I’ve had about 2 months off from active contracting. It’s been great. I’ve taken the time to play a lot of violin and bass guitar, acquired an AWS certification (thanks Anki!), and I’ve played around quite extensively with a few of the popular LLM coding workflows including copilot, avante.nvim, gp.nvim, claude, and codex. The notable item I haven’t used is Cursor. I’d like to continue using neovim and avoid VSCode if at all possible. No gripe with VSCode, I just love the terminal workflows.
By far, the Claude tool is supreme. Anthropic’s agent itself, Opus 4.5, appears to understand my code better than the other agents I’ve used. The workflow and tooling around the Claude CLI tool is also fantastic. I can’t speak highly enough of how much it’s increased my productivity. If there’s a software solution I’d like to build out and I have a strong knowledge of the domain-space, I can get so much done in just 30 minutes. And if I don’t have a strong knowledge of the domain space, I can leverage the tool to help me learn the codebase. Incredible, just incredible. There’s a lot of negatives, I know, but wow! It’s awesome.
I want to share a thought I haven’t seen expressed yet on the deluge of HackerNews posts about AI. As I stated above, I’m pretty into the terminal / neovim workflow of software development. One of the big sources of my interest in that side of computing is saving keystrokes and mouseclicks, my fingers, from doing more than they need to. A lot of scripting and vim usage is automation to save myself from more interaction with the computer. If these coding LLMs are looked at simply as autocomplete tools, this is a jump forward in my “saving keystrokes” workflow that makes all other moves before pale in comparison.
Now the autocomplete has to be reviewed by myself and it costs SAAS-service charges to generate. Nonetheless. Oh my gosh! This is 4 steps forward where all my previous work for my “cracked out vim workflow” was 1. There’s the important caveat that the “1 step” enables me to better understand computers and effectively use that “4 step” jump. Combining my terminal knowledge with LLMs, it is automation galore and I can get a lot done quickly.
And if these LLMs totally replace software development, I’ll fall back to the violin. I’ll play a song as the software-development ship is sinking, like LLMs are the iceberg. A good plan? More of a plan than a lot of us software developers have. Crazy times we live in!