Last night I wanted to build a small but useful feature on top of an existing codebase. I knew what I wanted, roughly where to change things.
A) Use an AI agent and spend an hour promoting and it will still get it wrong
B) Just code it in 5-10 minutes and get it right first
A) Use an AI agent and spend an hour promoting and it will still get it wrong
B) Just code it in 5-10 minutes and get it right first
Comments
https://www.microsoft.com/en-us/research/publication/the-impact-of-generative-ai-on-critical-thinking-self-reported-reductions-in-cognitive-effort-and-confidence-effects-from-a-survey-of-knowledge-workers/
That would be the highest value added feature.
All I see is trivial programming examples an. intern can do with the docs.
Hard to see this gap closing when working in complex codebases where it’s hard to even explain the work
But it was doable because I was the one who wrote the code in the first place. I
I would write a single test, then write the name of the second test and it gets the second right 90% of the time if they are similar.
Other than that, it's mostly in my way
I am noticing that either:
1. I get in “the flow” and can stay more thanks to being able to turn on integrated AI where I can add a comment and see suggestions immediately (not needing to leave IDE)
So it’s a net productivity gain for me! A big one.