Industrial
Linguistics
Slide 27 of 55

Claude Code - Advanced

Slide 27 / 55

Course concept

Big legacy code bases

  • Bug fixes or feature additions take more than 2 hours

    • Success probability for a one-shot request is low (it will take a long time)

  • The code base may fill up context (or be too big for context)

  • Documentation and policies matter more

  • Following procedures and checklists is more valuable

  • Lots of irrelevant code that has never been cleared out

Narration

Listen or continue with the transcript

Transcript

Working with big legacy codebases presents unique challenges. The main issue is context - you can't just load the entire codebase into Claude's context window. It's too large, and even if you could, it would be overwhelming.

When you're working in a greenfield project, you know exactly what files are relevant because you just created them. But in a brownfield project, figuring out which files matter for your task is half the battle.

You need to be strategic. Instead of loading everything, you need to identify the specific subsystem or component you're working on. Use grep to search for relevant code, use your IDE to understand the dependency graph, and only load the files that are directly relevant to your current task.

The good news is that Claude is excellent at navigating unfamiliar code. You can ask it to explain what a function does, trace through call chains, or identify where a particular feature is implemented. But you need to guide it to the right area first.

This is where your domain knowledge becomes crucial. You know the architecture, you know which services talk to which, and you can help Claude focus on the right parts of the codebase.