I've used Copilot ih VSCode quite extensively, using both GPT-5(.1) and Sonnet 3.7 and Sonnet 4.5 as the LLM backend. (Before our company got the deals in place for OpenAI and Anthropic to use their tooling directly)
The results are measurably worse every time vs. doing the exact same operation with Codex Cli or Claude Code.
There's something in the Microsoft AI framework that makes the worse and more "stupid" than the competition.
The results are measurably worse every time vs. doing the exact same operation with Codex Cli or Claude Code.
There's something in the Microsoft AI framework that makes the worse and more "stupid" than the competition.