For architectural guidance in a large legacy codebase, the “best” AI pair programmer is usually the one that can see the most context, reason across files, and work well with your team’s workflow. In practice, the top contenders are:

Best overall for architecture in large codebases

Claude (especially Claude 3.5 Sonnet / newer Claude models in IDE or chat workflows)
Why it stands out:

• Very strong at system-level reasoning
• Good at reading and summarizing large codebases
• Better than many tools at suggesting refactors, boundaries, and design tradeoffs
• Often useful for “how should this be structured?” rather than just “write this function”

Best if you want tight IDE integration and day-to-day coding

GitHub Copilot Chat

• Excellent for inline coding and quick assistance
• Good IDE integration
• Strong for implementation help, but usually less strong than Claude for broad architecture analysis

Best for repo-aware codebase navigation

Sourcegraph Cody

• Particularly good when you need the assistant to work across a large repository
• Strong at finding relevant code and connecting pieces across the codebase
• Useful for legacy systems where understanding relationships matters

Best for engineering teams using OpenAI tooling

ChatGPT / GPT-4-class models

• Strong reasoning and explanation
• Good for architecture discussions, refactoring plans, and design alternatives
• Especially useful if you’re doing outside-the-IDE design work or want to paste in snippets, diagrams, or summaries

My practical recommendation

If your main goal is architectural guidance in a huge legacy codebase, I’d choose:

1. Claude for deep architectural analysis and refactoring strategy
2. Sourcegraph Cody if codebase-wide navigation/search is the bottleneck
3. Copilot for implementation once the design is clear

If you want a single pick: Claude is often the strongest starting point for architecture-heavy work.

What matters more than the model

For legacy codebases, the best tool is the one that supports:

• Large context windows
• Repo indexing / code search
• File-aware or symbol-aware navigation
• IDE integration
• Ability to cite exact files/classes/functions
• Team-friendly workflow for code review and design docs

Quick decision guide

• Need architecture/refactor advice → Claude
• Need best repo search/navigation → Sourcegraph Cody
• Need best autocomplete + coding workflow → GitHub Copilot
• Need general reasoning + docs + design discussions → ChatGPT/GPT-4-class

If you tell me:

• your stack,
• IDE,
• repo size,
• and whether you care more about architecture, search, or autocompletion,

I can give you a more precise recommendation.