Since ChatGPT came out, I’ve mostly felt smarter than the LLM or at least on the same level. At the Swift / iOS level, it’s been bad. I check every commit and find mistakes every single time. So of course I’m smarter in the domain of my expertise. But that is slowly changing. With the proper skills and now SwiftFairy code checks at every commit, it’s becoming A LOT better!
But even with things I’m not an expert in, like writing Python scripts and Typescript functions, I feel that if I took some effort to learn these languages, I could definitely write the same thing that the AI writes. It’s pretty basic stuff. And despite not knowing the language directly, I’ve still had to steer it to implement certain details in the scripts, such as making sure it can restart from the point it left off at if it is stopped mid-way for any reason. So I haven’t found it to be perfect in Python either.
But then something changed… Last week, when I was building an agent workflow for a complicated RAG + Jev implementation, the agent ran many variations of the flow and landed on one that had the best results from a golden example set of prompts it was testing. Instead of just going with its suggestion, I decided to look at the Typescript code. Well, I did NOT like the code. It was not implemented the way I thought of it. It even hard-coded some exceptions, which I immediately told the agent to take out. I also didn’t like its decision to run two tracks in parallel.
Well, fast forward today, and as I was modifying some database backend data and testing out the search flow, I was not happy with it. So I kept working on it… and lo and behold! I came to the same conclusion as the agent had initially suggested. Hard-coding certain exceptions and running two tracks in parallel… It now works perfectly…
Anyway, as I keep working with Fable and now the incredible Opus 5.5, I’m sure this sort of thing will happen more often. And I find this to be a good thing… There is a saying that A players hire A+ players, while B players hire C players. I WANT to work with an intelligence that is smarter than me. Why not?!! I want to build more advanced things, so I’d rather have it be smarter than me than dumber. It hasn’t been fun to steer a dumber Swift agent. It’s also a weird feeling and hurts the ego for sure. YEARS of work and learning to build expertise only for AI to come and know more than you. Pretty wild!