What we think about copilots

anon:

Copilot is fantastic to the point where I get quite irritated when it’s not available. It’s great for getting the syntax right for the next thing you want to do, like map/filter. I don’t find it to be that slow, but sometimes it just doesn’t show up, and it’s unclear why.

Claude feels like a significant step up from anything else when it comes to coding. It’s nice to break out of the IDE and chat to help construct bigger functions, but as the human, you still have to copy and paste the correct bits. Sometimes you just want the code back from the LLM without all the waffle.

Maybe Cursor solves this, but I haven’t used it enough to say for sure. I’m worried it might get the balance wrong, generating too much code to easily work with (as Copilot does), but not enough for me to consider it a new abstraction layer over the code that I can rely on for all syntax.

anon:

Cursor is feeling nice—I used it yesterday to create a working OpenAPI spec from a bunch of copied and pasted text from the Airtable docs.

Cursor’s tab autocomplete is a bit less nice than Supermaven’s, but I haven’t used it long enough to be confident in this opinion. I mostly use ChatGPT for direct questions, but I’ve heard that people are using Claude/Anthropic Workbench as a ChatGPT-alike, but better.

Supermaven’s speed has an outsized positive effect on the coding experience, in my opinion.

anon:

I’ve only used Copilot. When it’s not painfully slow, it’s great for menial tasks or things I used to do with copy-paste. It does suggest very odd stuff at times, though.