This is a special post for quick takes by Viliam. Only they can create top-level comments. Comments here also appear on the Quick Takes page and All Posts page.
Today I tried to reverse engineer an old DOS program with the help of Claude.
First, Claude wrote me a Python code that disassembles the program and writes the output in a text file. Then I gave the output to Claude chat, and... Claude can just look at 1000 lines of assembly code and immediately know what happens where. Got some small details wrong, so I didn't feel completely useless when I helped to clarify them, but the big picture was right.
I did in a few hours of pleasant interaction what would otherwise be a frustrating week (and I would probably give up). It's not just about the speed, but also that the whole process feels less frustrating. You know that if you find some piece of puzzle, Claude will help you connect it to other pieces, so even the parts that you need to do yourself get a big multiplier on outcome.
I think there are some things that Claude is really good at, and others which it can't do at all. Some people mostly do work in the former, and get big productivity gains, and others don't. (see also)
I guess one difference is what tasks you want to solve... mine are probably sufficiently unambitious.
I suspect another difference is that Claude cannot do some tasks 100% alone, but can be insanely productive if you provide the 1% it cannot do. I have one or two moments during a project where I need to start debugging. It is still much faster than doing all of that alone.