Frustrated with AI assistance in app development

I’m modifying an app using Python Flask for the backend and SQLAlchemy as the ORM. I needed to add a new non-nullable boolean field to an existing model. While ChatGPT helped with the migration code, things became problematic when I tried to make this field ‘read-only’ in the admin panel.

The AI added the field to the appropriate views and had the logic to assign it a value during creation, but encountered an issue during editing: if the model’s existing value was True and I saved it as a disabled field, it got saved to the database as False because disabled fields don’t get included in the form submission. Despite trying different prompts, ChatGPT struggled to fix this.

This left me frustrated because I feel that while it’s great for quick snippets, it can’t handle even moderately complex logic without introducing bugs. I find more joy in researching and coding myself than in attempting to direct AI towards achieving my goal.

Has anyone else faced similar issues? Is there something I’m missing in how I’m utilizing AI tools?

They don’t build complex apps with AI.

Nico said:
They don’t build complex apps with AI.

“They”?

Nico said:
They don’t build complex apps with AI.

I do build with AI! :man_shrugging:

Vale said:

Nico said:
They don’t build complex apps with AI.

I do build with AI! :man_shrugging:

Good to know there’s an extreme edge case where it works! :ok_hand:

@Vale
Ask your AI for advice!

Rory said:
@Vale
Ask your AI for advice!

I’m genuinely interested in hearing how you find success using it.

Low-code platforms are often used instead of building complex apps from scratch; they simplify the process but can be limited.

Many users overestimate their prompting abilities and AI’s context understanding. You usually know more than the AI does, so you have to give it precise instructions.

I’ve been trying similar tasks with LLMs. While they can assist with simple tasks, they may not produce reliable results for complex coding.

Drew said:
I’ve been trying similar tasks with LLMs. While they can assist with simple tasks, they may not produce reliable results for complex coding.

It reminds me of blueprints for game creation without knowing C++. Sure, they help, but not enough in many cases.

@Skyler
These LLMs are limited compared to specialized coding assistants that have better integration and functionality.

Drew said:
@Skyler
These LLMs are limited compared to specialized coding assistants that have better integration and functionality.

I believe improving my coding skills is more beneficial than becoming adept at prompting AI.

Drew said:
I’ve been trying similar tasks with LLMs. While they can assist with simple tasks, they may not produce reliable results for complex coding.

AI can be a productivity enhancer, especially for larger projects, but it requires skill to utilize effectively.

@Ash
Have you worked on projects without AI assistance before? Would you mind sharing your work?

As a web developer, I find LLMs help setting up initial structures but not in debugging; you still need a strong foundational knowledge.

Rowan said:
As a web developer, I find LLMs help setting up initial structures but not in debugging; you still need a strong foundational knowledge.

I completely agree.

AI tools are evolving, but you still need to provide direction; they don’t replace understanding how to code.

Quill said:
AI tools are evolving, but you still need to provide direction; they don’t replace understanding how to code.

YouTube often showcases people claiming they can code without prior experience, but my experience contradicts that.

@Skyler
Yes, they often mislead for clicks; true programming requires foundational knowledge.