In my ongoing journey with Productlogz, I’ve started tracking how much of our code is written by AI tools like Cursor.ai and Claude. Cursor’s new multi-file edit feature is a game-changer, making it feel like I’m working with 1-2 interns. Here are my observations over a few days of tracking AI’s contribution.

Day 1: Mixed Results

On the first day of tracking, I had mixed results:

• Creating a new component from an image: AI did 90% of the work. 🤯
• Refactoring existing code: AI contributed only 5%. 🤷🏻‍♂️
• Modifying existing UI code to match a payload: AI handled 50%.

Day 2: Steady Contribution

By the second day, AI’s contribution stood at 50-60% for a mix of new code and modifying existing logic. It was impressive to see AI’s consistent performance across different tasks.

Day 3: Bug Fixing Process

On the third day, something remarkable happened. I was about to write a crucial bug fix, but AI beat me to it and wrote the fix itself. 🤯 The AI’s contribution for the day was 100%, a real game-changer.

Day 4: Self-Correcting Code

On the fourth day, AI continued to impress:

• It created new state variables by referencing existing conditions.
• It formulated new conditions based on changes in other files. 🤯
• It even corrected buggy statements on its own. 🤯 Overall, AI’s contribution was again 100% for the day.

Day 5: Consistency and Speed

On the fifth day, I noticed that AI contributed 40-50% when writing code from scratch. When it came to writing repetitive conditions, AI’s contribution increased to around 70%. The speed and efficiency of these rewrites were astounding, taking only seconds. 🤯

Conclusion

Tracking AI’s contribution has been eye-opening. From creating components almost entirely on its own to fixing bugs before I even start, AI is becoming an invaluable part of my development process.

The way you can work through a UI code problem with Claude is extraordinary, and I find Claude better for coding than ChatGPT. Once you understand the time efficiencies LLMs provide in the work you do, there’s no going back. It’s clear that learning how to code and prompt is the future of software development. And it’s only going to get faster from here.