[zh3]

As a 'professional coder', I've been very much pondering - and experimentinng - with how best to use AI coding tools.

For big, complex, programs I work on (multiple C programs up to 100k LoC) it's hard to frame a question in such a way that it doesn't need to know details of the codebase. For peripheral stuff it's pretty useful as a quick way to get some cut'n'paste code ('write me code to read the load average on linux and log it to a file every second' - I know exactly how to do that, but it's a faster typer).

So initial take is it's a very useful new tool - so sort-of a corollary, people who learn how to use it quickly and well are going to have an advantage in the short term. And even if/when everyone uses it, some people are going to be better at using it (i.e. asking the right question - like I find myself groping around for, when asking it about UK company law earlier today).

[908B64B197]

> For peripheral stuff it's pretty useful as a quick way to get some cut'n'paste code ('write me code to read the load average on linux and log it to a file every second' - I know exactly how to do that, but it's a faster typer).

For public APIs that might be the killer feature; a better way to look at doc (or rather, rely less on doc) if it can look at different code bases and find snippets of code using a popular API.

[mike_hearn]

It definitely works well for that stuff, albeit it tends to be easily confused by APIs that have evolved over time. I find it frequently giving me code that uses APIs that don't exist, and when I point that out, it apologises, tells me that API was removed in version xyz and then gives working code instead.

[zh3]

Yes, could well be you've hit it there (i.e. stackoverflow has had its day).