This week, I converted the Merchants' Village software from Rebol to Flask, using Fable.
It only took a few prompts. I did it using Deb's $20 per month Claude account, so had to wait for rate limits to pass every few hours, which spread the conversion across a couple casual days.
I did a few revisions after testing Fable's, but for the most part, that model just understood all the details of the Rebol code, and handled everything expertly. I even converted a big revision of the Merchants' Village software, which I created for a client around 2015. It took Fable only a single additional pass to complete that version too, which in the past encompassed many weeks of hard work.
The one thing I notice now is how much dramatically simpler the entire Merchants' Village project was, than project requirements I complete every day with GPT and other LLMs. It's not at all unusual to complete more than 100x the work, for production deployment in mission critical environments, using GPT, in a single day, than I could have completed with Rebol in weeks - I'm quite literally hundreds to thousands of times more productive these days, using GPT and other models to write code, and I experience none of the stress, fatigue, or pain that was involved in extremely challenging projects in the past. That 100-1000x increase in productivity is basically effortless. It's more challenging for me to get solid requirements from clients these days - but even a significant portion of that communication and time-consuming work is handled largely by LLMs in my daily workflows.
What an eye-opening comparison.