand or touch the code? Why not just go for foolang, since that works the best with AI agents?

My day as a coder consists of 10-30% coding. The rest are meetings, meetings, meetings, emails, reports, code reviews and meeting preparations. I wish AI would help me with the 70-90% of the non-coding work, so that I can do the fun bit. Instead, we're investing SO much money to solve the 10-30% part, so that we can spend that time reviewing the code the bot generated.

As with all challenges at work, I believe the key is to do what Beatles suggested and "get by with a little help from my friends":, i.e. focus on the people you work with and ensure you are in a place where you enjoy the company you're keeping.

@amszmidt

Technically, two things wrong: (1) using an LLM, and (2) the thing that made you think you needed an LLM.