Hey, so I have absolutely LOVED cursor ever since I found out about it since around October, but as everyone has found out, this last version has had it’s fair share of issues.
We just want to know, will there be a reasonably significant step up when we get the next version? I personally haven’t been able to use cursor due to models being unable to edit for around 24 hours now, and it’s just getting frustrating. Can we get a little news from the devs, not feature wise but bug wise, and usability wise?
The real question i have is how will Cursor respond to the effective integration of all their features into VSCode core directly in the current insider version.
Agent mode, context awareness, model selection.
I don’t think they have Tab that works as well as cursor, but I’m sure with MS & Github based resources it wouldn’t take long.
If the free version of VSCode incorporates all of what makes cursor great, and then lets you use your own API keys.
I do hope Cursor pushes ahead and keeps MS copying, means Cursor is still the competitive advantage, but they really need to get the v0.46.* issue worked out ASAP, timing couldn’t be worse.
Take any ‘traditional’ IDE maker (MS, JB,…) with their AI extensions. Sadly they are not that great which is what drives Cursor forward
Not saying Cursor is perfect but given the CF thats happening with other ‘traditional’ IDEs its so much better. Yes some modern AI IDE competitors have certain features that Cursor has and vice versa.
Completely agree Cursor must improve their stability and keep UI for a while stable. (its fine to release beta versions as beta! but keep mainline version stable).
I wanna know if the will bring back chat mode - where I don’t wanna use agent but with previous apply button where I can apply only parts of code manually. This worked so much better for me before and I LOVED it. Right now I unsubscribed for now and if they don’t provide feedback if non agent going back then when will have to look for better pasture. Agent mode crashed also too many times sometimes in last step and this completely kill vibe.
For me this drive towards agent is actually a negative and I think cursor have been groomed by hype in the industry. I’m a senior developer. I don t want an agent most of the time. what I personally need are these features :
improved codebase chat and example (mermaid) flows of how aspects of the codebase work together. (this is one major aspect I use every day for different clients to understand quickly)
Select the context ‘I’ want, anything else is ignored and never edited. unless enabled
review and cherry pick rollback of any change.
agent only if I specifically request
45 was a great layout, it needed improvements in code understanding (something I think could be a huge feature for cursor).
It’s all about control and confidence, I can’t afford to have code altered anywhere other than where I specify and in the context of the change requested.
So much depends on experience, project, prompts and kinds of usage as well as tools (Cursor), LLM (3.5, 3.7, 4o, 4.5) and the task at hand.
Not sure what ‘grooming’ would there have been about agentic tools, sounds like a wild speculation without any proof.
Overall, so many different players in the industry are going towards automatization, agents and tool usage. Traditional IDE makers have slept through the last year of AI advancements and are about a year behind in DX.
You can use Cursor without Agent mode, its not forbidden to use Ask or Edit mode
As LLM are for now still widely just a regurgitation of statistical probabiliies of tokens not sure how you would have ‘control’ or ‘confidence’ unless you read all the code you receive, guide it effectively and steer it back on track when it weers of.
Code understanding depends so much on how well a project is structured, having faced from huge enterprise projects to tiny ‘vibe coding’ projects i tend to say that understanding is less of an issue than recognization of the right elements.
Not sure what prevents you from using mermaid now already.
You can quite well guide LLMs not to touch ‘unrelated’ code. again, depends on experience, language and code structure.
I am gonna be honest, the tab doesn’t feel nearly as good as it used to. I tend to still write a lot of my code and it doesn’t feel nearly as fast or as smart as it used to in the previous versions. I know someone will say “you just got used to it” but I don’t think so, I remember tab being faster and better because I specifically paid for it and not the agent mode.
There was an update on twitter about the new version of Cursor coming, summary:
Cursor AI is improving Sonnet 3.7 in version 0.47 to allow users to delegate more complex and demanding tasks to the AI model. The key is to utilize the model for long tasks, complex problems, and in-depth thinking. The update optimizes the AI for asynchronous work, which involves delegating tasks, and will be available with usage-based pricing due to the increased costs associated with the enhanced capabilities. A future post will detail the changes and how to enable them.
===
My guess is that they are really going to push the agent mode - which makes sense (and is what I would want) because otherwise as someone points out above you don’t have a product much difference from vscode with copilot.
