by talldayo on 1/26/2025, 5:30:15 PM
Context ain't free. The larger your project becomes, the more money you will spend trying to infer code from it. It's possible that the cost of writing complex code with AI exponentially correlates to the complexity of the working codebase.
by ericalexander0 on 1/26/2025, 5:41:23 PM
The key is modularity with good interface design, then you have AI generate each component and play more of a QA role to validate each component is functional.
When I try to code a somewhat complex project (backend with login capabilities, jwt, refreshtoken, and e2e tests) and I put all the requirements in the initial prompt, the output is usually condensed and incomplete, like it has a maximum output length that it can't exceed.
When I give it small prompts, I get more verbose and complete answers, but the prompts seem to forget the previous prompts, and throw out logic that was written previously.
What's the best way to feed instructions to these AI systems for relatively complex projects?