Cursor is still the best AI editor I’ve used. It multiplies my output more than anything else, which is why I’m bothering to write this instead of quietly leaving (in the process lol). I’d also ask the team not to take this personally. It’s feedback from someone who wants to keep using the product, so you blocking my posts feels rather petty!
I picked Cursor over plain VS Code because I wanted fine-grained control and visibility into what the agent was doing, inside an editor I could take over at any moment. It did that better than anything else, so I switched.
Lately it feels like the product is being steered away from that. A while back it was the agent’s command output no longer opening in the real terminal pane (my earlier thread link). Then agent reasoning got harder to follow. Now Cmd+K is gone: inline edit, inline questions, and terminal command generation. The answer I’ve seen is basically “it’s retired, use Agent.”
I don’t buy that those are the same thing. Selecting ten lines, asking a question or making a quick edit, or getting a shell command in a second and running it myself doesn’t need an agent session. Agent is great for a multi-file refactor. It’s overkill for the small stuff, and it’s slower and hides more of what’s happening.
I get the business logic => Agent workflows burn more tokens and more tokens equals more revenue. I can’t know that’s why these calls were made, but when the lightweight, hands-on options keep getting removed and nobody explains why, people are going to wonder - I know I am. What was the reason? Low usage, maintenance cost, a conflict with Agent? And why couldn’t they just coexist? “It’s gone” tells us what happened, and we already know that.
The bigger thing for me is this. I think the future is AI plus human judgment, and that only works if the human can actually judge the output. A lot of the marketing, especially the influencer “I built this in an afternoon” demos, sells the opposite: you approve what you can’t evaluate. Those projects look great until someone asks about security, scale, or what happens when it breaks. I use agents heavily and they’re good. I just don’t want an editor that treats me watching as the goal. I’ve always switched back to editor mode because I want to be in the code.
If the direction is fully hands-off, I don’t see what separates Cursor from Claude Code or Codex. A VS Code fork was supposed to be the difference.
This is the first time I’m looking at alternatives. I want AI-assisted dev first, agents when I choose to hand something off. What I need:
- A real editor, not an agent dashboard
- Inline edit and Q&A on selected code
- Natural-language terminal commands
- Full visibility into what agents run and change
- Easy manual override
- Ideally, any model provider
I’ve looked at VS Code + Copilot, Windsurf, Continue and Zed, but I’d rather hear from people who use them daily. What would you pick? And can you make any suggestions? Colin responding to this doesn’t count! I’d rather you not respond with your boilerplate stuff!
this quick UI window which gives semi-manual workflow was a 40% of reason we’ve been is using Cursor in our team.