It's extremely hard to measure productivity correctly and self-reports are worthless. I don't think AI tools are a net negative in the average case (people are definitely indexing too much on that goddamn METR article) but "i'm 10x more productive, source trust me bro" is equally nonsense.
Using AI tooling means, at least in part, betting on the future.
> Using AI tooling means, at least in part, betting on the future.
It means betting on a particular LLM centric vision of the future.
I’m still agnostic on that. I think LLMs allow for the creation of a lot of one off scripts and things for people that wouldn’t otherwise be coding, but I have yet to be convinced that more AI usage in a sufficiently senior software development team is more valuable than the traditional way of doing things.
I think there’s a fundamental necessity for a human to articulate what a given piece of software should do with a high level of specificity that can’t ever be avoided. The best you can do is piggy back off of higher level language and abstractions that guess what the specifics should be, but I don’t think it’s realistic to think all combinations of all business logic and ui can be boiled down to common patterns that an LLM could infer. And even if that were true, people get bored/like novelty enough that they’ll always want new human created stuff to shove into the training set.
It works by translating language abstractions to code.
The probability of plain english being correctly translated to code depends on existing code and documented abstractions describing lower level functionality.
Yet HN has what feels like one blog post or article to that effect every day. Some senior person describing their workflow, arguing that this makes them 10x of before.
It's refreshing to get the other perspective now and then.
Using AI tooling means, at least in part, betting on the future.