Because we're developers with things to build and we don't have time to play with every AI tool backed by the same LLM.
Like it or not, we're hitting the slope of enlightenment and some of us are ready to be done with the churn for a while.
Which is another reason why I'll stick with Cursor. Cursor's tab complete can barely be described as Copilot-style, it's nearly a different paradigm and it's what actually got me to pay for it in the first place. I only tried agent mode because it was included with the bundle.
> from day one
July 5, 2024 if going by initial commit. So, yes, technically before Cursor, but Cursor was building agent mode before Cline drew any attention in the mainstream. Cline's first interest on HN dates back to January.
I'll concede that it appears Cline did get to agents first, but it's still a new development in terms of actually drawing interest.
Cursor Agent works great, too.
Most importantly, everything is $20/month, instead of possibly $20/day with Cline or Roo.
It's just like VSC, which I was using, but it has these magical abilities. I was using it within a minute of downloading it. Unlike Cline, I guess, whatever that is.
You can use the full-context if you prefer that cost/speed tradeoff! Just have to turn on Max Mode.
Cline is great for many users, but a bit of a different product. Lots of Cursor's value come from custom models that run in the background (e.g. Tab, models that gather context, etc.).
There's still a ton of low hanging fruit that other Copilot-style autocomplete products don't seem to be picking up, like using clipboard contents, identifying the next place in the file to jump to, etc.
I primarily save time coding with AI with autocomplete, followed by chat, with agentic flows a very distant 3rd, so Cursor is a legitimately better product for me.
I didn’t say cursor has poor UX.
I tab too. And use agent for cheaper work I don’t care too much about. That said, the best autocomplete is arguably evolving and cursor does not own that.
And your answer is "A lot of power in social influence.", which is a bit silly when autocomplete is the first form of AI assistance a critical mass of people found intuitive + helpful and Cursor has the best implementation of it... meanwhile Cline/Roo Code don't provide it.
Your beloved cursor will go all in on this front, less and less priority on focused cursors in the editor.
And it's not "my beloved cursor", not sure why you're being such an absolute weirdo about this.
It is more effective when you have to do a bunch of similar changes. Or when code is standard enough that you just hit tab, and change perhaps one line. Or when parts of code are immediately deduced from context, and substitutions/changes are immediately provided.
I’d rather do that than painstakingly put my request into prose, cross my fingers, and hope the “agent” doesn’t burn 100,000 tokens making stupid and unrelated changes all over my codebase.
I believe in “show, don’t tell,” and autocomplete is the former while agents are the latter.
Agentic coding is fine, definitely helps me a lot with setup and boilerplate, but finer business logic details and UX changes are now it's strong suit especially if you know WHAT you want but not HOW to explain it in a clear enough format that it can do it without additional prompting.