Rosanna CantavellaDec 7, 2022
Dave Dustin
Dec 7, 2022 at 7:31amWeb
Show threadMartinDec 7, 2022

missing alt text:
a venn diagram with two circles

the left circle is labeled "things that are actually slowing down our dev teams" and contains "constant reprioritizing", "burnout", "pointless meetings", and "technical debt"

the right circle is labeled "things an ai coding assistant helps with" and contains "googling boilerplate code"

the intersection of the circles contains one item: "generating subtle, devastating bugs"

the lower right hand corner says "@forrestbrazeal" (i believe) and probably indicates the creator by their handle on social media

Show threadalcinnzDec 7, 2022

@venzann I worry AI assistents would only generate more "subtle, devastating bugs" for me to catch. Worsening that bottleneck!

Incidentally when I last tooted about this, some replies didn't seem to get I was referring to regarding "bottlenecks". That leaves me concerned...

Show threaddr 🛠️🛰️📡🎧Dec 7, 2022
@venzann Even more to the point, you shouldn't be googling boilerplate code either. If your language, architecture or design has that much boilerplate **fix it**.
Show threadJakub NerudaDec 7, 2022

@davidr @venzann I think this isn't issue of quantity, rather than muscle memory. Many folks would struggle implementing double dispatch for Visitor pattern from memory and they would google such boilerplate.

Even more folks would probably not expect something as the Visitor pattern to be provided to them by the standard library.

Show threadAbhi Beckert 🇺🇦Dec 9, 2022
@davidr @venzann Nah there's a place for boilerplate code. I often see people add a dependency with a thousand files when 10 lines of boilerplate would've been better (less bugs, less code to audit, etc). It's not something we need AI to fix though.
Show threadJake MannixDec 7, 2022
@venzann lol!
Show threadkechpajaDec 7, 2022
@venzann If they want my attention, they need to build an AI that generates documentation for existing code. Now _that_ I would probably use.
Show threadIgigogDec 7, 2022
@kechpaja I have actually seen it done with Codex. A dude generated a documentation for the whole lua codebase (90k loc) of a game. Don't know how, just saying it's a thing already.
Show threadJakub NerudaDec 7, 2022
@kechpaja @venzann It actually doesn't sound that hard to train ChatGTP to do exactly that. With decent enough function and parameter naming and exception annotations, it could probably do it right now!
Show threadBrennan StehlingDec 7, 2022
@venzann It's a great perspective. The focus has been too limited. Triaging tickets could actually be a good way to cut down on meetings and there is a long backlog of past tickets to prepare the ML model. Debugging is another one where it is possible to do the analysis to identify the most likely causes for a bug along with recommendations and even code changes. An automated PR that the team could review would be amazing, especially if the PR included a useful description.
Show threadBob ThomsonDec 7, 2022
@venzann Compromise - AIs have all the meetings so engineers can get peace to deliver what they decide in the meetings :)
Show threadRev. Bussy Body of The Church of America MLM and Shooting ClubDec 8, 2022
@venzann in before(?) they start putting these algorithms to use facilitating management and business decisions and a much worse "constant reprioritizing" is in the overlap
Show threadAsbjørn UlsbergDec 8, 2022

@venzann @yellowbrickc A Venn diagram of two partly overlapping circles.

Left circle labelled “Things that are actually slowing down our dev teams”: Reprioritizing, burnout, pointless meetings, technical debt.

Right circle labelled “Things an AI coding assistant helps with”: Googling boilerplate code.

Intersection: Generating subtle, devastating bugs.

@alt_text

Show threadFlorian EgermannDec 8, 2022
@venzann to be fair, automating "google boilerplate code" helps me tremendously.