What's Next?
As I begin the search for what’s next, work-wise, I’m forced to accept that it will likely involve using LLMs. Given all the shared ethical concerns, none of us have been using LLMs at work, but I’ve played a little with Claude in my spare time.
What I have noticed, is that it replaces the part of programming I wasn’t very good at - getting started from a blank sheet of paper… overcoming analysis paralysis, trying to get it right first time.
What it doesn’t replace, I have found, is the part I have traditionally been very good at - refactoring and design. Taking a complex, unstructured codebase and cleaning it up.
What the LLM produces “works”, but if I want to understand that code and be responsible for shipping it, there’s still (for me) a vital manual step in reading the code and spotting opportunities for clarity, or reuse, in finding places where there’s a missing abstraction, and two things are really the same if you look at them the right way….even if those insights just end up as more input to the model.
I recall Kent Beck saying (I’m paraphrasing) that LLMs made 90% of his skills irrelevant but 10% became vital. I don’t think that was true for him, and I don’t think it’s true for me.
Maybe 10% of my value to an employer (as a programmer…I do a bunch of other stuff too) became irrelevant. I think I know exactly which 10% and it was 10% I wasn’t great at anyway.