GradientDescent

@bob_zim @pier @martin @jk I totally agree with both of you. Without going into too much detail, I tried my own "throw everything and the kitchen sink" at codex and the results were not impressive at all. I've taken the stance with my team that they should curtail any LLM usage to small autocomplete suggestions for single code lines that they could have written themselves. For any real coding, there isn't any substitution for personal knowledge of the problem space and solution.

@martin @jk As long as "tools" remain "tools" and not replacements for actual thought, there is little harm in including it in a development workflow. LLMs can, and will, churn out crap on occasion. Safely constraining usage of any AI tool is paramount for the foreseeable future. Done properly, the impact of LLMs in a dev workflow is equivalent to syntax highlighting: nice to have and helps get the work done more quickly, but doesn't change the bottom line necessitating writing good code.

@martin @jk totally agreed with martin. It all comes down to implementation. Self-respecting orgs can, and do, use AI tooling to assist and accelerate programmers but maintain that their programmers are responsible for disclosure of AI tool use while keeping the programmers themselves in the driver seat. Hell, even the Apache Software Foundation has taken this approach and I can sparsely think of anything higher profile than the projects they maintain.