And google anounced yesterday agent mode for gemini.
I used its really smart, but the real Cursor killer for me is Claude code by far. Yes they have an expensive subscription but the way that the environment it self can be improved and the integration between the model and the environment is something else.
Thanks! just signed up. cannot wait to have another alternative.
how’s windsurf been?
I believe cursor is the most powerful at the moment - I think everyone on here can agree thats not its biggest downfall - its the pricing to leverage the available models
Yeah cursor is greedyasfff. They thought they were the only IDE. With AI moving so quick and big tech making their own IDE’s it’s only sooner or later Cursor will fall
1 Like
I like it because it’s more transparent and predictive. The only annoying issue is that it times out after a certain period if the task is too long if you still in trial mode but if you are on pro you won’t get the error: “Resource exhausted: Encountered retryable error from model provider: rate limit exceeded for model; upgrade to pro account or try again in about an hour”.
Claude costs 4 credits, but at least with it, I know exactly how much it will consume. They have a discount on Gemini 1.5 Pro (cost 0.75 credits) and GPT4.1 (cost 0.25 credits), which can be used for simple tasks like modifying colors or arranging elements.
I never have luck with Gemini. I have tried it multiple times and every time it messed up. Grok 4 worked for me once then never get its response again, it simply returns nothing.
Kiro’s built-in workflow is so great. But since it is free, it is so slow at the moment so I don’t know if the actual implementation is as good as planning process. I’ll see in a few days.
Grok broken in Cursor. Only one-shot tasks in new chat
not really, I found it one-shots when there’s a long chat, new chat and it works
I think … yes. it is not because the model is better or what ever, the key point is: Cursor do not have a direction, Kiro have it: “spec-driven”. WHen talk about Kiro, we talk about spec-driven. when talk about Cursor, we talk about “i feel good, i think it work well”.
1 Like
In Cursor, yes. In Claude Code, nope. Like using two different LLMs.
2 Likes
Rubbish. I’ve worked with Claude up to 150K context, and it has never stripped out functionality (I have thousands *2555) unit tests ensuring nothing is removed.
Honestly you seem to have a beef with Claude, maybe it’s a user problem vs LLM problem.
Exactly! Cursor completely retards most models with their middle of the man re-prompting. Claude code just listens.
This is because Cursor has been built for pro users, anyone can replicate Kiro spec-driven workflow with rules like I show in my thread, some UI design could be taken like showing MCPs in a sidebar, enabling/disabling them really fast instead of going into settings or model options everytime
More like the Cursor vs LLM problem…
Cursor has till the end of the month to figure their stuff out or I will cancel and move to another option… (I understand that my singular cancellation will not really make a dent).
Cursor is unfortunately going to be left behind if they dont come up with a solution. If i were a competitor, i would be monitoring user pain points here and then solve them and use it as a hook. Im pretty sure the next platform that fixes their problems will see a mass migration away from cursor.