@mattpocockuk Agreed, Copilot autocomplete is so bad that it makes me not want to edit files manually anymore and let Claude Code do everything. Its hallucinations and lack of context have actually introduced bugs that I'm sure Cursor Tab would not have. For example when completing patterns.
@housecor Depends on the task. Rarely review when stakes aren't high or vibe coding internal tools, or when tests are comprehensive enough. Always review when logic is more complex and there's a feeling that LLM is taking shortcuts while testing.
@ErvistheGreat@badlogicgames I don't want to build a thing that does everything, and it doesn't have to be offline, it actually needs to be online and shared. Everyone's requirements are different.
@GitMaxd@badlogicgames Thanks, I didn't even know that! But that's the issue with TUIs that everyone keeps reinventing shortcuts, visual affordances etc. With the IDE I've been working with for X years, there's just much less friction.
@simonw@charliermarsh When asking ChatGPT and Claude they confirm that the page should be mostly self contained, but can include links for further details.
@Pragmatic_Eng@cursor_ai Contrary to your article, the Tab model is not using Sonnet 4 or Gemini 2.5 for prediction. Those would be too slow. See https://t.co/o2f0tpCEbN
@charliermarsh uv has been the one thing making me appreciate working with Python again. It removes so much friction in setting up dev environments and making it easy for users to run my scripts.