I'm spending 20-30% less time writing (better) code
A high-level overview of how I'm using AI to write better code at a faster pace.
I write over 95% of the lines I ship. The other 5% comes from AI (mostly Claude, but there's some GPT-4o and o1-preview in there). This is a rough estimate, but I write most of the code overall.
While 5% of code may not seem like a lot, remember that this code is usually clutch! These are things I don't know how to do or would take me a long time to figure out.
So, I estimate AI saves me 20% to 30% of the time it would take me to do things cold turkey.
This is good, but there's an added benefit of using AI that gives me a ton of value: These models are awesome at helping me simplify my code!
I first write the code, and when everything works, I ask the model to simplify it:
"Simplify the code below so a developer with experience writing Python can easily understand it. Remove anything unnecessary. Fewer lines are better but don't sacrifice readability. Prioritize Pythonic constructs wherever possible."
Half the time, the model either writes a better version or gives me ideas on how to improve things myself.
As a side note, simplifying code becomes much more dynamic and secure if you have unit tests. Be extra careful if you are doing this without a safety net.
I'm not sure how to quantify the value I get from shipping code with a few extra degrees of readability and simplicity, but I'm sure it's a lot.
Hard to go back to pre-AI development.
I experience something similar. Currently, I work for Volvo Cars. It took them over a year to decide to get Github Copilot for the whole company. A year after, the benefits of adopting Copilot are undeniable on a company level. AI won't replace you, someone using AI will.
What's your current editor/IDE setup? I'm using PyCharm + Copilot but was wondering about Cursor+Claude or other alternatives