The REPL: Issue 133 - September 2025
Yet another LLM rant
Dennis Schubert discussed LLMs – like half the internet.
LLMs can be a useful tool, maybe. But don’t anthropomorphize them. They don’t know anything, they don’t think, they don’t learn, they don’t deduct.
This is true. Although they can fake it quite a bit, especially agents that break down the next steps, read files, and come up with a list of todos, etc.
I find myself thinking that AI is both unreasonably hyped-up and incredibly useful. I’ve been able to get proof-of-concept projects working in minutes with only superficial knowledge of the underlying technologies. Denying that agentic AI is powerful seems foolish. Also, thinking that they can replace engineers wholesale is equally foolish.
We are in for a time of disruption.
Pick the wrong tool for the job
Sometimes the wrong tool for the job is right for you at a given moment in time. In this case, the author wanted to use Ruby to be able to iterate faster. Being able to learn what the users really want was more important than having the perfect technical solution
Find me on:
- Bluesky at @ylan.segal.family.com
- Mastodon at @ylansegal@mastodon.sdf.org
- By email at
ylan@{this top domain}