I’ve recently moved to Claude Code for my current one-man project as my primary AI code assistant and the productivity boost in terms of time to feature fully working is mind-blowing. A very poor metric to quantify this, but one I can do publicly, is via git commits, and by that metric I’m going twice as fast as with just Cursor, which I started using in late March and was already a significant productivity boost and felt like magic.
Category “Other” includes mostly .json schemas and .md files both for me and Claude Code.
On a “is this helping me solve real problems faster?,” you will have to take my word for it, but it is. UX features that in the past would take me a couple of hours or three, I can now get done in a minute or two. Code health refactoring tasks that you know would improve development speed but had a prohibitive opportunity cost, now become no brainers so I can improve code quality and future speed much faster.
The cost of writing and maintaining software is rapidly changing and I’m sure it’s going to really change the world. With my limited experience working with AI to write code, I’m already solve problems that a few years ago, working a big tech company, would take several engineers. And the rate at which these tools are improving is also incredible.
Claude Code isn’t just auto-completing names, for me it’s solving tasks in all these domains: data analysis, writing AI gents (analyze companies, resolve forecasts, merge datasets), typescript / react frontends, golang api / data pipelines, code health and refactorings, security review, devops, monitoring, performance debugging, and it’s one-shooting ~60-80% of the tasks with minimal tuning.
I’ve done experiments with broad, recurrent tasks like “look for code health / api issues and propose refactorings” and it does a good enough job to be worth the effort of writing the tasks and reviewing its output. And I have a simple setup to queue tasks 24/7 so that when I wake up I have work to review that Claude did overnight.
I still have to review code and do manual tests because Claude still hallucinates, and I need to balance my time to actually use the tool to solve the problems it solves, so the default $20 Claude Pro subscription is enough for now.
Reality looks like sci-fi.