Comments

• 

  2 days ago by amunoz | Reply

  This was my first time pair programming with Gemini AI (or with an AI agent in general). My HW25 project was more an excuse to work with such coding mate and evaluate the experience. I chose to develop a flask application implementing an API and a simple frontend, as I thought there might be plenty of Flask applications used for AI training. My reasoning was that Gemini would be good in that area of expertise. Additionally the project would allow me to experiment a bit in frontend technologies where I am a total noob.

  Here are my thoughts about developing with an AI coding assistant: - Overal I think this kind of development flow is: - better for implementing PoCs or simple custom tools that were not worth the effort before. - bad in general for software quality. This will happen because the humans supervising the code generated by the AI will become lazy and trust the code without a proper check.

  • In my experience, Gemini works better when you assing it to implement small tasks. Examples where it shines are implementing unit tests or refactoring some functions.

  • About my setup, I used Gemini (NOT as an agent) with the VsCodium extension. Works well but I found some issues:

    • If you make code changes without any AI interaction, the version the AI has of the files is NOT the same you have locally and you need to create a new chat from scratch to make them in sync again.
    • I only was able to accept the diff of each file individually. The accept all was not working for me.

  • You can't trust blindly the AI and a detailed review is required for each task. The AI can implement 10 unit test perfectly in a row but for some reason in the next request it will implement the test in a file in some other directory or will use a different name in the fixtures it was using before.

  • With small tasks, you don't need to be super specific with your prompts, Gemini will probably understand what you are trying to get (at least with simple projects as a Flask API.)

×
Edit Comment 12237
# A First Level Header
## A Second Level Header

Use one asterisk to *emphasize*

Use two asterisks for **strong emphasis**

- Use hyphens
- for unordereed
- lists

This is an [link to example.com](http://example.com/)

This is an image ![an openSUSE geeko icon](https://en.opensuse.org/images/d/d0/Icon-distribution.png)

This is a user link @hans

This is a project link hw#some-cool-title
More Complex Markdown Help
Formatting Help
• Edit
• Preview
This was my first time `pair` programming with Gemini AI (or with an AI agent in general). My HW25 project was more an excuse to work with such coding mate and evaluate the experience. I chose to develop a flask application implementing an API and a simple frontend, as I thought there might be plenty of Flask applications used for AI training. My reasoning was that Gemini would be good in that area of expertise. Additionally the project would allow me to experiment a bit in frontend technologies where I am a total noob. Here are my thoughts about developing with an AI coding assistant: - Overal I think this kind of development flow is: - better for implementing PoCs or simple custom tools that were not worth the effort before. - bad in general for software quality. This will happen because the humans supervising the code generated by the AI will become lazy and trust the code without a proper check. - In my experience, Gemini works better when you assing it to implement small tasks. Examples where it shines are implementing unit tests or refactoring some functions. - About my setup, I used Gemini (NOT as an agent) with the VsCodium extension. Works well but I found some issues: - If you make code changes without any AI interaction, the version the AI has of the files is NOT the same you have locally and you need to create a new chat from scratch to make them in sync again. - I only was able to accept the diff of each file individually. The accept all was not working for me. - You can't trust blindly the AI and a detailed review is required for each task. The AI can implement 10 unit test perfectly in a row but for some reason in the next request it will implement the test in a file in some other directory or will use a different name in the fixtures it was using before. - With small tasks, you don't need to be super specific with your prompts, Gemini will probably understand what you are trying to get (at least with simple projects as a Flask API.)
 
×
Reply to amunoz
This was my first time pair programming with Gemini AI (or with an AI agent in general). My HW25 project was more an excuse to work with such coding mate and evaluate the experience. I chose to develop a flask application implementing an API and a simple frontend, as I thought there might be plenty of Flask applications used for AI training. My reasoning was that Gemini would be good in that area of expertise. Additionally the project would allow me to experiment a bit in frontend technologies where I am a total noob.
Here are my thoughts about developing with an AI coding assistant: - Overal I think this kind of development flow is: - better for implementing PoCs or simple custom tools that were not worth the effort before. - bad in general for software quality. This will happen because the humans supervising the code generated by the AI will become lazy and trust the code without a proper check.
• In my experience, Gemini works better when you assing it to implement small tasks. Examples where it shines are implementing unit tests or refactoring some functions.

• About my setup, I used Gemini (NOT as an agent) with the VsCodium extension. Works well but I found some issues:

  • If you make code changes without any AI interaction, the version the AI has of the files is NOT the same you have locally and you need to create a new chat from scratch to make them in sync again.
  • I only was able to accept the diff of each file individually. The accept all was not working for me.

• You can't trust blindly the AI and a detailed review is required for each task. The AI can implement 10 unit test perfectly in a row but for some reason in the next request it will implement the test in a file in some other directory or will use a different name in the fixtures it was using before.

• With small tasks, you don't need to be super specific with your prompts, Gemini will probably understand what you are trying to get (at least with simple projects as a Flask API.)
---
# A First Level Header
## A Second Level Header

Use one asterisk to *emphasize*

Use two asterisks for **strong emphasis**

- Use hyphens
- for unordereed
- lists

This is an [link to example.com](http://example.com/)

This is an image ![an openSUSE geeko icon](https://en.opensuse.org/images/d/d0/Icon-distribution.png)

This is a user link @hans

This is a project link hw#some-cool-title
More Complex Markdown Help
Formatting Help
• Edit
• Preview