Did you get a chance to try Cursor? I use Copilot only rarely, when using Github codespaces, so hard to get a feel for it.
But I'm impressed with the latest default model from Cursor. Comprehensive like the other "thinking" models, but can also do simpler targeted / repetitive tasks like the simpler models.
Seems it can go 10 or more cycles (e.g. 1. implement, 2. fix that, 3. address next point in todo, 4. fix that, etc...). And mostly correct (it keeps the previous constraints in place). For tests the maximum I tried so far was to let it fix ~50 tests, one by one. It was done in ~7 minutes, but correct.
Nice to hear from you, Juro! 👋I haven't used Cursor much but want to have a closer look next. For my current project I'm using Claude Code exclusively - tbh I'm not noticing a huge difference to Copilot at the codebase complexity I'm at.
Did you get a chance to try Cursor? I use Copilot only rarely, when using Github codespaces, so hard to get a feel for it.
But I'm impressed with the latest default model from Cursor. Comprehensive like the other "thinking" models, but can also do simpler targeted / repetitive tasks like the simpler models.
Seems it can go 10 or more cycles (e.g. 1. implement, 2. fix that, 3. address next point in todo, 4. fix that, etc...). And mostly correct (it keeps the previous constraints in place). For tests the maximum I tried so far was to let it fix ~50 tests, one by one. It was done in ~7 minutes, but correct.
For context, this was a project with ~10K lines of code, and relevant code being spread only across ~6 files. Greenfield Rust project, see https://github.com/django-components/djc-core/pull/17.
Nice to hear from you, Juro! 👋I haven't used Cursor much but want to have a closer look next. For my current project I'm using Claude Code exclusively - tbh I'm not noticing a huge difference to Copilot at the codebase complexity I'm at.