GPTEngineer VS tzap.io

As you know, Stockholm was home to Stieg Larsson, has also great sailing culture, and serves so-so expressos. It also leads in patents per capita, investment per capita and I dare say GPT-based open-source projects (DM me if wrong). Not only it is the birth-place of "kthGPT" by Ludwig Kristoffersson - a youtube search engine for lectures that you skipped while not attending those interesting Kungliga Tekniska h?gskolan lectures, but it is also home to two more tools that you might be using very soon -- if you are a geek.

In this post, I compare two open-source gpt based projects, both conceived in my beloved ???? . These are tzap.io by the great Alexander Faraj and GPTEngineer Anton Osika and also great Emil Ahlb?ck among others.

For this comparison, we will ask both to make a Nokia snake game of the 80s by using Python. Therfore, we use the prompt "We are writing snake in python. MVC components split in separate files. Keyboard control"**

Results

Both produced a preliminary working snake controllable by keyboard.

• GPTEngineer did the job in 131 lines of code and 9 files
• tzap.io did the job in 82 lines of code, and 4 files

source code in the links-job.

Bugs & features

• In the code generated by tzap io... I found an indent error in one file; on the other hand,
• "my" gptEnigneer -- see how I quickly defaulted to genitive --- could not use gpt4 with the key I gave... so it defaulted to gpt 3.5.
• In tzap, I had to prompt twice... first it created a correct skeleton with lots of "add your implementation of this here..." ; then in the second pass she completed the missing implementations... I like this step wise method even though I had to reprompt with "complete missing implementations" the second time.

Remarks

What is your favorite prompting tool? Share in the comments!

** (please forgive the grammar ?? - GPT does not care so much about it.)