๐ช๐ต๐ ๐ฌ๐ผ๐๐ฟ ๐๐ ๐๐๐๐ถ๐๐๐ฎ๐ป๐ ๐๐ฒ๐๐ ๐๐ผ๐๐ ๐ถ๐ป ๐๐ฎ๐ฟ๐ด๐ฒ ๐ฅ๐ฒ๐ฝ๐ผ๐๐ถ๐๐ผ๐ฟ๐ถ๐ฒ๐
AI coding tools are great. They write code. They explain functions. They write tests. Small projects feel easy.
Things change in big repositories. Enterprise apps. Monorepos. Legacy systems. Your AI starts giving odd answers.
You think the AI is not smart. You are wrong. The problem is context.
AI reads code. It does not understand systems. A big repo has thousands of files. It has hundreds of APIs. It has years of decisions.
Ask about user onboarding. The logic lives in many places.
- Frontend code.
- Backend services.
- Event queues.
- Databases.
AI finds code snippets. Finding is not understanding. You search for a function. You find 15 results. You still do not know:
- Which one is used.
- What calls it.
- What happens next.
Systems are about relationships.
- Service A depends on Service B.
- Component X triggers Event Y. Connections matter more than files.
AI models have limits. Repositories are bigger than these limits. The AI chooses some files. It ignores others. It fills gaps with guesses. This leads to wrong answers.
Writing code is easy. Understanding architecture is hard. New tools use:
- Repository Memory.
- Knowledge Graphs.
- Architectural Maps. These treat code as a connected system.
AI needs to understand connections. Code is a network of decisions. Reading code is one thing. Understanding a system is another.
Source: https://dev.to/md_mijanur_molla/why-your-ai-assistant-gets-lost-in-large-repositories-3c9i