/dev/world 2025

What's AI good for? (Probably not that, but not nothing.)

As developers, we're pretty overwhelmed with AI-everything right now. You've probably heard all the arguments ranging from "all programmers will be out of a job by December" to "lying autocompletion engines have no practical uses." You've probably seen posts from people claiming incredible "one-shot" prompting that rewrites Safari from scratch in an afternoon. Maybe you've tried it out, and it wasn't quite...that. Are you doing it wrong? Is everyone lying? What's going on? I spend much of my work time developing software and processes based on LLMs. I've talked extensively to people in other teams and other companies about what does and doesn't work. And I've learned a few things that aren't obvious. AI is not particularly good at most of the things your boss probably wants it for. But there are problems it can help you solve that would be impractical without it. In this talk, I'll help you think a little more clearly about where AI will probably be unhelpful, where it may be harmful, and where it's worth considering.

About the presenter

Rob has been a Cocoa developer since 2003, and began developing for iPhone when the first public SDK came out. Before coming to Cocoa, he made his living sneaking into Chinese facilities in broad daylight, among other InfoSec risk assessment duties. He's given dozens of talks internationally on software development, infosec, and cryptography. Today he's a senior developer for Audible's iOS team.