I've been using Amazon CodeWhisperer for the past year, mostly integrated into VS Code for Salesforce Apex and Lightning Web Components. It's fine for predictable boilerplate, but the suggestions often felt a bit "dumb" and context-blind beyond a few lines.
Now I'm on a Copilot trial. The hype is overwhelming, so I'm coming in skeptical. I want to move beyond the basic "it writes a function" demos. I need to know where it actually saves time in a real, messy dev workflow versus just being a fancier autocomplete.
Based on my CodeWhisperer experience, here's my initial test list. What else should I prioritize?
* **Context handling across multiple open files:** CodeWhisperer was weak here. Does Copilot actually use my open Apex class and its test class to suggest relevant, syntactically correct mocks and assertions?
* **Complex comment-to-code translation:** Not just "create a function that sorts a list," but something like "// Query all Opportunities where Stage is Closed Won, roll up Amount by Account, and map to a custom DTO." Does it build the SOQL, the loop, and the data structure?
* **Workflow interruption:** Does it constantly suggest unhelpful completions mid-typing and break your flow, or can it be tuned to be less intrusive?
* **API/database schema awareness:** If I have a `schema.sql` file open, how well does it generate model layer code or repository methods that correctly reference those tables and columns?
* **Handling of verbose, enterprise-style patterns:** My org uses a specific wrapper pattern for all HTTP callouts. Will Copilot recognize and continue that pattern, or will it try to write generic, non-compliant code?
I'm less interested in greenfield toy projects. My main pain points are navigating legacy codebases and adhering to strict org-specific patterns. If it can't help there, it's just a party trick.
-- CRM Surfer
Your CRM is lying to you.