I did something similar to that at one of my jobs. We had a Java web app that we wanted to port to Python. I created a new Python web app and ran it alongside the Java app. Then I started reimplementing the Java code in Python, one endpoint at a time. Every time I had a new Python endpoint working, I went into the corresponding Java endpoint, and I replaced its code with a call to the Python app. The Java app, in effect, was becoming a proxy to the Python app.

The plan was that once all the code had been ported over, the Java app could just go away. I skipped over some of the details here, but this was a nice way to port the app over without feeling in a rush, and without having that big stressful "throw the giant switch" moment. Ultimately, the company decided to scrap the original app altogether and rewrite from scratch because a lot of the original app was irrelevant to the new direction we were taking... but I got about 20% of the original app ported over in a week or so.

We tried that in the company I worked, project was abandoned 1 year later. Improvements and better practices were incorportated into the old codebase which is still scaling today.

I worked somewhere where this is all we did. Changing things was discouraged (versioned api endpoints) unless it was to fix a bug. I actually loved it. Each time you copy-pasted an endpoint to make a significant change, you didn’t have to. You could just rewrite it from scratch, taking everything you knew from the old versions.

One of my colleagues recently described[0] how we applied this at Deliveroo and it worked nicely for incrementally decomposing

[0]: https://youtube.com/watch?v=LP0sTNruvaM