Post

Skill issue.

Dan Tappan@dantappan.net

Cory Doctorow describes what has been my thesis about “AI” coding: it’s great for small productivity projects, it’s a source of crippling tech debt for production code and libraries.

Accurate that it's a skill issue. Why is being a bit flippant about it but it's accurate. Good engineers know how to get good work out of llms; bad ones produce tech debt. Same as ever

lol I know you're joking but it's been weird how often people seem to think talent and experience aren't still a factor Like, in both directions: 1, rookies don't need to know what you're doing anymore, and 2, experienced engineers aren't capable of evaluating whether AI coding is good or not

Obviously item 2 gets me pretty frustrated. I'm surrounded by the most talented engineers I've ever met. Seems remarkable to me that they all agree ai coding is effective, and that theyre producing incredible software with it. Seems like we should believe the experts on this

Hmm. While I agree that talented software engineers are very good at using imperfect tools, it's also true that software engineers tend to overestimate their productivity, the quality of the code they generate, and the impact of tools. (And I say that as a long-time software tool developer!)

So ... I dunno. If your team has been using AI more extensively over the last year, at least from the outside it doesn't seem like it's translated to significant additional functionality, improved quality, or delivering improvements that would lead to user growth.

An honest accounting of this would be, it started about 6 months ago, has delivered huge immediate benefits to the internal systems management, and the outward facing software projects are grander in scope than before and require commiserate rigor - meaning they still take time

0 reposts9 likes

A significant portion of the speed gains have been eaten by increasing the scope of what we try to ship and increasing the correctness work before releasing Jetstream v2 will be the first directly visible instance of it

Fair enough. Like I say, this is just my impression from the outside. In general (ignoring the ethical, cognitive, power-centralizing, etc issues) I do believe that skilled devs can get a lot of value from these tools ... but it's still very much an open question what the impact in practice will be.

also @pfrazee.com Cory's thread does distinguish between your situation -- skilled and *empowered* team -- and the situation where it's leading to huge amounts of tech debt in his discussion of why some devs really do see significant benefits from LLMs. bsky.app/profile/doct...

But as soon as I realized that the former group were "centaurs" (workers who get to decide and direct their adoption of automation) and the latter were reverse centaurs (workers who were conscripted to serve as peripherals for automation systems), it all snapped into place. 6/

Yeah there's definitely (I assume) a big difference between "smart team with smart leadership in a small startup where everyone has a lot of independence" vs. "corpo where managers are standing with a whip above each employee ready to beat them if they don't hit the KPI of tokens used"…