I love Cursor and I couldn’t be more grateful for a tool that brought me back to programming after a decade. Anyone with a CS or SW engineering background who diverged, once you’ve brushed the rust off your architectural skills through a few frustrating cycles, is essentially hiring a junior SWE for pennies on the dollar. Thank you!
(This opening was, clearly, to prepare the reader, hopefully the cursor team, for some tough love.)
I understand you need to keep shipping new features and focus on the future (like background agents) at least as much if not more than the present. I also understand you need to balance the economics of huge growth with purely variable costs and mostly fixed (or let’s say MUCH less variable) per-user revenue. It is not hard to empathize with some hard decisions made under these conditions. However, please be more transparent. As in, fully transparent about the changes that go into a release.
2 fundamental things at play here:
- UX for vibe-coders is extremely intangible and intuition-driven: It’s about how effectively Cursor accomplishes the current task, and how added complexity makes subsequent tasks harder—something mostly captured as subtle intuition.
- Model performance is the primary determinant of UX, followed closely by your context building & prompt engineering. Even minor tweaks to these dramatically alter the UX, often without immediate clarity as to why the hard earned intuition is no longer working.
The influx of traffic onto Claude 4, combined with somewhat increased verbosity on reasoning tokens and 0.75 requests promotional pricing could’ve had significant financial impact, forcing a quick turn off thinking. Understandable. Less understandable is why the only indication was the quiet removal of the thinking icon and the reasoning tokens from chat answers.
Having been on both sides of the argument, I know intimately the tradeoff you’re trying to balance: Too much transparency amplifies user feedback but lowers its SNR, making triage more effortful—an obvious downside. Yet too little transparency, while clearly saving immediate effort, comes with subtle, hard-to-detect costs: user frustration that remains largely unvoiced or blends unnoticed into the already noisy vibe-coding feedback.
I don’t expect every Cursor release to tell me what changes were made in the secret sauce of context building for the agent. But I sure do cross my fingers really hard some information about how that context is built and how it is evolving in major versions, on broad strokes. And what I sure do expect is transparency on how model properties are being leveraged, or not.
In closing: Thank you again for making it possible to bring the joy of building back into my life. Please, pretty please, refrain from jerking the machine too hard. Or, if & when you have to, just be up front about it. Starting with a secondary changelog page for patch versions might be good start.
